remote sensing Article
Automatic Power Line Inspection Using UAV Images Yong Zhang 1 1 2
*
ID
, Xiuxiao Yuan 1,2, *, Wenzhuo Li 1 and Shiyu Chen 1
School of Remote Sensing and Information Engineering, Wuhan University, 129 Luoyu Road, Wuhan 430079, China;
[email protected] (Y.Z.);
[email protected] (W.L.);
[email protected] (S.C.) Collaborative Innovation Center of Geospatial Technology, Wuhan 430079, China Correspondence:
[email protected]; Tel.: +86-27-6877-1228
Received: 26 June 2017; Accepted: 9 August 2017; Published: 10 August 2017
Abstract: Power line inspection ensures the safe operation of a power transmission grid. Using unmanned aerial vehicle (UAV) images of power line corridors is an effective way to carry out these vital inspections. In this paper, we propose an automatic inspection method for power lines using UAV images. This method, known as the power line automatic measurement method based on epipolar constraints (PLAMEC), acquires the spatial position of the power lines. Then, the semi patch matching based on epipolar constraints (SPMEC) dense matching method is applied to automatically extract dense point clouds within the power line corridor. Obstacles can then be automatically detected by calculating the spatial distance between a power line and the point cloud representing the ground. Experimental results show that the PLAMEC automatically measures power lines effectively with a measurement accuracy consistent with that of manual stereo measurements. The height root mean square (RMS) error of the point cloud was 0.233 m, and the RMS error of the power line was 0.205 m. In addition, we verified the detected obstacles in the field and measured the distance between the canopy and power line using a laser range finder. The results show that the difference of these two distances was within ±0.5 m. Keywords: power line inspection; power line extraction; dense matching; obstacles within power line corridors; power line monitoring
1. Introduction The inspection of power lines is a critical factor for the safe operation of power transmission grids [1]. The monitoring of power lines incorporates two aspects: power line components and their surrounding objects, especially vegetation. The conditions of the components require regular checking to detect faults that are caused, for example, by corrosion and mechanical damage. There is also a need for the regular inspection of vegetation both inside and near the power line corridor. Trees or tree branches that are too close to power lines should be trimmed, because vegetation is conductive, which will lead to electric arcs. To ensure the safe operation of power lines, it is necessary to ensure that a range of empty space without conductive objects exists around an extra high voltage (EHV) power line. However, vegetation within the power line corridor will naturally grow after a power transmission grid becomes operational. A discharge may be generated when the distance between the vegetation and the power line is less than the safety threshold, thereby endangering the safe operation of the power transmission grid [2]. Therefore, research with the aim of discovering a highly efficient method for detecting obstacles along a power line corridor across a large area has received much attention. There are seven fundamental types of data that can be used for the inspection of power line corridors [3]: synthetic aperture radar (SAR) images, optical satellite images, optical aerial images, thermal images, airborne laser scanner (ALS) data, land-based mobile mapping data, and unmanned aerial vehicle (UAV) images. SAR image pixels include the radar backscattering intensity, the phase of the backscattering signal and the range from the sensor to the target. This information can be easily Remote Sens. 2017, 9, 824; doi:10.3390/rs9080824
www.mdpi.com/journal/remotesensing
Remote Sens. 2017, 9, 824
2 of 19
used to map power lines and towers and to monitor disasters, such as earthquakes and typhoons, that can harm power lines [4,5]. Optical satellite images and optical aerial images can also be used to monitor vegetation in power line corridors [6,7], although the detected results are rather coarse because of the lower spatial and temporal resolution of satellite images compared to aerial visible wavelength images. The advantages of aerial images are their high resolution and availability; thus, they are regularly used for the reconstruction of power line corridors [8]. Thermal images are sometimes used to inspect electrical faults in high voltage electric utility transmission and distribution lines. The study conducted by Tabor showed that, even at short distances, accurate temperature measurements of electrical faults are impossible to quantify [9]. ALS is an active remote sensing technique that can be applied to power line mapping, vegetation mapping and power line monitoring. For example, ALS-based inspection methods use a laser scanner mounted on an aircraft to scan along the power line corridor, thereby obtaining a 3D point cloud to conduct a 3D reconstruction of the power lines and the ground in order to detect obstacles within the corridor. At present, this method is applied only within a certain range [10,11], and for these applications, the point cloud density is the key factor in the 3D reconstruction of power lines [12]. However, the ALS technique has not been popular because of the high costs associated with laser scanning equipment possessing large scanning ranges and high scanning frequencies. Because of its large size and heavy weight, such equipment is usually mounted on a manned helicopter; thus, the cost is high, making it more difficult to arrange inspection work. Although small laser scanner equipment can be mounted on UAVs [13], the small scanning range and slow scanning frequency of this equipment makes it difficult to efficiently obtain sufficiently dense 3D point clouds. With the development of ALS equipment in the direction of miniaturization, a number of new types of ALS have been brought into the market [14,15], which is worthy of continuous attention. A land-based mobile mapping technology is based on the integration of various positioning, navigation and imaging data collection sensors that constitute a mobile mapping system (MMS), which is mounted on a kinematic platform, such as a car or a human. Unfortunately, many EHV transmission lines are mainly located in areas that lack a transportation network. A UAV equipped with a digital camera is a very convenient method which is emerging for power line inspection [16]. UAVs exhibit significantly lower operating costs than helicopters [17]; thus, the inspection of power line corridors across large areas can be implemented by installing a digital camera on a UAV [18,19]. Different UAV platforms have been applied to power line surveys. Generally, fixed-wing UAVs can fly higher and faster, and they are more suitable for vegetation monitoring and the rough inspection of long power lines, while helicopter and multi-rotor UAVs can be used to acquire detailed pictures by hovering in the air close to the objects [20]. The automatic measurement of power lines using stereo images is similar to automatic stereo mapping for a specific target in a particular scene, which is worthy of further study. When performing an inspection, a UAV equipped with a lightweight digital camera flies along the power line while acquiring stereo images of the corridor according to a certain overlap. A 3D reconstruction of the power lines and the ground can be performed using photogrammetric methods, and the obstacles can be automatically identified and located by calculating the spatial distance between the power line and the ground. In reference [21], we proposed a dense matching algorithm for the 3D reconstruction of the ground in power line corridors known as SPMEC, which has accomplished the automatic 3D reconstruction of objects such as the crowns of trees and roofs within a power line corridor. For a 3D reconstruction of power lines, we adopted the traditional manual stereo measurement method, which restricts the level of automation of inspection by using UAV images and which needs further improvement. It is difficult to use image-matching methods to find the corresponding points along a power line because of the small diameter of power lines and the complexity of their backgrounds. To calculate the 3D coordinates of power lines automatically, we extract 2D vectors of the power lines from stereo images. Since the object points in the stereo image pairs must be located along their corresponding epipolar lines, we acquire the corresponding points on the power line by calculating the intersection
Remote Sens. 2017, 9, 824
3 of 19
of the epipolar line and the corresponding power line. To extract power line vectors from digital images, Ramesh et al. [22] used the k-means clustering method based on the gray values of images. Combining the shape index and the density index, the clustering results can fall into two categories: power lines and non-power lines. A skeletonisation algorithm derived from mathematical morphology then extracts the power line centerlines. Yan et al. [23] used a Radon transform to extract line segments from power lines, followed by a grouping method to link each segment, and finally applied Kalman filter technology to connect the segments to form an entire line. Compared to these approaches, our objectives are to not only extract power line segments from images but also acquire the 3D coordinates of power lines, thus achieving automatic power line measurements. In this paper, based on the work discussed in the literature [21], we focus on the automatic measurement of power line 3D coordinates using UAV images to improve the automation level of power line inspections. Details of the proposed power line automatic measurement method based on epipolar constraints (PLAMEC) are described in Section 2.1. Brief descriptions of the 3D reconstruction of the ground within the power line corridor and of the obstacle detection method are mentioned in Sections 2.2 and 2.3, respectively. The validity of the PLAMEC was tested in a typical experimental area. The details of the experimental area and the analysis of the experimental results are described in Section 3. 2. Methodology All of the objects in the power line corridor with a distance to the power line that is less than the safety distance are defined as obstacles [21]. As shown in Figure 1, to detect obstacles within the power line corridor, we first use UAV images and ground control points (GCPs) to obtain the exterior orientation of images (EOIs) via block bundle adjustment (BBA). Then, epipolar images are generated based on the relative orientation parameters of stereo images. Subsequently, PLAMEC is proposed to extract power line 3D vectors from stereo image pairs composed of corresponding images from different flight strips, and the semi patch matching based on epipolar constraints (SPMEC) is employed to extract dense point clouds from within the corridor for a 3D reconstruction of the ground objects, including canopies, buildings, and other ground attachments. Finally, obstacles are automatically identified and located by calculating the spatial distance between the power line and the ground point cloud extracted from the optical imagery. The rest of this section is organized as follows. Section 2.1 describes the PLAMEC, which is the emphasis of this paper. Section 2.2 gives a brief description of the SPMEC, and Section 2.3 demonstrates the automatic detection of obstacles in a power line corridor. 2.1. Power Line Automatic Measurement Method Based on Epipolar Constraints When inspecting the power line corridor for obstacles, the flight parameters mast be designed according the camera parameters mounted on a UAV. The flying height of UAVs should trade off the image ground sample distance (GSD) and the operational efficacy. The higher the above-ground height of the UAV, the higher the efficiency of the photography and the bigger the GSD value will become. The power line would be too thin to be imaged because of the small photographic scale. In general, the diameter of EHV power lines is about 4 cm. Therefore, in practice, we are inclined to design the flight height according to GSD ≈ 4 cm. The forward and side overlaps are also essential parameters, and we recommend that the forward and side overlaps both be set at 80%. A high forward overlap will increase the robustness of the automatic tie points measurement method in aerial triangulation (AT). A high side overlap will reduce the risk of the power line being outside of the overlap area of the stereo images. After determining the flight height, as well as the forward and side overlap, a UAV flies back and forth along the power line, collecting two strips as shown in Figure 2. Geo-referencing the images was conducted using a block bundle adjustment process and ground control points.
