Visualization of Concrete Slump Flow Using the Kinect Sensor - MDPI

sensors Article

Visualization of Concrete Slump Flow Using the Kinect Sensor Jung-Hoon Kim 1, * 1 2

*

ID

and Minbeom Park 1,2

Construction Robot and Automation Laboratory, Department of Civil & Environmental Engineering, Yonsei University, Seoul 03722, Korea; [email protected] Engineering & Construction Group, Samsung C&T Corporation, Seoul 13530, Korea Correspondence: [email protected]; Tel.: +82-2-2123-5804; Fax: +82-2-364-5300

Received: 7 February 2018; Accepted: 27 February 2018; Published: 3 March 2018

Abstract: Workability is regarded as one of the important parameters of high-performance concrete and monitoring it is essential in concrete quality management at construction sites. The conventional workability test methods are basically based on length and time measured by a ruler and a stopwatch and, as such, inevitably involves human error. In this paper, we propose a 4D slump test method based on digital measurement and data processing as a novel concrete workability test. After acquiring the dynamically changing 3D surface of fresh concrete using a 3D depth sensor during the slump flow test, the stream images are processed with the proposed 4D slump processing algorithm and the results are compressed into a single 4D slump image. This image basically represents the dynamically spreading cross-section of fresh concrete along the time axis. From the 4D slump image, it is possible to determine the slump flow diameter, slump flow time, and slump height at any location simultaneously. The proposed 4D slump test will be able to activate research related to concrete flow simulation and concrete rheology by providing spatiotemporal measurement data of concrete flow. Keywords: visualization of flowable concrete; 3D depth sensor; Kinect; digital image processing; coordinate transform; slump flow test; concrete workability test; structural health monitoring during construction; automated measurement in construction

1. Introduction Monitoring the workability of freshly mixed concrete is necessary to ensure that the concrete is properly placed during construction and that adequately hardened strength is achieved after construction [1]. Both are related to structural integrity, safety, and construction productivity. For the safe and easy construction of high-quality concrete structures, it is necessary to use fresh concrete with proper workability. Workability such as flowability, viscosity, yield stress, and resistance to material separation required at construction sites depends on the type of target structure, the spacing of reinforcements, dimensions, cross-sectional shape, construction method, and concrete pumping distance [2]. Monitoring of concrete workability [3,4], as well as strength [5–7], is regarded as a key technology in the long-distance transport of concrete required for high-rise buildings and long-span bridge construction. In the field of innovative concrete 3D printing technology in the construction area, workability control will be critical to ensure consistent material properties and structural performance [8–10]. In the field of concrete materials, highly flowable self-consolidating concrete (SCC) is attracting attention for its ability to fill congested rebar spaces in a formwork under its own weight without using a vibrator. The characteristics of SCC are filling ability, passing ability, and stability [11]. The use of SCC can increase the construction efficiency because it can improve the quality of a concrete structure, reduce the labor requirement, and increase the construction speed. The use of SCC is expected to increase as the complexity of building design and the number of high-rise buildings increase [4]. Sensors 2018, 18, 771; doi:10.3390/s18030771

www.mdpi.com/journal/sensors

Sensors 2018, 18, x

2 of 15

use of SCC can increase the construction efficiency because it can improve the quality of a concrete structure, Sensors 2018,reduce 18, 771 the labor requirement, and increase the construction speed. The use of SCC 2 of is 16 expected to increase as the complexity of building design and the number of high-rise buildings increase [4]. As aa standard standard test test method method to to assess assess the the filling filling ability ability of of flowable flowable concrete, concrete, the the slump slump flow flow test test As shown in Figure 1 has been established [12–14]. It is the simplest and most widely used test method for shown in Figure 1 has been established [12–14]. It is the simplest and most widely used test method SCC [1]. [1]. TheThe result of the flowflow test istest recorded by measuring the diameter of the circular spread for SCC result of slump the slump is recorded by measuring the diameter of the circular of the concrete after lifting up the slump cone. As non-mandatory information, relative viscosity can spread of the concrete after lifting up the slump cone. As non-mandatory information, relative be measured during the slump flow test [13]. The T50 value indicating the relative viscosity is the time viscosity can be measured during the slump flow test [13]. The T50 value indicating the relative for the outer thethe spreading concrete mass to reach a diameter ofreach 500 mm. Most test viscosity is theedge timeoffor outer edge of the spreading concrete mass to a diameter of methods, 500 mm. such as the slump flow test and the slump test [15], are carried out by operators using a ruler or Most test methods, such as the slump flow test and the slump test [15], are carried out by operators a stopwatch, measurement errors are inherent errors and theare measured maymeasured vary depending using a ruler and or ahence stopwatch, and hence measurement inherentvalue and the value on the operator. There is also a possibility that the obtained value can be wrongly recorded easily may vary depending on the operator. There is also a possibility that the obtained valueorcan be manipulated after measurement. If concrete with insufficient workability is used for the construction wrongly recorded or easily manipulated after measurement. If concrete with insufficient workability ofused structures of inaccurate measurement results or incorrect data records, there will be future is for thebecause construction of structures because of inaccurate measurement results or incorrect data problems in terms of structural safety. records, there will be future problems in terms of structural safety.

