2013 IEEE 4th Control and System Graduate Research Colloquium, 19 - 20 Aug. 2013, Shah Alam, Malaysia
Tree Diameter Measurement Using Single Infrared Sensor for Non-Stationary Vehicle Context in Agriculture Field Norashikin M. Thamrin, Nor Hashim Mohd. Arshad, Ramli Adnan, Rosidah Sam, Noorfadzli Abdul Razak, Mohamad Farid Misnan and Siti Fatimah Mahmud Faculty of Electrical Engineering Universiti Teknologi MARA 40450 Shah Alam, Selangor, Malaysia
[email protected] Abstract— One of the important aspects in building an autonomous vehicle with the cognitive capability in agriculture field is the potentiality of it to recognize the trees or significant landmarks and map it for future navigation process. Therefore, it is crucial for a machine to detect the trees and keep tracking between them in order to perform the plantation activities autonomously. One of the challenges faced by this implementation is the need of a real-time tree detection technique which will not burden the processor as well as lightweight hardware that suits the needs of an autonomous aerial vehicle. Due to that, this paper presents a new and relatively simple tree diameter measurement technique using a high-performance and non-intrusive infrared sensor. The experiments are performed with a various coloured pole at a static velocity implementation to develop a promising technique that is useful and promising in this context.
II.
Measurement of tree diameter for forest harvester is discussed by Jaakko et al [3] in their research. They implement a SICK 2D laser range finder as the primary sensor which has been mounted on an ATV. In their method, a validation algorithm is introduced to accept an only tree feature which is a bare tree trunk without branches, and diminish the significant gaps between them and other clusters. By using a rotational 2D laser scanner, the circle of a tree can be estimated by utilizing only the edges point of the tree as well as its shortest range measurement. However, in their experiment, the average motion of the ATV is 15mm between two scans. They reported that the mean of the error in diameter is 6 mm and standard deviation is 22 mm with the relative error less than 4% with this approach. Similar approach has been carried out by YuShin Chou et al [4] in deciding the width as well as height measurement of an object by implementing additional 3D laser range sensing technique on a stationary and moving platform. According to their method, the laser range finder is rotated to perform a total of 500 scans, one scan for each 1.8 degree of rotation starting from 0 degree to 90 degree forward. They reported that, the width discrepancies between the three measured boxes dimension and the real measurement are 7.9mm, 4.5mm and 4mm due to errors inherent in the laser range finder and the linkage transformation of the rotating bar.
Keywords- infrared detection, edge detection, diameter measurement
I.
INTRODUCTION
Advance techniques in automation are highly demanded in the agriculture field to aid in reducing the costs of labours and production as well as projects the positive impacts of greater net income and quality crops in this sector. However, introducing self-governing mobile robots in commercial farm or forested environment requires a specific and peculiar motion control strategy, object identification scheme, constructive and accommodative local map as well recognizing local position within the map. Simultaneous Localization and Mapping (SLAM) is one of the technique which allows a mobile robot to build a map of the environment and at the same time localize itself in this map to determine its location [2]. One of the important characteristic in building a constructive map is landmark observation. In this context, the existence natural landmark in the structured and facilitative agriculture fields is the plants or crop itself. Therefore, in this paper, a method is suggested to recognize the targeted tree is by measuring the diameter of it to distinguish it from others based on the appropriate threshold value.
978-1-4799-0552-2/13/$31.00 ©2013 IEEE
PREVIOUS IMPLEMENTATION ON DIAMETER MEASUREMENT
A diameter measurement based on laser triangulation method for spherical object has been explored by Michael et al [5]. In their method, the diameter of a spherical object is determined based on the laser triangulation technique under structured lighting which creates laser plane lines on the object under test and these lines are observed by the charge-coupled device camera. The radius of the sphere is calculated by two approaches, namely, (1) three-tangent method and (2) gradientdescent method. In their technique, a PC-connected preprototype with a 10-m W 670-nm laser diode and a monochrome CCD camera are used. They found that the threetangent method shows the best results with relative error below 1%.