Remote Sens. 2017, 9, 824 Remote Remote Sens. Sens. 2017, 2017, 9, 9, 824 824
4 of 19 44 of of 19 19
Images Images and and GCPs GCPs
Block Block bundle bundle adjustment adjustment
Exterior Exterior orientation orientation of of images images
Epipolar Epipolar images images
PLAMEC PLAMEC
SPMEC SPMEC
Power Power line line positions positions
3D 3D point point clouds clouds
Obstacles Obstacles
Figure 1. the automatic detection of obstacles in power line corridors. Figure The workflow for Figure 1. 1. The The workflow workflow for for the the automatic automatic detection detection of of obstacles obstacles in inpower powerline linecorridors. corridors.
Figure 2. Schematic diagram of the strip arrangement. The The green dots represent camera centers, and Figure Figure 2. 2. Schematic Schematic diagram diagram of of the the strip strip arrangement. arrangement. The green green dots dots represent represent camera camera centers, centers, and and the gray rectangles below the green dots are the captured images in world coordinates. The smallsmall dots the gray rectangles below the green dots are the captured images in world coordinates. the gray rectangles below the green dots are the captured images in world coordinates. The The small at the bottom are the projected ground points where the colorthe represents different elevationselevations indicated dots dots at at the the bottom bottom are are the the projected projected ground ground points points where where the color color represents represents different different elevations by the legend on the right in the figure. indicated indicated by by the the legend legend on on the the right right in in the the figure. figure.
The The diameter diameter of of aa power power line line is is small small (about (about 44 cm); cm); hence, hence, the the power power line line in in an an image image is is very very thin and is usually approximately one to three pixels. The texture is homogenous, while the thin and is usually usually approximately one to three pixels. pixels. The texture is homogenous, homogenous, while the image image image-matching background background is is composed composed of of complex complex textures. textures. It It is is difficult difficult to to use use image-matching image-matching methods methods to to find find points along the power line. We propose PLAMEC to perform automatic power corresponding points along the power line. We propose PLAMEC to perform automatic power corresponding points along the power line. We propose PLAMEC to perform automatic power line line measurements since the ground points in the stereo line measurements ground points stereoimage imagepairs pairsmust mustbe be located located along along measurements sincesince the the ground points in in thethe stereo image pairs must be located along the the flow chart of the proposed method. corresponding epipolar lines. Figure 3 shows a flow chart of the proposed method. corresponding epipolar lines. Figure 3 shows a flow chart of the proposed method. As As shown shown in in Figure Figure 3a, 3a, aa pair pair of of images images from from two two adjacent adjacent strips strips are are selected selected as as the the stereo stereo image image making the direction the power line approximately perpendicular to of the pairs, the power line approximately perpendicular to that of the line. pairs,making makingthe thedirection directionofof of the power line approximately perpendicular to that that ofepipolar the epipolar epipolar line. The epipolar images are generated based on the coplanarity condition, and the relative line. The epipolar images are generated based on the coplanarity condition, and the relative orientation orientation parameters parameters of of the the stereo stereo image image pair pair are are calculated calculated using using the the relative relative orientation orientation
Remote Sens. 2017, 9, 824
5 of 19
The epipolar images are generated based on the coplanarity condition, and the relative orientation Remote Sens. 2017, 9, 824 5 of 19 parameters of 9,the Remote Sens. 2017, 824stereo image pair are calculated using the relative orientation algorithm [24]. 5 of 19Then, the power line extraction algorithmextraction is used to extract the 2D vectors the2D power lines algorithm [24].automatic Then, the power line automatic algorithm is used to extractofthe vectors algorithm [24]. Then, theepipolar power line automatic extraction algorithm is used toinextract the 2DFinally, vectorsin the from the left and right images (as shown by the green solid lines Figure 3b). of the power lines from the left and right epipolar images (as shown by the green solid lines in of the power lines from the left right epipolar images (as shown byepipolar the green solid in direction of the y parallax the and epipolar pairs corresponding lines arelines obtained at Figure 3b). Finally, in theof direction of theimage, y parallax ofofthe epipolar image, pairs of corresponding Figure 3b). Finally, in the direction of the y parallax of the epipolar image, pairs of corresponding a epipolar certain interval shown by yellow dashed in Figure Thedashed intersections the epipolar lines are(as obtained at athe certain interval (aslines shown by the 3b). yellow lines inof Figure 3b). epipolar lines arevectors obtained at certain (as shown by theare yellow dashed lines Figure 3b). and line thea left andinterval right epipolar images obtained, and thein two intersections Thepower intersections of theinepipolar and power line vectors in the left and right epipolar images are The intersections of the epipolar and power line vectors in the left and right epipolar images are are a pair ofand corresponding points onare theapower The 3D coordinates obtained forward obtained, the two intersections pair ofline. corresponding points oncan thebe power line. by The 3D obtained, and the two intersections are a pair of corresponding points on the power line. The 3D coordinates[24]. can be obtained by forward intersection [24]. intersection coordinates can be obtained by forward intersection [24]. Left epipolar image Left epipolar image
Right epipolar image Right epipolar image
Extraction of power line 2D vectors Extraction of power line 2D vectors
Generation of power line 3D vectors Generation of power line 3D vectors
(a) (a)
(b) (b)
Figure3.3.Schematic Schematicdiagrams diagramsof ofthe theautomatic automaticpower powerline linemeasurement measurement method. method. (a) Flow Flow chart chart for for Figure Figure 3. Schematic diagrams of the automatic power line measurement method. (a)(a) Flow chart for the the power line automatic measurement method based on epipolar constraints (PLAMEC); (b) power line automatic measurement method based based on epipolar constraints (PLAMEC); (b) Visualization the power line automatic measurement method on epipolar constraints (PLAMEC); (b) Visualization of the PLAMEC. of the PLAMEC. Visualization of the PLAMEC.
2.1.1. Automatic Extraction of Power Line 2D Vectors from Epipolar Images 2.1.1. Automatic AutomaticExtraction ExtractionofofPower Power Line Vectors from Epipolar Images 2.1.1. Line 2D2D Vectors from Epipolar Images The automatic extraction of power line 2D vectors requires epipolar images as input and 2D The automatic automaticextraction extraction power line vectors requires epipolar images as input and 2D The ofof power line 2D2D vectors requires epipolar images as input and 2D vectors of power lines as output. A flow chart for power line 2D vector extraction is shown in Figure 4. vectors of ofpower powerlines linesasasoutput. output. flow chart power 2D vector extraction is shown in Figure 4. vectors AA flow chart forfor power lineline 2D vector extraction is shown in Figure 4. Epipolar image Epipolar image Step 1: Power line feature detection Step 1: Power line feature detection Step 2: Power line feature refinement Step 2: Power line feature refinement Step 3: Knowledge-based power line Step 3: Knowledge-based power line vector extraction vector extraction Power line 2D vectors Power line 2D vectors Figure 4. Flow chart of power line 2D vector extraction. Figure 4. 4. Flow chart of power lineline 2D 2D vector extraction. Figure Flow chart of power vector extraction.
The process contains three steps: the detection of power line features using a gray ratio operator, The process contains three steps: thethe detection of power lineline features using a gray ratio ratio operator, processof contains three steps: detection of power features a gray operator, the The refinement power line segments based on the same line conditions, andusing the extraction of power the refinement of power line segments based on the same line conditions, and the extraction of power line vectors based on prior This section these three steps. better illustrate the refinement of power lineknowledge. segments based on thedescribes same line conditions, andTo the extraction of the power line vectors based on prior knowledge. This section describes these three steps. To better illustrate the proposed method, we visualize the results for each step: the left epipolar image of a stereo is shown in the line vectors based on prior knowledge. This section describes these three steps. To better illustrate proposed method, we visualize the results for each step: the left epipolar image of a stereo is shown in Figure 5a, the detected power line featuresfor areeach shown Figure 5b, the refined line segment proposed method, we visualize the results step:inthe left epipolar imagepower of a stereo is shown in Figure 5a, the detected power line features are shown in Figure 5b, the refined power line segment results5a, arethe shown in Figure 5c,d,line andfeatures the extracted power in line vectors arethe shown in Figure Figure detected power are shown Figure 5b, refined power5e.line segment results are shown in Figure 5c,d, and the extracted power line vectors are shown in Figure 5e. results are shown in Figure 5c,d, and the extracted power line vectors are shown in Figure 5e.
Remote Sens. 2017, 9, 824
6 of 19
Remote Sens. 2017, 9, 824
6 of 19
(a)
(b)
(c)
(d)
(e)
Figure 5. Visualization of power line extraction. (a) Left epipolar image of a stereo; (b) Result of
Figure 5. Visualization of power line extraction. (a) Left epipolar image of a stereo; (b) Result of power line feature detection using a gray ratio operator; (c) Result of the refinement of power line power line feature detection using a gray ratio operator; (c) Result of the refinement of power line features using mathematical morphology; (d) Result of the refinement of power line features using features using mathematical morphology; (d) Result of the refinement of power line features using the the equivalent line condition; (e) Result of power line 2D vector extraction by prior knowledge. equivalent line condition; (e) Result of power line 2D vector extraction by prior knowledge.