(a)

(b)

(c)

(d)

(e)

(f)

Figure 1. Slump flow test: (a) Slump cone filled with fresh concrete; (b) Start of slump flow test; Figure 1. Slump flow test: (a) Slump cone filled with fresh concrete; (b) Start of slump flow test; (c) Flow (c) Flow of concrete from cone; (d) Concrete slump after lifting up cone; (e) Spreading concrete slump; of concrete from cone; (d) Concrete slump after lifting up cone; (e) Spreading concrete slump; (f) Final (f) Final concrete slump. concrete slump.

With the development of SCC and high-performance concrete, instead of the conventional test With the development ofintroduced SCC and high-performance concrete, parameters instead of the conventional test devices, a rheometer has been to measure the rheological of concrete [16]. For devices, a rheometer has been introduced to measure the rheological parameters of concrete [16]. example, it has been used for workability monitoring of high-rise buildings [3,4]. Recently, concrete For example, been used for of high-rise [3,4]. Recently, flow based on ita has computational fluidworkability dynamics monitoring (CFD) analysis has been buildings actively studied with a concrete flow based on a computational fluid dynamics (CFD) analysis has been actively studied rheometer [17,18]. However, there is a lack of understanding about concrete rheology in the with a rheometer However, is a lack understanding about in the construction field. [17,18]. Furthermore, withthere the price of aofrheometer ranging fromconcrete $20,000 rheology to $180,000, its construction field. Furthermore, with the price of a rheometer rangingequipment from $20,000 to $180,000, use in the field is limited by its costliness. The workability measuring required in the its use in thefield fieldthus is limited bybeitsable costliness. The workability equipment construction should to quantitatively evaluatemeasuring the workability at lowrequired cost. in the construction field thus should be able to quantitatively evaluate the workability at low cost. a depth In this study, a new framework for qualitatively evaluating concrete workability using Inisthis study, aKinect, new framework qualitatively concrete the workability using a depth sensor proposed. a low-costfor depth sensor, is evaluating utilized to measure dynamically changing sensor is proposed. Kinect, a low-cost depth sensor, is utilized to measure the dynamically changing 3D 3D concrete surface during a slump flow test, as shown in Figure 2. The Kinect was originally concrete surface duringmotion a slumprecognition flow test, asinshown in Figure 2. Theconsole Kinect was originally developed for human the X-box, a game developed by developed Microsoft for human motion recognition in the X-box, a game console developed by Microsoft Corporation [19]. Corporation [19]. Its resolution and its scanning speed are excellent and due to mass production it is Its resolution and its scanning speed are excellent and due to mass production it is inexpensive. Its use inexpensive. Its use is expanding in various fields, such as robotics [20,21], rehabilitation engineering is expanding in various fields, such as robotics [20,21], rehabilitation engineering [22], and even in the construction field in monitoring structural health [23,24]. Visualization of spatiotemporal data using

Sensors 2018, 18, x Sensors 2018, 18, 771

3 of 15 3 of 16

[22], and even in the construction field in monitoring structural health [23,24]. Visualization of spatiotemporal using sensors is useful analysis. Just as the thermography sensors is usefuldata for numerical analysis. Justfor asnumerical the infrared thermography is infrared useful for heat transfer is useful of forconcrete heat transfer analysis of4D concrete the will 4D slump using on Kinect will analysis surfaces [25], the slump surfaces test using[25], Kinect activatetest research concrete activate research on concrete flow simulation. flow simulation. This paper is organized as follows. Section Section 2 presents the data processing algorithm, related to the workability test, which can produce the spreading diameter over time and slump flow time from the Kinect the depth sensor areare represented in the coordinate frame on the Kinect data. data.The Thedata dataacquired acquiredinin the depth sensor represented in the coordinate frame on ground plane by by coordinate transformation, which is is well the ground plane coordinate transformation, which wellused usedinincomputer computergraphics graphics and and robot kinematics. data areare then used to create the the surface of the and kinematics. The Thetransformed transformedpoint pointcloud cloud data then used to create surface of concrete the concrete the is reorganized in aingrid form forfor further data andsurface the surface is reorganized a grid form further dataprocessing. processing.AAfinal final4D 4Dslump slump image image is constructed constructed by collecting collecting the extracted cross-sections cross-sections of of the concrete slump at each instant surface image. In In Section Section 3, 3, the the experimental experimental setup setup for for 4D 4D slump, slump, including including the the test test material material and and procedure, procedure, is described. In In Section Section 4, the experimental experimental results show that concrete flow visualization is possible during the slump flow test by processing the time-varying time-varying surface surface shape shape of of the the concrete. concrete.