38
2013 IEEE 4th Control and System Graduate Research Colloquium, 19 - 20 Aug. 2013, Shah Alam, Malaysia
A sensor fusion of vision sensor and ultrasonic has been proposed by Muhammad et al [6] to detect obstacle and measure its size. According to their method, the size of the object is determined by computing the occupied pixels with the respect to the distance between the camera and the object as well as respect to the real size of the object by fixing the view angle of the camera. Once the geometric similarity is identified between the real object and the image, the equivalent ratio is used to calculate the sizes, which are the width and the height of the detected object. Jiangming Kan et al [7] has researched the means of trunk and branch diameter measurement of a standing tree based on computer vision technique which is really important working parameters of the intelligent pruning robots. They presented the six steps of extracting the trunk and branch parameters which include the use of a CCD camera to acquire the trunk and branch images, pixels counting with the respect of the real size of each pixels to the real measurement of it, intersection of trunks and branches detection means as well as the real diameter determination algorithm of the said trunks and branches. They reported that, with their technique, the experiments show that the mean absolute error is 0.67cm and the mean relative error is 1.9%. However, they stated that the complex background of the trunks and branches can deteriorate their detection severely. Grape fruit diameter measurement by using a non-contact optical method has been carried out by Qingbing Zeng et al[8]. The images which are gained through a CMOS camera undergo several image processing algorithms such as filtering, image segmentation and occupied pixels calculation to determine the exact diameter of the fruit. They reported that the repeatability accuracy of the diameter measurement by the vision sensor is ±7 µm.
III.
INFRARED SENSOR
Infrared (IR) can be used either to detect obstacle or measure distance by providing a binary output or analog outturn to the respective application. In our case, a GP2Y0A02YK0F IR sensor produced by Sharp is used to detect the edge of the object and maintaining the distance between the mobile vehicle and the target. In our case, the distance between the sensor and the targeted object is maintained approximately at 80 cm. However, this sensor is known to be unstable under direct sunlight explosion [1], but under shady environment such as canopied environment e.g. papaya, palm oil and rubber fields, the capability of it is still acceptable. In our approach, the edge of the tree is detected by non-stationary detection means via IR sensor which is mounted on a mobile vehicle, that later calculates the diameter of the tree by substituting current location from the previous known position. Therefore, in this case, it advised by [1] to set the sensor that the moving direction of the object and the line between emitter and detector centers are vertical as illustrated in Fig.2. This technique decreases the deviation of measurement distance by moving direction of the mobile robot. This method is selected preliminarily as it offers a great advantage in enforcing real-time tree detection for a nomadic aerial vehicle in the agriculture field which contradicts from what have been offered in the earlier mentioned techniques, fact that most of the diameter measurement is done stationary.
All of the above techniques mentioned, has significantly contributed to the diameter measurement of an object to be implemented in the autonomous machine. Nevertheless, those approaches requires extensive matrices computation [4], bulky sensor dimension[3-5], expensive image processing time[6-7] as well as lighting condition [8] which is not suitable to be implemented on an autonomous aerial vehicle for real-time individual tree detection method in a canopied and structured environment such as papaya plantation in Fig. 1. Thus, in this paper, a less complex technique based on dead reckoning approach is proposed to suit the need of real-time tree detection for an unmanned aerial vehicle implementation.
Figure 2.
Infrared detection technique on a moving platform[1]
IV.
METHODOLOGY
A. Hardware Testing Platform
Fig. 3 shows the components involved in the hardware development board in testing the diameter measurement means of an object. It consists of a specific microcontroller from dsPIC30F family, high performance IR sensor which can provide detection range between 20cm to 150cm with 1-cm resolution, a sonar range sensor with a measurement range of 0mm to 5500mm with 1-mm resolution, DC motors and the drivers, LCD display for showing the sensor reading values, switches, auto-calibration line following sensors (series of IR sensors) as well as XBee wireless board for online data acquisition through personal computer. On the other hand, the dsPIC30F microcontroller is equipped with a line following algorithm that allows the designed robot to move in a controlled straight line during the data capturing process. An ADC is used to convert the analog values received from the IR range finders into understandable digital values which later converted into a meaningful SI unit. For communication
Figure 1. Papaya Plantation Field.
39
2013 IEEE 4th Control and System Graduate Research Colloquium, 19 - 20 Aug. 2013, Shah Alam, Malaysia
protocol controller between development board and personal computer, a wireless Universal Asynchronous Receive Transmit (UART) protocol is utilized. There are two interrupts used in this platform, namely, timer and UART interrupts. Timer interrupt is used to capture the interval times during sensor readings. The latter is used to send motion control signals from personal computer to the robot. A Pulse Width Modulation (PWM) is designed to provide adequate speed controller to the motor to ensure the robot can follow the provided line track. Fig. 4 shows the omnidirectional robot used to test the proposed detection means of tree. Omnidirectional orientation is applied due to its similarity of non-holomonic movement of a autonomous aerial vehicle such as quadrocopter.