Figure 5a shows the left epipolar image of a stereo. Figure 5b shows the results for line feature detection a gray to the of the background texture, therefor areline many Figure 5ausing shows theratio left operator. epipolarDue image of complexity a stereo. Figure 5b shows the results feature non-power line features in the feature extraction results. Based on an analysis of the difference detection using a gray ratio operator. Due to the complexity of the background texture, there are between the power line features and non-power line features, a filter was designed to refine the feature many non-power line features in the feature extraction results. Based on an analysis of the difference detection results. The refinement result is shown in Figure 5c. After an initial refinement, the line between the power line features and non-power line features, a filter was designed to refine the feature segments belonging to the same power line are approximately collinear. Each line segment is detection results. refinement result is shown in Figure 5c. After anisinitial extended, and aThe search is conducted for other collinear line segments, which taken refinement, as a conditionthe to line segments to current the same powerisline are approximately collinear. line segment is extended, judgebelonging whether the segment a power line segment, after whichEach the power line extraction and aresult search is conducted otherincollinear segments, which is taken condition to judge is refined further asfor shown Figure 5d.line After being refined by two steps, as theadetection results are power line segments. Based the prior of thewhich power the linespower as shown the image, the whether the current segment is a on power lineknowledge segment, after lineinextraction result is power line segments are linked, and a power line 2D vector result is output as shown in Figure 5e. refined further as shown in Figure 5d. After being refined by two steps, the detection results are power
line segments. Based on the prior knowledge of Gray the power lines as shown in the image, the power line (1) Detection of Power Line Features Based on Ratio Operators segments are linked, and a power line 2D vector result is output as shown in Figure 5e. In the process of power line inspection, the direction of flight and flight height of a UAV are
designed of to ensure power line is clearly imaged. In Operators general, the width of the power line in an (1) Detection Powerthat Linethe Features Based on Gray Ratio image is one to three pixels. Power lines are distinguishable from other features, which can be used In design the process poweroperators line inspection, direction flightthree andtypical flightobjects heightinof a UAV are to featureof detection for powerthe lines. Figure 6of shows a power designed to ensure the power line is clearly imaged. In general, the width the power line in an line corridor thatthat have strong edge responses with their corresponding gray valueofprofiles.
image is one to three pixels. Power lines are distinguishable from other features, which can be used to design feature detection operators for power lines. Figure 6 shows three typical objects in a power line corridor that have strong edge responses with their corresponding gray value profiles.
RemoteSens. Sens.2017, 2017,9,9,824 824 Remote Remote Sens. 2017, 9, 824
of19 19 77of 7 of 19
Edge Edge
Normal line Normal line
Power line Power line Figure 6. Typical features of power lines that differ from other objects. Blue curves are the gray value Figure 6. 6. Typical features of of power lines that differ from other objects. Blue curves are the gray value Figure Typical features lines that differ profiles of the red segments inpower the images on the left.from other objects. Blue curves are the gray value profiles of the red segments in the images on the left. profiles of the red segments in the images on the left.
From Figure 6, we can see that a power line is brighter (compared with the background, the From Figure 6, we can that a power is brighter (compared with thepeak background, the From Figure we canline see see that a power line isline brighter (compared with the background, metal material of6,power usually has higher reflectance) and has a narrower inthe themetal gray metal material of power line usually has higher reflectance) and has a narrower peak in the gray material of power to linethe usually higher such reflectance) andThe has monotone a narrowerminimum peak in the grayon profile profile compared edges has of objects as houses. values both profile compared to the edges of objects such as houses. The monotone minimum values on both compared to the edges of objects such as houses. The monotone minimum values on both sides of sides of the wave peak are approximately the same because power lines have small diameters and sides of the wave peak are approximately the same because power lines have small diameters and the wave peak are approximately the same because power lines have small diameters and have an have an approximately uniform texture on either side, which can be used to design feature detection have an approximately uniform texture on either which be usedfeature to design feature detection approximately uniform texture on either side, which can be usedcan to design detection operators operators for power lines, as shown in Figure 7. side, for power lines, shown foroperators power lines, as shown in as Figure 7. in Figure 7.
wc wc
wl wl
wr wr
Figure 7. Schematic diagram of a power line feature extraction operator. Figure 7. Schematic diagram of a power line feature extraction operator. Figure 7. Schematic diagram of a power line feature extraction operator.
In InFigure Figure7,7,the theblue bluecurve curveshows shows the the profile profileof ofthe thegray grayvalue valueof ofaarow rowof ofpixels pixelsin inan anepipolar epipolar Figure 7, the blue curve shows thethe profile of the value ofline, a row of pixels in marked an epipolar image. Three windows are along direction of gray thethe epipolar which are are marked as was image.In Three windows arearranged arranged along the direction of epipolar line, which l, image. Three windows are arranged along the direction of the epipolar line, which are marked ww wr (illustrated by the red squares shown in Figure 7). The size of each window is w pixels, andas c and l , wc and wr (illustrated by the red squares shown in Figure 7). The size of each window is w wr (illustrated by the squares shown in Figure size of each window is w thewdistance between the windows is δ red pixels. The gray value sum of7). theThe three windows is calculated l , wc and pixels, and the distance between the windows is δ pixels. The gray value sum of the three and marked as g , i ∈ ( l, c, r ) , after which the gray ratio is calculated for the area between the center pixels, and thei distance between the windows is δ pixels. The gray value sum of the three , after which the as gray is calculated for windowswisand calculated and marked i , i ∈ (l, c, window the windows wl andas wr .gWindows wlr) and wr are shown theratio windows on the left r) , after which the gray ratio is calculated for windowsc is calculated and marked as g i , i ∈ (l, c, and respectively, to Equation (1). If ratio r is greater threshold ε,wthe power wc and wl andthan wr .a Windows thethewindows the right, area between the according center window l and wr wof w w w and the windows and . Windows and thefeature area between theand center window line is detected, the gray value the central pixel in w is marked as 255; otherwise, thewr cl are shown as the windows on the left and cright, respectively, according to rEquation (1). If l the ratio gray isas marked as 0. Inon this way, the imageisof a power linethe feature detection is shown theawindows left andbinary right, respectively, according togray Equation (1). Ifresult the ratio isvalue greater than threshold the power line feature detected, and value of the central rare ε ,the generated as shown Figure 5b.ε , the power line feature is detected, and the gray value of the central than ainthreshold r is greater pixel in wc is marked as 255; otherwise, the gray 2gc −value gl − gisr marked as 0. In this way, the binary image the gray value is marked as 0. In this way, the binary image pixel in wc is marked as 255; otherwise,r = (1) of a power line feature detection result is generated shown in Figure 5b. 1 | gl − gas r| + of a power line feature detection result is generated as shown in Figure 5b. In Equation (1), the numerator is the sum in the gray values between the 2g c of− the g l differences − gr r =left 2g (1) − g − g center window and the windows on the and right. Since a power line is brighter than the r = g l c− g r l + 1 r (1) background, larger differences indicate a strongerg contrast between the power line and the background. − g +1 l
r
Remote Sens. 2017, 9, 824
8 of 19
Equation (1), RemoteIn Sens. 2017, 9, 824
the numerator is the sum of the differences in the gray values between8 ofthe 19 center window and the windows on the left and right. Since a power line is brighter than the background, larger differences indicate a stronger contrast between the power line and the The denominator is the absolute of the difference between the between gray sumthe of gray the left window background. The denominator is value the absolute value of the difference sum of the and the gray sum of the right window, as a power line has a small diameter with an approximately left window and the gray sum of the right window, as a power line has a small diameter with an uniform textureuniform on eithertexture side. To the operator from division byfrom 0, a value of 1by is added to the approximately onprevent either side. To prevent the operator division 0, a value of denominator. Given that the ground resolution of the UAV image is 4 to 5 cm and that the width of the 1 is added to the denominator. Given that the ground resolution of the UAV image is 4 to 5 cm and power is approximately pixel with one mixed the one power line pixel and the on that theline width of the power one line pure is approximately one purepixel pixelofwith mixed of ground the power bothand the the leftground side and right the power feature detection parameters set as line onthe both theside, left side and theline right side, the power operator line feature detectionare operator w = 1 pixel, δ = 1 pixel, and ε = 5.0. To ensure that the width in the binary image is a single pixel, parameters are set as w = 1 pixel, δ = 1 pixel, and ε = 5.0 . To ensure that the width in the a thinning algorithm basedpixel, on mathematical morphology [25] is to process the binary [25] image binary image is a single a thinning algorithm based onadopted mathematical morphology is of the power line features. During the power feature detection process, only ratio between the adopted to process the binary image of theline power line features. During thethe power line feature sum (the numerator) and the difference (the denominator) is calculated, which compared with(the the detection process, only ratio between the sum (the numerator) and isthe difference threshold as a is basis for judging theispresence a power line. Noise the image is inevitable; there denominator) calculated, which comparedofwith the threshold as in a basis for judging the presence are a large number of small, discontinuous and randomly shaped non-power line segments in the of a power line. Noise in the image is inevitable; there are a large number of small, discontinuous detection results, as shown in Figure 8. and randomly shaped non-power line segments in the detection results, as shown in Figure 8.
Figure result. Figure 8. 8. Partially Partially enlarged enlarged image image of of the the power power line line feature feature extraction extraction result.