Figure 2. Utilization ofof a Utilizationof ofthe thedepth depthsensor sensorduring duringthe theslump slumpflow flowtest; test;recording recordingand andprocessing processing dynamically changing surface a dynamically changing surfaceofoffresh freshconcrete. concrete.

2. 2. Data Data Processing Processing Algorithm Algorithm for for the the 4D 4D Slump Slump Test Test The to acquire acquire spatial spatial information information of the dynamically dynamically changing The Kinect Kinect is is used used to of the changing surface surface of of fresh fresh concrete duringthe theconcrete concreteslump slump flow test. cannot be directly utilized to analyze concrete during flow test. TheThe rawraw datadata cannot be directly utilized to analyze useful useful information on the workability; concrete workability; several stages of dataareprocessing are needed to information on the concrete several stages of data processing needed to produce slump produce slump flow diameter, slump flow time, and slump height. The proposed algorithm for data flow diameter, slump flow time, and slump height. The proposed algorithm for data processing consists processing consists of the following procedures: (1) spatial representation of 3D spatial information the of the following procedures: (1) representation of 3D information at the camera frame {C}atfrom camera frame {C} from the depth image; (2) determination of a ground plane equation; (3) calculation the depth image; (2) determination of a ground plane equation; (3) calculation of a transformation of a transformation the and camera and {S} the where slump the frame {S} where slump matrix between thematrix camerabetween frame {C} the frame slump{C} frame slump cone the is initially cone is initially located; (4) reconstruction of the 3D surface for concrete slump; and (5) cross-section located; (4) reconstruction of the 3D surface for concrete slump; and (5) cross-section extraction and extraction and of a 4D slump image.is Each procedure is described in the following construction of aconstruction 4D slump image. Each procedure described in the following sections. sections. 2.1. Representation of 3D Spatial Information at Camera Frame {C} 2.1. Representation of 3D Spatial Information at Camera Frame {C} While projecting a known speckle pattern of near-infrared light, the Kinect sensor acquires While projecting known pattern of time near-infrared light, thedisparity Kinect sensor acquires disparity disparity imagesaat the IRspeckle camera in real [26], where the d is represented at the images at the IR in real time [26], where the disparity represented the pixel (u, pixel location (u,camera v). Figure 3 represents the pinhole camerad is model, which at shows howlocation the point v). Figure 3 representsspace the pinhole camera model, whichplane showsofhow point inThe three-dimensional in three-dimensional is projected onto the image the the IR camera. point P(X,Y,Z) space is projected onto the image plane of the IRto camera. Thep(u,v) pointon P(X,Y,Z) described in the is camera described in the camera coordinates is projected the point the image plane, which apart coordinates is projected to the point p(u,v) on the image plane, which is apart from the camera frame from the camera frame by a focal length f. Here, the coordinate system of (X,Y,Z) is referenced to by a focal length f. Here, the coordinate system of (X,Y,Z) is referenced to the camera frame {C}. Using the proportionality of two similar triangles OPQ and Opc, the following equations for coordinates X and Y are obtained:

Sensors 2018, 18, 771

4 of 16

Sensors 2018, 18, x

4 of 15

the camera frame {C}. Using the proportionality of two similar triangles OPQ and Opc, the following equations for coordinates X and Y are obtained: (𝑢 − 𝑐𝑥 )𝑍 𝑋= (u−𝑓c x ) Z X= (1) f (1) (𝑣(v− −c𝑐y 𝑦 ) Z)𝑍 𝑌Y== 𝑓f and cyy are are optical optical center center pixels pixels in in the IR camera. The The focal length and where f is focal length and cxx and optical center pixels can be determined by a standard calibration of the camera.

Figure 3. Depth measurement geometry in the pinhole camera model. Figure 3. Depth measurement geometry in the pinhole camera model.