B. Testing Setup The experiments are done in a controlled laboratory environment under ambient light condition produced by a fluorescent lamp. A cylindrical pole is used as the tree to realize the proposed diameter measurement technique which will be implementing in real-time tree detection in agriculture field later. A straight line track with 3.5m long is provided to facilitate the movement of the autonomous omnidirectional mobile robot. The velocity of the vehicle is strictly consistent at 0.215 ms-1 during the detection test. These approaches are used to eliminate other unwanted disturbances which can decrease the effectiveness and persistence in the sensor reading. The diameter of the pole is approximately 9 cm and placed 80 cm away from the line track. This experiment is done in a free space to produce a free-cluttered and almost ideal profile of perpendicular object detection with a 4 different of coloured poles namely, light green, dark green, dark cocoa and dark grey. These colours are chosen due to various colours available on the bare tree of the example agriculture fields like papaya. Fig. 5 shows the overall experiment setup for perpendicular object detection and diameter measurement. Later, this experiment is extended in differentiating the target pole from other various diameter of cylindrical through this method as depicted in Fig. 6. The IR sensor is placed 90 degree towards the object for perpendicular object detection. The details of the
dsPIC30F Dev Board INFRARED SENSOR MOTOR & MOTOR DRIVERS
LINE FOLLOWING ALGORITHM DATA ACQUISITION INTERRUPT
TIMER
LCD UART
PWM
ADC
SWITCHES
ULTRASONIC SENSOR
AUTO-CALIBRATION LINE FOLLOWING SENSOR
9 cm WIRELESS COMM. PROTOCOL
Figure 1.
Hardware Testing Platform 80 cm
Figure 5
Testing environment
result are discussed in the next section.
B
C
A
Figure 4 Aotonomous omnidirectional mobile robot platform for testing the proposed technique Figure 2 Obstacle experiment. From left: Pole A with diameter of 1 cm, Pole B with diameter of 2 cm and Pole C (object under test) with diameter of 9 cm.
40
2013 IEEE 4th Control and System Graduate Research Colloquium, 19 - 20 Aug. 2013, Shah Alam, Malaysia
V.
RESULT & DISCUSSION
The experiment is successfully performed on a controlled line tracking operation with an omnidirectional mobile robot in a clear environment. The object is placed at approximately 82 cm from the line track, which is total of 80 cm from the sensor surface, to realize the perpendicular object diameter measurement by the mobile robot. These preliminary experiments show the capableness and effectiveness of the object diameter measurement based on edge detection perpendicularly by the IR sensor that will benefit the usage of it in detecting individual tree in the agriculture field. Fig. 7 shows the diameter measurement of a tree obtained from the IR sensor based on this approach. Referring to this figure, the diameter, d, of the tree can be estimated by detecting the edges of it, E1 and E2, through the sensor. In order to calculate d, the total elapsed time during consecutive two edge detection is multiplied with the constant velocity, v. In this experiment, the capability of the sensor is limited programmatically up to 150 cm only even though the longest range can be detected is 550cm, due to the lack of space available in the laboratory in order to differentiate the object under test from its background. Therefore, the limitation of the range favors the experiment to be performed to acquire the ideal diameter measurement of the pole with the average detection of the object from the sensor is ± 3.35 cm.
Figure 3. Histogram of relative error in pole diameter measurement based on various colours
error less than 0.2 cm for each colours with average of 0.2 cm in standard deviation. Even though IR sensor is not the best solution to be used as range detection in outdoor application or under direct sunlight as well as ambient light , but the narrow beam width offers through it still preferable to detect the proximity of an obstacle like the edge of the pole in this case. Therefore, the use of the IR sensor must be complemented by other sensor which can detect the range of the object accurately. TABLE I. SUMMARY FIGURE OF POLE DIAMETER MEASUREMENT AT VARIOUS COLOURS : LIGHT GREEN & DARK GREEN mean measurement mean ρR %mean ρR σ
d E1
Figure 4.
Light Green
Dark Green
0.08 0.02 83.17 0.02
0.08 0.02 83.17 0.02
TABLE II. SUMMARY FIGURE OF POLE DIAMETER MEASUREMENT AT VARIOUS COLOURS : DARK COCOA & DARK GREY
E2
mean measurement mean ρR %mean ρR σ
Tree diameter measurement based on the edge detection technique
Fig. 8 shows the histogram of relative error result obtained from the experiment of four different colours of pole. This experiment is repeated 10 times for each colour. From the graph, it can be seen that, 72.5% out of the experiment exhibits relative error from the expected diameter measurement less than or equal with 0.2 cm. The rest of the other results give relative error approximately between 4 to 5 cm, that is considered above the average of permitted diameter measurement. This drawback is due to the reason of technical problem occurs during the experiment such as instability of omni tires motion and sensor defect. When this happens, the edge of the pole cannot be detected properly and corrective action is necessery.