It can be seen from Figure 8 that, compared with power line features, non-power lines are shorter, It can be seen from Figure 8 that, compared with power line features, non-power lines are shorter, more curved, and exhibit more space in their surroundings. The filter design starts by considering more curved, and exhibit more space in their surroundings. The filter design starts by considering these these characteristics and filtering out the micro scratch-like noises in the feature detection results characteristics and filtering out the micro scratch-like noises in the feature detection results without without having a significant impact on the real power line features. First, a vector tracking algorithm having a significant impact on the real power line features. First, a vector tracking algorithm [26] is [26] is used for the vectorization of the binary image of power line features, removing features that are used for the vectorization of the binary image of power line features, removing features that are shorter shorter than seven pixels. A shape index (SI) of the remaining features is calculated according to than seven pixels. A shape index (SI) of the remaining features is calculated according to Equation (2), Equation (2), thereby removing the curved features that are larger than 0.25 pixels. thereby removing the curved features that are larger than 0.25 pixels. (2) SI = A L SI = A/L (2) In Equation (2), A is the area of the polygon surrounded by the features, and L is the perimeter of the(2), polygon. HMTsurrounded (hit or miss transform) In Equation A is the Secondly, area of the An polygon by the features, based and L isonthemathematical perimeter of morphology is adopted to delete isolated features. the polygon. [25] Secondly, An HMT (hit the or miss transform) based on mathematical morphology [25] is adopted to delete the isolated features. (2) Power Line Segment Refinement Based on Equivalent Line Conditions (2) Power Line Segment Refinement Based on Equivalent Line Conditions The power lines are actually consecutive line in the images; however, after the initial refinement of the line are features, are represented discontinuous linethe segments because of Thedetected power lines actuallythey consecutive line in the by images; however, after initial refinement background and features, sources they of noise, as shown by in Figure 5c. There not only straight power line the detected line are represented discontinuous lineare segments because of background segments butofalso straight segments and curved segments other linearpower objects,line as shown in Figure 9. and sources noise, as shown in Figure 5c. There are notof only straight segments but also straight segments and curved segments of other linear objects, as shown in Figure 9.
Remote Sens. 2017, 9, 824 Remote Sens. 2017, 9, 824
9 of 19 9 of 19
Figure refinement of of power power line line features. features. Figure 9. 9. Partially Partially enlarged enlarged image image of of the the results results of of the the first first refinement
A vector tracking algorithm [26] is used for the vectorization of the binary image, as shown in A vector tracking algorithm [26] is used for the vectorization of the binary image, as shown in Figure 5c. For any line segment in the vector result, the extension line and the line segment Figure 5c. For any line segment in the vector result, the extension line and the line segment intersecting intersecting the extension line for the first time are obtained. If the direction of the intersected line the extension line for the first time are obtained. If the direction of the intersected line segment is segment is consistent with the current line segment, then the current line segment is retained; consistent with the current line segment, then the current line segment is retained; otherwise, the otherwise, the current line segment is removed as noise. Non-power line sources of noise are current line segment is removed as noise. Non-power line sources of noise are significantly decreased, significantly decreased, as a comparison between Figure 5c and Figure 5d shows. Consequently, as a comparison between Figures 5c and 5d shows. Consequently, equivalent line conditions of line equivalent line conditions of line segments effectively distinguish between power line and segments effectively distinguish between power line and non-power line features. non-power line features. (3) Power Line Vector Extraction Based on Prior Knowledge (3) Power Line Vector Extraction Based on Prior Knowledge In general, within an EHV transmission network, the distance between two towers is about 300 to In general, within an EHV transmission network, the distance between two towers is about 300 500 m. For a UAV image, the power lines run through the whole image, forming a set of parallel lines. to 500 meters. For a UAV image, the power lines run through the whole image, forming a set of Alternatively, there is one tower in the image, dividing the power lines into two sets of parallel lines. parallel lines. Alternatively, there is one tower in the image, dividing the power lines into two sets of First, the power line segments are grouped according to the condition regarding whether they are parallel lines. First, the power line segments are grouped according to the condition regarding parallel. For each set of parallel lines, a straight line is set parallel to them, and the distances between whether they are parallel. For each set of parallel lines, a straight line is set parallel to them, and the this straight line and the line segments are calculated. Subsequently, the power lines in the same set distances between this straight line and the line segments are calculated. Subsequently, the power are divided into different groups by the differences in the distance. The line segments in the same lines in the same set are divided into different groups by the differences in the distance. The line group can be considered to belong to the same power line and are then connected end to end. Finally, segments in the same group can be considered to belong to the same power line and are then the two ends of the power line are extended to make it the same length as the longest line in the same connected end to end. Finally, the two ends of the power line are extended to make it the same group of parallel lines. length as the longest line in the same group of parallel lines. 2.1.2. Automatic Generation of Power Line 3D Vectors 2.1.2. Automatic Generation of Power Line 3D Vectors As shown in Figure 3b, after extracting the 2D vectors of the power lines from the epipolar images, shown in Figure 3b, after the at 2D vectors of the in power lines from they epipolar pairsAs of corresponding epipolar linesextracting are obtained certain intervals the direction of the parallax images, pairs of corresponding epipolar lines are obtained at certain intervals in the direction of are the of the epipolar images. The intersections of the epipolar lines and corresponding power lines y parallax of the epipolar images. The intersections of the epipolar lines and corresponding power calculated, and they form a pair of corresponding image points on the power line. The 3D coordinates lines calculated, and they form a pair of corresponding points on the power Theway, 3D of theare corresponding points can be calculated using a forwardimage intersection algorithm [24].line. In this coordinates of the corresponding points can be calculated using a forward intersection algorithm 2D vectors of power lines can be converted into 3D vectors. [24]. There In thisare way, vectors power can be converted vectors. two2D issues thatof should belines addressed. One issue isinto the 3D determination of the corresponding There are two issues that should be addressed. One issue is the determination the power line vectors extracted from the epipolar images. The other issue is the improvement of of the corresponding power line vectors extracted from the epipolar images. The other issue is positioning accuracy of the corresponding image points on the power line at the sub-pixel level. the improvement of the positioning accuracy of the corresponding image points on the power line at the (1) Selection of Corresponding Power Lines sub-pixel level. Automatically matching Power power Lines line vectors is very difficult because the image textures (1) Selection of Corresponding surrounding the power lines are very similar. Thus, we applied manual work to match the power Automatically matching line vectors is very difficult because the image textures line vectors and consider thempower as the seeds. As shown in Figure 10, power lines a and a0 (Stereo 1) surrounding the power lines are very similar. Thus, we applied manual work to match the power represent the manually selected corresponding power lines, and their 3D coordinates can be calculated line vectors and consider them as the seeds. As shown in Figure 10, power lines a and a′ (Stereo 1) represent the manually selected corresponding power lines, and their 3D coordinates can be
Remote Sens. 2017, 9, 824
10 of 19
Remote Sens. 2017, 9, 824
10 of 19
by forward intersection. In the next stereo image pair (Stereo 2 in Figure 10), the corresponding power calculated by forward intersection. In the next stereo image pair (Stereo 2 in Figure 10), the lines can be automatically determined by the re-projection of the 3D coordinates. The correspondence corresponding power lines can be automatically determined by the re-projection of the 3D of the rest of the power lines can be conducted in the same manner. coordinates. The correspondence of the rest of the power lines can be conducted in the same manner. a
Stereo 1
b
a'
c
Left image
b'
c'
Right image
Stereo 2
Flight direction Figure10. 10. Illustration Illustrationof ofthe thesselection of corresponding corresponding power power lines. lines. Figure election of
(2) Sub-Pixel Level Positioning of the Corresponding Points along the Power Lines (2) Sub-Pixel Level Positioning of the Corresponding Points along the Power Lines To ensure that the accuracy during the automatic measurement of power lines is consistent with Tomanual ensure measurements, that the accuracy the of automatic measurement of power is consistentatwith that of theduring accuracy the intersection positioning mustlines be determined the that of manual measurements, the accuracy of the intersection positioning must be determined at the sub-pixel level. The positioning accuracy of the 2D power line vectors in the epipolar images sub-pixel by level. positioning accuracy of the2.1.1 2D power line vectors in theand epipolar achieved achieved theThe method discussed in Section is expressed in pixels, thus, images the positioning by the method in Section 2.1.1 expressed pixels,lines and is thus, accuracy of accuracy of the discussed intersections between theisepipolar andinpower onlythe at positioning the pixel level. During the intersections between the epipolar and power lines is only at the pixel level. During manual stereo manual stereo measurement, the positioning accuracy of the stereo marks in both the left and right measurement, theispositioning accuracy the stereo marks insub-pixel both the left and right images epipolar images approximately 0.5 ofpixels. To achieve accuracy, weepipolar calculate the is approximately 0.5 pixels. To achieve sub-pixel accuracy, we calculate the intersection between the intersection between the epipolar line and the power line. The gray ratio coefficient r between the epipolar line and the power line. The gray ratio coefficient r between the intersection point and the intersection point and the adjacent pixels on the left and right sides are calculated according to adjacent pixels onadopt the left and right sides calculated according to Equation (1). Wefor adopt parabolic Equation (1). We a parabolic fittingare algorithm [24] to acquire the coordinates the amaximum fitting algorithm [24] to acquire the coordinates for the maximum value of the coefficients, which are value of the coefficients, which are used as the true intersection between the epipolar line and the used as the true intersection between the epipolar line and the power line 2D vector. power line 2D vector. 2.2. Power Line Corridor 3D Reconstruction Using the SPMEC 2.2. Power Line Corridor 3D Reconstruction Using the SPMEC Modern 3D reconstruction algorithms include patch base multi-view system (PMVS) [27], Modern 3D reconstruction algorithms include patch base multi-view system (PMVS) [27], multi-photo geometry constrains (MPGC) [28] and semi-global matching (SGM) [29]. For an efficient multi-photo geometry constrains (MPGC) [28] and semi-global matching (SGM) [29]. For an efficient matching algorithm, we applied the SPMEC, which was proposed in reference [21]. The principle matching algorithm, we applied the SPMEC, which was proposed in reference [21]. The principle of of the SPMEC is as follows: first, an initial parallax mesh is constructed using tie points; second, the the SPMEC is as follows: first, an initial parallax mesh is constructed using tie points; second, the match window slides along the epipolar line within a certain search window guided by the initial match window slides along the epipolar line within a certain search window guided by the initial parallax; finally, a tentative match is determined by applying a similarity measure for the shape of parallax; finally, a tentative match is determined by applying a similarity measure for the shape of the correlation coefficient curve within a certain search window, and the 3D reconstruction process the correlation coefficient curve within a certain search window, and the 3D reconstruction process is completed. is completed.
Remote Sens. Sens. 2017, 2017, 9, 9, 824 824 Remote
11 of of 19 19 11
2.3. Automatic Detection of Obstacles within a Power Line Corridor 2.3. Automatic Detection of Obstacles within a Power Line Corridor After extracting the power line and the 3D point clouds of the power line corridor, we divide After extracting power with line and point clouds of theδ power corridor, divide the power line into n the segments δ asthe the3D step distance. When is smallline enough, eachwe segment the line into segments with δ as is the step distance. When δ is small each segment can power be regarded as anstraight line. If there a point within the 3D point cloudsenough, at a distance from the can be regarded a straight line. If there is a point within the clouds a distance from the straight line thatas is less than the safety distance, an obstacle can3D bepoint detected, andatthe exact position of straight line is that is less than the safety distance, an obstacle can be detected, and the exact position of the obstacle determined. the obstacle is determined. 3. Results and Discussion 3. Results and Discussion 3.1. Basic Conditions of the Experimental Data 3.1. Basic Conditions of the Experimental Data A typical working environment of a power line corridor inspection area located in Guizhou A typical working environment of a power line corridor inspection area located in Guizhou Province, southwestern China, is selected to test the proposed method. The province has a Province, southwestern China, is selected to test the proposed method. The province has a subtropical subtropical humid monsoon climate, abundant rainfall, dense vegetation cover, and undulating humid monsoon climate, abundant rainfall, dense vegetation cover, and undulating terrain. There are terrain. There are approximately 19,537 km of power lines with a voltage degree above 110 kV in approximately 19,537 km of power lines with a voltage degree above 110 kV in Guizhou Province [30]. Guizhou Province [30]. We select a section of a 220 kV power line that is approximately 3.9 km in We select a section of a 220 kV power line that is approximately 3.9 km in length. The elevation length. The elevation difference along the power line corridor is approximately 200 m. There are difference along the power line corridor is approximately 200 m. There are nine towers within the nine towers within the selected power line corridor (Numbered: T108-T116). Each tower has three selected power line corridor (Numbered: T108-T116). Each tower has three separated power lines, and separated power lines, and thus, the total length of the power lines is 11.7 km ( 11.7 km = 3.9 km × 3 ). thus, the total length of the power lines is 11.7 km (11.7km = 3.9km × 3). The same fixed-wing UAV The same fixed-wing UAV found in reference [21], with a Nikon D810 camera (focal length: 51.42 found in reference [21], with a Nikon D810 camera (focal length: 51.42 mm, complementary metal oxide mm, complementary metal oxide semiconductor (CMOS) sensor, 35.9 mm × 24.0 mm, 7360 × 4912 semiconductor (CMOS) sensor, 35.9 mm × 24.0 mm, 7360 × 4912 pixels, pixel size: 4.88 µm) is adopted pixels, pixel size: 4.88 μm ) is adopted for the photography of the power line corridor. The camera for the photography of the power line corridor. The camera interior parameters were calibrated interior parameters were calibrated before commencing aerial photography. The images were shot at before commencing aerial photography. The images were shot at a fixed distance. The forward a fixed distance. The forward and side overlaps were both 80%. The relative height of the UAV to the and side overlaps were both 80%. The relative height of the UAV to the ground is approximately ground is approximately 400 m. The height of each power line tower is approximately 25 m. The 400 m. The height of each power line tower is approximately 25 m. The GSD is approximately 3.8 cm. GSD is approximately 3.8 cm. Two strips were flown, and 178 valid photos were acquired. The strip Two strips were flown, and 178 valid photos were acquired. The strip arrangement and topography arrangement and topography are shown in Figure 11. The edges of the image are sharp and without are shown in Figure 11. The edges of the image are sharp and without obvious blurring, and the obvious blurring, and the contrast is moderate, without over-exposure or under-exposure; therefore, contrast is moderate, without over-exposure or under-exposure; therefore, they meet the requirements they meet the requirements of power line inspection. All of the images are geo-referenced using an of power line inspection. All of the images are geo-referenced using an aerial photogrammetric aerial photogrammetric triangulation process. triangulation process.
Figure 11. 11. Topography Topography and Figure and strip strip arrangement arrangement within within the the experimental experimental area. area.
The PLAMEC algorithm was verified by experiments, including the 2D power line vector extraction capability and extraction accuracy. The accuracy of the automatic 3D reconstruction of power lines was evaluated by comparing the automatic generation of 3D power line vectors and the results of manual power line measurements. The 3D reconstruction of the ground of the power line corridor was achieved using the SPMEC. Taking 6.5 m as the threshold, the obstacles in the corridor were automatically automatically detected. detected. An obstacle distribution table was output, and the values were checked were investigation to to verify verify the the practicability practicability of of the the proposed proposed method. method. through a field investigation
Remote RemoteSens. Sens.2017, 2017,9,9, 9,824 824 Remote Sens. 2017, 824
1212ofof1919 12 of 19
3.2. 3.2.Analysis Analysisofofthe theBlock BlockBundle BundleAdjustment Adjustment 3.2. Analysis of the Block Bundle Adjustment We We employed employed the the block block bundle bundle adjustment adjustment (BBA) (BBA) software, software, PATB PATB [31] [31] toto obtain obtain EOIs. EOIs. Tie Tie We employed the block bundle adjustment (BBA) software, PATB [31](SIFT) to obtain EOIs. Tie points points were initially matched using the scale-invariant feature transform [32] descriptor points were initially matched using the scale-invariant feature transform (SIFT) [32] descriptorand and were initially matched matching using the scale-invariant feature transform (SIFT) [32] descriptor and refined by refined refinedby byleast leastsquare square matching[33]. [33].Part Partofofthe thetie tiepoints pointsare areshown shownininFigure Figure12. 12. least square matching [33]. Part of the tie points are shown in Figure 12.
Figure12. 12.Tie Tiepoint pointdistribution distributionininimages. images.The Thedashed dashedrectangles rectanglesare areimages. images.The Theblue bluedots dotsare aretie tie Figure Figure 12. Tie point distribution in images. The dashedfrom rectangles arestrip. images. The blue dots are tie points from inter-strips and the pink dots are tie points the same points from inter-strips and the pink dots are tie points from the same strip. points from inter-strips and the pink dots are tie points from the same strip.
FromFigure Figure12, 12,we wecan cansee seethat, that,for foreach eachimage, image,there thereare are200~300 200~300evenly evenlydistributed distributedtie tiepoints. points. From From Figure 12, we can see that, for each image, there are 200~300 evenly distributed tie points. We surveyed the coordinates 15 ground points along the power line corridor and used 11 points as We surveyed the coordinates of ground points along the power line corridor and used 11 points We surveyed the coordinates of 15 ground points along the power line corridor and used 11 points GCPs in BBA and four of them as check points. The distribution of these ground points is shown in as GCPs in BBA and four of them as check points. The distribution of these ground points is shown as GCPs in BBA and four of them as check points. The distribution of these ground points is shown Figure 13. inin Figure 13. Figure 13. 9001 9001 9002 9002
9003 9003 9007 9007
9006 9006 9008 9008 9009 9009
9011 9011 9013 9013 9010 9015 9010 9015 9012 9012 9014 9014
9016 9016
9018 9018
Figure Figure13. 13.Distribution Distributionofof ofthe theground groundcontrol controlpoints points(GCPs) (GCPs)and andcheck checkpoints. points.The Thedashed dashedrectangles rectangles Figure 13. Distribution the ground control points (GCPs) and check points. The dashed rectangles are images. The red triangles are GCPs and the green triangles are check points. are images. The red triangles are GCPs and the green triangles are check points. are images. The red triangles are GCPs and the green triangles are check points.
All and check points are located on identifiable corners of objects. The AllGCPs GCPs located oneasily easily identifiable corners ofground ground objects. The GCPs and andcheck checkpoints pointsareare located on easily identifiable corners of ground objects. coordinates were surveyed using a global navigation satellite system (GNSS) real-time kinematic coordinates were surveyed using a global navigation satellite system (GNSS) real-time kinematic The coordinates were surveyed using a global navigation satellite system (GNSS) real-time kinematic (RTK) calculated using the continuously operating reference stations (CORS) system in the (RTK) and and calculated using the continuously operating reference stations (CORS) (CORS) system system in the CGCS-2000 coordinate system. The re-projected image coordinates of the GCPs were measured CGCS-2000 CGCS-2000 coordinate coordinate system. system. The The re-projected re-projected image image coordinates coordinates of of the the GCPs GCPs were measured manually. manually.The Thelocation locationand andimage imagemeasurement measurementof ofGCP GCPNo. No.9012 9012are areshown shownininFigure Figure14. 14. Figure 14.