Dark Cocoa
Dark Grey
0.10 0.01 83.54 0.02
0.09 0.01 88.15 0.02
In the obstacle experiment, that is used to discriminate the target pole from other various diameter poles, it can be seen that the probability of the TRUE condition is approximately 64% from the overall tests. TRUE condition shows that the target pole with diameter of 9 cm can be discriminate effectively through this method from other poles. Otherwise, FALSE condition is produced. The pass percentage of probability of recognizing the pole is still low and this experiment is only testing the individual pole. It can be improved by introducing distance threshold between two consecutive poles or other possible conditions. Consequently, these experiments show another option in measuring the diameter of the pole or tree based on dead reckoning technique which significantly contributes in realtime tree detection technique for UAV implementation in the agriculture field. However, further study must be drawn to
Table I and II shows the overall summary of the effectiveness of the proposed technique based on various colours existed on the bare trunk in estimating the diameter of a tree in agriculture field. It can be observed that more than 80% out of the experiments show a good result in lower relative
41
2013 IEEE 4th Control and System Graduate Research Colloquium, 19 - 20 Aug. 2013, Shah Alam, Malaysia
increase the accuracy of the tree diameter measurement as well as the probability of tree recognizing based on the measured parameter. VI.
REFERENCES [1] [2]
SUMMARY
From the experiments, it can be concluded that the diameter of a pole with various colours can be measured through the computation based on the consecutive edge detection made by a non-intrusive infrared sensor at a straight motion of mobile vehicle with a fix velocity of 0.215 ms-1. The result shows that each colour produces more than 80% pole diameter measurement from the overall tests. Unfortunately, the probability test that is used to differentiate the pole from other obstacles gives a moderate estimation of 64% only. However, this experiments needs to be explored deeply to increase the probability of tree recognition as well as measurement accurateness and removes any unwanted parameter that can cause faulty to the diameter measurement. Nevertheless, these preliminary experiments inspire a useful technique in realizing real-time tree detection in the agriculture field for aerial vehicle implementation.
[3]
[4]
[5]
[6]
[7]
ACKNOWLEDGMENT This research is sponsored by Research Incentive Faculty (RIF) Fund, Universiti Teknologi MARA (Contract number: 600-RMI/DANA 5/3/RIF (7/2012)). We thank the Research Management Institute (RMI), UiTM, FKE and Hulu Selangor Agriculture Department for their support.
[8]
42
S. Corporation, "GP2Y0A02YK0F : Distance Measuring Sensor Unit Measuring distance: 20 to 150 cm Analog output type," ed, 2006. T. B. Hugh Durrant-Whyte. (2006, June 2006) Simultaneous Localization and Mapping: Part 1. IEEE Robotics & Automation Magazine. 99-108. K. K. Jaakko Jutila, Arto Visala, "Tree Measurement In Forest by 2D Laser Scanning," in Proceedings of the 2007 IEEE International Symposium on Computational Intelligence in Robotics and Automation, Jacksonville, FL, USA, 2007, pp. 491 496. J.-S. L. Yu-Shin Chou, "A Robotic Indoor 3D Mapping System Using a 2D Laser Range Finder Mounted on a Rotating Four-Bar Linkage of a Mobile Platform," International Journal of Advanced Robotic Systems, INTECH, vol. 10, p. 10, 2013. D. R. Michael Demeyere, Christian Eugene, "Diameter Measurement of Spherical Objects by Laser Triangulation in an Ambulatory Context," IEEE Transaction of Instrumentation and Measurement, vol. 56, pp. 867 - 872, June 2007 2007. F. S. C. Muhammad Wali Ullah Bhuiyan, "Obstacle Detection And Size Measurement For Autonomous Mobile Robot Using Sensor Fusion," Bachelor OF Science In Computer Science And Engineering, Department Of Computer Science And Engineering, Islamic University Of Technology (IUT), 2012. B. W. L. R. S. Jiangming Kan ; Beijing Forestry Univ., "Automatic measurement of trunk and branch diameter of standing trees based on computer vision," in 3rd IEEE Conference on Industrial Electronics and Applications, 2008. ICIEA 2008. , Singapore, 2008, pp. 995 - 998. C. L. Qingbing Zeng, Yubin Miao, Shengwei Fei, Shiping Wang, "A Machine Vision System for Continuous Field Measurement of Grape Fruit Diameter " in Second International Symposium on Intelligent Information Technology Application, 2008. IITA '08. , Shanghai 2008, pp. 1064 - 1068.