Remote Sens. 2017, 9, 824 Remote Sens. 2017, 9, 824
13 of 19 13 of 19
(a)
(b)
Figure 14. 14. The Thelocation locationand andimage imagemeasurement measurement GCP 9012. location of No. (b) Figure of of GCP No.No. 9012. (a) (a) TheThe location of No. 9012;9012; (b) The The image measurement of 9012; No. 9012; thecrosses red crosses represent the location the GCP the images. image measurement of No. the red represent the location of theofGCP in theinimages.
Before applying BBA, we used the precise point position (PPP) algorithm [34] to calculate the Before applying BBA, we used the precise point position (PPP) algorithm [34] to calculate the coordinates of the antenna phasing center associated with the onboard GNSS receiver, for each image coordinates of the antenna phasing center associated with the onboard GNSS receiver, for each image at the instant of exposure. Subsequently we input all tie points, GCP coordinates, and antenna phasing at the instant of exposure. Subsequently we input all tie points, GCP coordinates, and antenna phasing center coordinates to the PATB software. Since image distortion was corrected, we did not use camera center coordinates to the PATB software. Since image distortion was corrected, we did not use camera self-calibration during PATB analysis. After BBA, we used adjacent image pairs from the same strip as self-calibration during PATB analysis. After BBA, we used adjacent image pairs from the same strip as stereo image pairs and measured all check points using stereo-measurement software. The difference stereo image pairs and measured all check points using stereo-measurement software. The difference between the field surveyed coordinates and software measured coordinates are shown in Table 1. between the field surveyed coordinates and software measured coordinates are shown in Table 1. Table 1. The coordinate differences of check points from in-strip stereo image pair (unit: m). Table 1. The coordinate differences of check points from in-strip stereo image pair (unit: m). Coordinate Difference Check Point No. Stereo No. Coordinate Difference Easting Northing Elevation Stereo No. Check Point No. DSC_8780-DSC_8781 0.115 −0.019 Easting Northing 0.327 Elevation DSC_8781-DSC_8782 −0.080 0.043 0.324 DSC_8780-DSC_8781 0.115 −0.019 0.327 DSC_8782-DSC_8783 0.082 −0.038 0.294 DSC_8781-DSC_8782 −0.080 0.043 0.324 9003 DSC_8809-DSC_8810 0.125 0.052 0.212 0.294 DSC_8782-DSC_8783 0.082 −0.038 9003 DSC_8810-DSC_8811 0.046 0.070 −0.183 0.212 DSC_8809-DSC_8810 0.125 0.052 DSC_8811-DSC_8812 −0.052 −0.106 −0.280−0.183 DSC_8810-DSC_8811 0.046 0.070 DSC_8820-DSC_8821 −0.132 −0.062 0.058−0.280 DSC_8811-DSC_8812 −0.052 −0.106 9011 DSC_8821-DSC_8822 −0.049 −0.081 0.214 0.058 DSC_8820-DSC_8821 −0.132 −0.062 DSC_8842-DSC_8843 −0.089 −0.053 −0.194 DSC_8821-DSC_8822 −0.049 −0.081 0.214 9011 DSC_8928-DSC_8929 0.072 0.154 0.255−0.194 DSC_8842-DSC_8843 −0.089 −0.053 DSC_8929-DSC_8930 −0.071 −0.026 0.283 DSC_8928-DSC_8929 0.072 0.154 0.255 DSC_8948-DSC_8949 0.082 −0.017 0.223 0.283 DSC_8929-DSC_8930 − 0.071 − 0.026 9013 DSC_8949-DSC_8950 −0.090 0.037 0.176 0.223 DSC_8948-DSC_8949 0.082 −0.017 9013 DSC_8952-DSC_8953 0.088 −0.034 −0.265 DSC_8949-DSC_8950 −0.090 0.037 0.176 DSC_8973-DSC_8974 −0.083 0.036 0.157−0.265 DSC_8952-DSC_8953 0.088 −0.034 DSC_8973-DSC_8974 −0.083 0.036 DSC_8960-DSC_8961 0.009 0.086 −0.153 0.157 DSC_8987-DSC_8986 0.011 0.166 −0.298−0.153 DSC_8960-DSC_8961 0.009 0.086 DSC_8986-DSC_8985 −0.004 −0.097 −0.385−0.298 DSC_8987-DSC_8986 0.011 0.166 9016 DSC_8985-DSC_8984 −0.039 0.058 −0.185−0.385 DSC_8986-DSC_8985 −0.004 −0.097 DSC_8985-DSC_8984 − 0.039 0.058 9016 DSC_9023-DSC_9022 −0.017 −0.115 0.090−0.185 DSC_9023-DSC_9022 − 0.017 − 0.115 DSC_9022-DSC_9021 −0.017 −0.057 0.275 0.090 DSC_9022-DSC_9021 − 0.017 − 0.057 DSC_9021-DSC_9020 −0.004 −0.122 0.078 0.275 DSC_9021-DSC_9020 −0.004 −0.122 0.078 Root mean square error of the coordinate difference 0.073 0.081 0.233 Root mean square error of the coordinate difference 0.073 0.081 0.233
From Table 1, we can see that the manually-measured coordinates of check points using stereo-measure software are quite consistent with the coordinates surveyed in the field. The maximum differences of the easting, northing, and elevation of the check points are 0.125 m, 0.166 m and −0.385 m, and the root mean square (RMS) errors are 0.073 m, 0.081 m and 0.233 m, respectively.
Remote Sens. 2017, 9, 824
14 of 19
From Table 1, we can see that the manually-measured coordinates of check points using stereo-measure software are quite consistent with the coordinates surveyed in the field. The maximum differences of the easting, northing, and elevation of the check points are 0.125 m, 0.166 m and −0.385 m, and the root mean square (RMS) errors are 0.073 m, 0.081 m and 0.233 m, respectively. Since the coordinates of check points were measured using a stereo image pair within a strip, the accuracy thus can be considered representative of the accuracy of SPMEC. In this same way, we used adjacent images pairs from inter-strip as stereo image pairs and measured all check points using the stereo-measure software. The difference between the field surveyed coordinates and software measured coordinates are shown in Table 2. Table 2. The coordinate differences of check points from inter-strip stereo image pair (unit: m). Coordinate Difference Check Point No.
Stereo No.
Easting
Northing
Elevation
9003
DSC_8780-DSC_8809 DSC_8781-DSC_8810 DSC_8782-DSC_8811 DSC_8783-DSC_8812
0.047 0.038 0.065 0.080
0.055 0.046 0.063 0.121
−0.217 −0.205 −0.250 −0.209
9011
DSC_8820-DSC_8842 DSC_8821-DSC_8843 DSC_8822-DSC_8843
−0.016 −0.123 −0.038
−0.056 −0.064 −0.078
0.194 0.125 0.203
9013
DSC_8928-DSC_8948 DSC_8929-DSC_8949 DSC_8930-DSC_8950 DSC_8952-DSC_8973 DSC_8953-DSC_8974
−0.156 −0.080 −0.101 0.084 0.090
0.058 0.122 0.047 −0.076 −0.100
0.305 0.232 0.058 0.219 0.240
9016
DSC_8960-DSC_8983 DSC_8961-DSC_8983 DSC_8987-DSC_9023 DSC_8986-DSC_9022 DSC_8985-DSC_9021 DSC_8984-DSC_9020
−0.055 −0.040 0.094 −0.018 0.009 0.004
−0.112 0.097 −0.110 0.068 0.079 −0.109
−0.141 −0.286 −0.225 −0.166 −0.187 0.050
0.075
0.085
0.205
Root mean square error of the coordinate difference
As shown in Table 2, the coordinates of check points measured from inter-strip stereo image pair were also quite consistent with the coordinates measured in the field. The maximum differences of the easting, northing and elevation of the check points are −0.156 m, 0.122 m and 0.305 m, and the RMS errors are 0.075 m, 0.085 m and 0.205 m, respectively. Since the coordinates of check points was measured using inter-strip stereo image pair, the accuracy thus can be used as the accuracy of PLAMEC. From Tables 1 and 2, we can see that the planar coordinates are almost the same using either the stereo image pair from in-strip or inter-strip. When the overlap of in-strip and inter-strip stereo images are both 80% and the shape of the onboard sensor is a rectangle, then the base line of the inter-strip image pair is slightly longer than an in-strip image pair. Thus, the elevation accuracy from the inter-strip stereo pair was slightly higher than the accuracy of the in-strip stereo pair. 3.3. Analysis of the Results of Automatic Power Line Measurements To verify the feasibility of the PLAMEC technique proposed in Section 2.1, the power lines in the experimental area were automatically measured to have a total length of 10.9 km. In the experimental area we chose, there were three power lines on the towers, reaching a total length of 11.7 km. Thus, the success rate for automatic extraction of power line was 93.2%.
Remote Sens. 2017, 9, 824
15 of 19
To further verify the measuring accuracy of power lines using the PLAMEC, we registered the epipolar images with the extracted power line 2D vectors, as shown in Figure 15, and evaluated the Remote Sens. 2017, 9, 824 15 of 19 accuracy by visual comparison.
Enlarging scale: 1:5
Figure 15. Registration of a 2D vector and epipolar images of the power line. Figure 15. Registration of a 2D vector and epipolar images of the power line.
As shown in Figure 15,15, power line beregistered registeredaccurately accurately with power As shown in Figure power line2D 2Dvectors vectors can can be with the the power lines lines in theinepipolar image. WeWe manually checked 89)stereo stereo image pairs found the epipolar image. manually checked89 89(178/2 (178/2 ==89) image pairs andand found that that the the estimated registration error is approximately one pixel.The Theregistration registration results results are consistent estimated registration error is approximately one pixel. consistentwith with the the accuracy location accuracy of the proposed line feature detector. location of the proposed powerpower line feature detector. According to the method introduced in Section 2.1.2, after the generation of 3D vectors from thefrom According to the method introduced in Section 2.1.2, after the generation of 3D vectors power line 2D vectors, we compared the automatic measurement and stereoscopic manual the power line 2D vectors, we compared the automatic measurement and stereoscopic manual measurements on the stereo image pairs, which were performed on photogrammetric software measurements on the stereo image pairs, which were performed on photogrammetric software MapMatrix [35]. Fourteen of the checking points were evenly extracted to compare the elevation MapMatrix [35]. Fourteen of the checking points were evenly extracted to compare the elevation differences between the automatic and the manual measurements. The results are shown in Table 3. differences between the automatic and the manual measurements. The results are shown in Table 3. Table 3. Distribution of the elevation differences among sampling points in the power line profile
Table(unit: 3. Distribution of the elevation differences among sampling points in the power line profile (unit: m). m). Elevation of Manual Measurement Elevation Elevation of of the the PLAMEC PLAMEC Elevation Difference Elevation of Manual Measurement Elevation Difference 1261.858 1261.906 0.048 1261.858 1261.906 0.048 1261.571 1261.397 −0.174 1261.571 1261.397 −0.174 1315.674 1315.705 0.031 1315.674 1315.705 0.031 1316.065 1315.822 −0.243 1316.065 1315.822 −0.243 1279.093 1279.090 −0.003 1279.093 1279.090 −0.003 1278.859 1279.044 0.185 1278.859 1279.044 0.185 1248.168 1248.040 −0.128 1248.168 1248.040 −0.128 1248.187 1248.217 0.030 1248.187 1248.217 0.030 1248.440 1248.560 0.120 1248.440 1248.560 0.120 1247.950 1247.721 −0.229 1247.950 1247.721 −0.229 1264.729 1264.682 −0.047 1264.729 1264.682 −0.047 1264.611 1264.793 0.182 1264.611 1264.793 0.182 1298.366 1298.251 −0.115 1298.366 1298.251 −0.115 1298.145 1298.316 0.171 1298.145 1298.316 0.171 Root mean square error of the elevation difference: 0.148 m. Root mean square error of the elevation difference: 0.148 m.
It canItbe seen 3 that 3the RMS the elevation differences between the automatic can be from seen Table from Table that theerror RMSoferror of the elevation differences between the automatic results measurement results and the manual measurement is 0.148 The maximum measurement and the manual measurement results is results 0.148 m. The m. maximum error was was 0.243 m, that which shows that the accuracy of the measurement automatic measurement is consistent 0.243 error m, which shows the accuracy of the automatic is consistent with thewith manual the manual measurement method. We infer that our proposed approach can replace the manual of measurement method. We infer that our proposed approach can replace the manual measurement measurement of power lines during the inspection of power line corridors. power lines during the inspection of power line corridors. 3.4. Detection of Obstacles within the Power Line Corridor
3.4. Detection of Obstacles within the Power Line Corridor We use 6.5 m as the safe distance threshold between the power line and ground objects. As a
We use 6.5 m as the safe distance threshold between the power line and ground objects. As a result, result, we detect eight obstacles, as shown in Figure 16. we detect eight obstacles, as shown in Figure 16.
Remote Sens. 2017, 9, 824 Remote 2017, 9, 824 Remote Sens.Sens. 2017, 9, 824
T115 T113 T116T115 T116 T112 T114T113 T112 T114
16 of 19 1616 ofof 1919
T111 T111
T110 T110
T109 T109
T108
T108
T110
T110
Power
Power Obstacles
Obstacles Figure 16. Three-dimensional reconstruction of the power line corridor and the distribution of obstacles. Figure 16. of the line corridor and theand distribution of obstacles. Figure 16. Three-dimensional Three-dimensionalreconstruction reconstruction of power the power line corridor the distribution of
obstacles. In Figure 16, the ground 3D point clouds were extracted using the SPMEC, while the blue,
In Figure the ground 3D point cloudsusing were extracted usingThe the SPMEC, while blue, orange, orange, and16, white wires were measured the PLAMEC. red points are the obstacles. The In Figure 16, the ground 3D point clouds were extracted using the SPMEC, while the andtowers white wires were measured using the PLAMEC. The red points are obstacles. The towers with with existing 3D models were included in the scene to make it more intuitive. Taking blue, the orange, and white wires were measured using the PLAMEC. The red points are obstacles. existing 3D models were included in the scene to make it more intuitive. Taking the power line corridor power line corridor between T109 and T110 in the experimental area as an example, Figure The 17 towers with 3Dinmodels included in the scene toFigure make17 it shows more intuitive. the between T109 and T110 theofexperimental area an example, the profileTaking analysis of shows theexisting profile analysis thewere power line andas the ground. power lineline corridor between the power and the ground.T109 and T110 in the experimental area as an example, Figure 17
shows the profile analysis of the power line and the ground. Elevation (m)
1319.0 1304.4
(a)
Elevation (m)
1319.0 1289.8 1304.4 1275.2
(a)
1289.8 1260.6 1275.2 1246.0 1260.6
Distance (m)
Distance (m)
1246.0
4.5 28.5 52.5 76.5 100.5 124.5 148.5 172.5 196.5 220.5 244.5 268.5 292.5 316.5 340.5 364.5 388.5 412.5 Span of T109-T110
4.5 28.5 52.5 76.5 100.5 124.5 148.5 172.5 196.5 220.5 244.5 268.5 292.5 316.5 340.5 364.5 388.5 412.5 33.0 27.2
(b) Span of T109-T110
33.0 21.4 15.6 (b) 27.2 9.8 21.4 4.0 15.6 4.5 28.5 52.5 76.5 100.5 124.5 148.5 172.5 196.5 220.5 244.5 268.5 292.5 316.5 340.5 364.5 388.5 412.5 9.8
Span of T109-T110
4.0
4.517.28.5 52.5 76.5 100.5 124.5analysis 148.5 172.5 196.5of220.5 244.5 268.5 316.5between 340.5 364.5 Figure Diagram of the profile results the power line292.5 corridor T109388.5 and412.5 T110. Diagram theand profile of the power linethe corridor and T110. (a) Figure Profile 17. of the powerofline theanalysis ground;results (b)Span Distance between powerbetween line andT109 the ground. of T109-T110 (a) Profile of the power line and the ground; (b) Distance between the power line and the ground.
Figure 17. Diagram of the profile analysis results of the power line corridor between T109 and T110. (a) Profile of the power line and the ground; (b) Distance between the power line and the ground.
Remote Sens. 2017, 9, 824
17 of 19
Figure 17a represents the profile of the power line corridor, where the horizontal axis shows the distance to T109 and the vertical axis shows the elevation. Figure 17b represents the distance between the power line and the ground, where the horizontal axis shows the distance to T109 and the vertical axis shows the distance. It can be seen from Figure 17b that, when the safe distance between the power line and the ground is set to 6.5 m, an obstacle that is 388.5 m away from T109 can be detected. The positions of the eight obstacles and the distances between them and the power line are listed in Table 4. To verify the reliability of the obstacle detection results, a field investigation was conducted. The distances between the obstacles and the power line were measured using TruPulse 200, a laser range finder produced by Laser Technology, Inc. [36]. These results are also listed in Table 4. Table 4. Distribution of obstacles within the power line corridor (safety distance threshold: 6.5 m). Tower Section
Position of the Obstacles (m)
T108~T109 T109~T110 T110~T111 T111~T112 T112~T113 T115~T116
674.0–679.0 375.5–400.0 84.0–116.0 142.0–144.5 93.5–96.0 112.0–122.5 584.0–600.0 235.0–247.0
Distance Between the Obstacles and the Power Line (m) Proposed Method
Field Measurement
Distance Difference
5.945 4.626 5.012 5.820 6.021 5.161 6.113 5.889
6.3 4.8 5.4 5.4 5.9 5.3 5.9 5.5
0.355 0.174 0.388 −0.420 −0.121 0.139 −0.213 −0.389
Root mean square error of the distance difference: 0.319 m.
In Table 4, the first column is the identifications of the towers, the second column represents the distances between the obstacles and the first tower, and the third column gives the results of the distances between the obstacles and the power line obtained using the proposed method and the field measurements. The differences between the distances obtained using the two methods are also given in the fourth column. As shown in Table 4, the distances between the obstacles and the power lines obtained by the proposed method conform to the field measurement results. The RMS error of the differences between the distances was 0.319 m, and the maximum error was 0.420 m. While verifying the distance between obstacles and power lines, we use a laser range finder instead of a total station, because a laser range finder is easier to operate. Another reason for using a laser range finder is that, even when using a total station, it is very hard to determine the location of a point on a power line with a minimum distance to the canopy. Conversely, a canopy is a large object and therefore it is a difficult task to determine the location of a point on a canopy when the minimum distance to the power line depends on the viewpoint of an operator. Thus, we followed the traditional specifications, and used a laser range finder to measure multiple times to confirm the detected obstacles in our proposed method. The accuracy of the laser range finder used in the experiments is about 0.3 m [37], which is consistent with the proposed method. Thus, it cannot be used as ground truth to check the accuracy of the proposed obstacle detection method. Although the method is not designed to assess the accuracy of the measured distance between obstacles and power lines, the distance measured by this method is consistent with the traditional methods, and meets the requirements (better than 2 m) for the inspection of obstacles within power line corridors. To further verify whether there are undetected obstacles in the experimental power line corridors, we manually checked all the stereos in the stereo measurement system, and found that no obstacle was missed in the automatic detection, which meet the needs of power line inspection. 4. Conclusions In this paper, we proposed an automatic inspection method for power lines using UAV images. We focused on solving three problems: the automatic measurement of power lines, the automatic 3D reconstruction of the ground along the power line corridor, and the automatic recognition of obstacles. We proposed the PLAMEC for automatic measurements of power lines, based on the
Remote Sens. 2017, 9, 824
18 of 19
principle that the intersection point of the epipolar line and the corresponding power line must be the corresponding image point on the power line. The SPMEC algorithm was adopted for a 3D reconstruction of the ground along the power line corridor, and the obstacles could be detected automatically by calculating the distance between the 3D vectors of the power lines and the ground point clouds. The results showed that the success rate of the automatic measurement of power lines was 93.2%, and the accuracy was consistent with those of manual measurements, indicating that it can replace the manual measurement of 3D coordinates of power lines. The height RMS error of the point cloud was 0.233 m and the RMS error of the power line was 0.205 m. During the test, when the distance threshold was set to 6.5 m, eight obstacles were found. The distances between the obstacles and the power lines conformed approximately to the field-measured results; the measured distance difference between the two methods was better than ±0.5 m. Due to the limitations inherent to actual application scenarios on the ground, at this point, we cannot acquire the true distance between the obstacles and power lines to assess the true distance measurement accuracy, which will be addressed in our future studies. Acknowledgments: This work was supported by the National Natural Science Foundation of China (Grants. 41371432, 41401522) and the National Key Developing Program for Basic Sciences of China (Grant No. 2012CB719902). Author Contributions: Yong Zhang developed the algorithm, conducted the primary data analysis, and crafted the manuscript. Xiuxiao Yuan advised on the data analysis provided remote sensing experience and significantly edited the manuscript. Wenzhuo Li and Shiyu Chen carried out the experiments and revised the manuscript. Conflicts of Interest: The authors declare no conflict of interest.
References 1.
2. 3.
4. 5.
6. 7.
8. 9. 10. 11. 12.
Miralles, F.; Pouliot, N.; Montambault, S. State-of-the-art review of computer vision for the management of power transmission lines. In Proceedings of the IEEE International Conference on Applied Robotics for the Power Industry (CARPI), Foz do Iguassu, Brazil, 14–16 October 2014. Ahmad, J.; Malik, A.S.; Abdullah, M.F.; Kamel, N.; Xia, L. A novel method for vegetation encroachment monitoring of transmission lines using a single 2D camera. Pattern Anal. Appl. 2015, 18, 419–440. [CrossRef] Matikainen, L.; Lehtomäki, M.; Ahokas, E.; Hyyppä, J.; Karjalainen, M.; Jaakkola, A.; Kukko, A.; Heinonen, T. Remote sensing methods for power line corridor surveys. ISPRS J. Photogramm. Remote Sens. 2016, 119, 10–31. [CrossRef] Zhang, J.X.; Huang, G.M.; Liu, J.P. SAR remote sensing monitoring of the Yushu earthquake disaster situation and the information service system. J. Remote Sens. 2010, 14, 1–046. Eriksson, L.E.B.; Fransson, J.E.S.; Soja, M.J.; Santoro, M. Backscatter signatures of wind-thrown forest in satellite SAR images. In Proceedings of the IEEE International Geoscience and Remote Sensing Symposium (IGARSS), Munich, Germany, 22–27 July 2012. Kobayashi, Y.; Karady, G.G.; Heydt, G.T.; Olsen, R.G. The utilization of satellite images to identify trees endangering transmission lines. IEEE Trans. Power Deliv. 2009, 24, 1703–1709. [CrossRef] Mills, S.J.; Castro, M.P.G.; Li, Z.; Cai, J.; Hayward, R.; Mejias, L.; Walker, R.A. Evaluation of aerial remote sensing techniques for vegetation management in power-line corridors. IEEE Trans. Geosci. Remote Sens. 2010, 48, 3379–3390. [CrossRef] Sun, C.; Jones, R.; Talbot, H.; Wu, X.; Cheong, K.; Beare, R.; Buckley, M.; Berman, M. Measuring the distance of vegetation from powerlines using stereo vision. ISPRS J. Photogramm. Remote Sens. 2006, 60, 269–283. [CrossRef] Tabor, A.; Ottesen, B.S. Advances in applications for aerial infrared thermography. Proc. SPIE 2006, 156, 3642–3643. Guo, B.; Li, Q.; Huang, X.; Wang, C. An improved method for power-line reconstruction from point cloud data. Remote Sens. 2016, 8, 36. [CrossRef] Zhu, L.; Hyyppä, J. Fully-automated power line extraction from airborne laser scanning point clouds in forest areas. Remote Sens. 2014, 6, 11267–11282. [CrossRef] Mclaughlin, R.A. Extracting transmission lines from airborne LIDAR data. IEEE Geosci. Remote Sens. Lett. 2006, 3, 222–226. [CrossRef]
Remote Sens. 2017, 9, 824
13. 14. 15. 16.
17. 18. 19. 20. 21. 22.
23. 24. 25. 26. 27. 28. 29.
30. 31. 32. 33. 34. 35. 36. 37.
19 of 19
Ax, M.; Thamke, S.; Kuhnert, L.; Kuhnert, K.-D. UAV based laser measurement for vegetation control at high-voltage transmission lines. Adv. Mater. Res. 2012, 614–615, 1147–1152. [CrossRef] RIEGEL. Available online: http://www.riegl.com/products/unmanned%E2%80%90scanning/ (accessed on 20 July 2017). Velodyne LiDAR. Available online: http://velodynelidar.com/products.html (accessed on 20 July 2017). Toth, J.; Gilpin-Jackson, A. Smart view for a smart grid—Unmanned Aerial Vehicles for transmission lines. In Proceedings of the 2010 1st International Conference on Applied Robotics for the Power Industry (CARPI), Montreal, QC, Canada, 5–7 October 2010. Cai, J.; Walker, R. Height estimation from monocular image sequences using dynamic programming with explicit occlusions. IET Comput. Vis. 2010, 4, 149–161. [CrossRef] Jó´zków, G.; Vander Jagt, B.; Toth, C. Experiments with UAS imagery for automatic modeling of power line 3D geometry. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2015, XL-1/W4, 403–409. Jiang, S.; Jiang, W.; Huang, W.; Yang, L. UAV-based oblique photogrammetry for outdoor data acquisition and offsite visual inspection of transmission line. Remote Sens. 2017, 9, 278. [CrossRef] Deng, C.; Wang, S.; Huang, Z.; Tan, Z.; Liu, J. Unmanned Aerial Vehicles for power line inspection: A cooperative way in platforms and communications. J. Commun. 2014, 9, 687–692. [CrossRef] Zhang, Y.; Yuan, X.; Fang, Y.; Chen, S. UAV low altitude photogrammetry for power line inspection. ISPRS Int. J. Geo-Inf. 2017, 6, 14. [CrossRef] Ramesh, K.N.; Murthy, A.S.; Senthilnath, J.; Omkar, S.N. Automatic detection of powerlines in UAV remote sensed images. In Proceedings of the 2015 International Conference on Condition Assessment Techniques in Electrical Systems (CATCON), Bangalore, India, 10–12 December 2015. Yan, G.; Li, C.; Zhou, G.; Zhang, W.; Li, X. Automatic extraction of power lines from aerial images. IEEE Geosci. Remote Sens. Lett. 2007, 4, 387–391. [CrossRef] Zhang, Z.X.; Zhang, J.Q. Digital Photogrammetry; Wuhan University Press: Wuhan, China, 1997. Serra, J. Image Analysis and Mathematical Morphology; Academic Press: Orlando, FL, USA, 1982. Ren, M.; Yang, J.; Sun, H. Tracing boundary contours in a binary image. Image Vis. Comput. 2002, 20, 125–131. [CrossRef] Furukawa, Y.; Ponce, J. Accurate, dense, and robust multi-view stereopsis. IEEE Trans. Pattern Anal. Mach. Intell. 2010, 32, 1362–1376. [CrossRef] [PubMed] Zhang, L.; Gruen, A. Multi-image matching for DSM generation from IKONOS imagery. ISPRS J. Photogramm. Remote Sens. 2006, 60, 195–211. [CrossRef] Hirschmuller, H. Accurate and efficient stereo processing by semi-global matching and mutual information. In Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), San Diego, CA, USA, 20–26 June 2005. China Southern Power Grid. Available online: http://www.gz.csg.cn/web/news/43.html (accessed on 6 April 2017). k2 -Photogrammetry. Available online: http://www.k2-photogrammetry.de/products/patb.html (accessed on 15 June 2017). Lowe, D.G. Distincitive image features from scale-invariant keypoints. Int. J. Comput. Vis. 2004, 60, 91–110. [CrossRef] Gruen, A. Adaptive least squares correlation: A powerful image matching technique. S. Afr. J. Photogramm. Remote Sens. Cartogr. 1985, 14, 175–187. Teunissen, P.J.G.; Khodabandeh, A. Review and principles of PPP-RTK methods. J. Géod. 2015, 89, 217–240. [CrossRef] Visiontek Inc. Available online: http://en.visiontek.com.cn/pp1/newsCategoryId=37.html (accessed on 18 July 2017). Laser Technology Inc. Available online: http://www.lasertech.com (accessed on 6 June 2017). Laser Technology Inc. Available online: http://www.lasertech.com/ftp/techdocs/TruPulse_Height_ Performance.pdf (accessed on 20 July 2017). © 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