Depth Z is expressed as a nonlinear function of disparity d [21]. The mathematical model Depth Z is and expressed as adnonlinear ofgeometric disparity drelationship [21]. The mathematical between depth disparity is derivedfunction from the in Figure 3, model which between is given depth and disparity d is derived from the geometric relationship in Figure 3, which is given by by Equation (2) [27]: Equation (2) [27]: 𝑍00 Z 𝑍= = (2) Z (2) 𝑍 + Zf 0b0 d𝑑 11 + 𝑓𝑏 where Z is the the distance distance to to the the reference reference plane, plane, bb is is the the baseline baseline between between the the IR IRprojector projector and andcamera. camera. where Z00 is Since the which ranges from 0 to Since the output output value value by by the thesensor sensorisisthe thenormalized normalizeddisparity disparityd’d’ininpractice, practice, which ranges from 0 11 − 1, d should be replaced with md’ + n [27,28] and then Equation (2) becomes as follows: 2 11 to 2 − 1, d should be replaced with md’ + n [27,28] and then Equation (2) becomes as follows: " #  𝑛  𝑚  1 h i𝐶0 1 m −(𝑍 1 0−1n+ 0𝑑 ′ = 𝐶0 + 𝐶1 𝑑0 ′ = [1 𝑑′] [ ] C0 = ) + ( ) (3) = 𝑍 Z0 + + d = C0 + C1 d = 1 d0 𝐶1 (3) Z f b 𝑓𝑏 f b𝑓𝑏 C1 Equation (3) includes five parameters such as Z0, b, f, m, and n but depth Z can be determined Equation (3) includes such asshould Z0 , b, f,be m,estimated and n butby depth Z can be determined by two parameters, C0 andfive C1. parameters These parameters depth calibration. In the by two parameters, C and C . These parameters should be estimated by depth calibration. In the 0 several 1 sets of disparity information and the measured depth are collected. calibration experiment, calibration experiment, several sets ofrelationship disparity information and the in measured arecalibration collected. Since the inverse of depth has a linear with the disparity Equationdepth (3), the Since the inverse of depth has a linear relationship with the disparity in Equation (3), the calibration parameters C0 and C1 can be calculated in terms of pseudo-inverse form by the least square method. parameters C0 the andparameters, C1 can be calculated in terms of pseudo-inverse form bybe theeliminated least square method. By calibrating the systematic error due to Z0, b, and f can and depth By calibrating the parameters, the systematic error due to Z , b, and f can be eliminated and depth 0 accuracy can be improved [28]. accuracy can be improved [28]. To sum up, the final coordinates of X, Y, Z for the given (u,v,d’) are calculated as follows:

𝑋=

1 (𝑢 − 𝑐𝑥 )𝑍 (𝑣 − 𝑐𝑦 )𝑍 , 𝑌= , 𝑍= 𝐶 + 𝐶1 𝑑′ 𝑓 𝑓 0

(4)

Using the above equations with the calibration parameters, it is possible to reconstruct the points in 3D coordinates again from the disparity image.

Sensors 2018, 18, 771

5 of 16

To sum up, the final coordinates of X, Y, Z for the given (u,v,d’) are calculated as follows: X=

(u−c x ) Z , f

Y=

(v−cy ) Z f

,

Z=

1 C0 + C1 d0

(4)

Using the above equations with the calibration parameters, it is possible to reconstruct the points in 3D coordinates again from the disparity image. 2.2. Determination of the Ground Plane Equation During the slump flow test, the surface of the concrete is detected by a Kinect depth sensor, as shown in Figure 4. There are two coordinate systems, {C} and {S}. {C} is the camera frame where the Kinect IR camera is located and {S} is the slump frame where the concrete slump is located. In order to represent the points P(X,Y,Z) in the camera frame {C}, the expression of C P is used in this paper. Here, the leading superscript indicates the coordinate system to which the points P are referenced. For the data processing of the measured point cloud, it is convenient to use a description of the points in frame {S}, that is, S P. In order to change the description from frame {C} to {S}, the coordinate transformation can be adopted as a mathematical tool. The theory of coordinate transformation is well established in Sensors 2018, 18, x 5 of 15 the field of robot kinematics to express the position and orientation of a robot’s end effector, which is 2.2. Determination of the Ground Plane Equation serially connected by several links and joint angles [29]. As the first step, the slump frame {S} should During the slump flow test, the surface of the concrete be mathematically represented in the camera frame {C}. is detected by a Kinect depth sensor, as shown in Figure 4. There are two coordinate systems, {C} and {S}. {C} is the camera frame where the In Figure 4, theIRslump is isdefined a frame whose plane is the ground plane and Kinect camera isframe located {S} and {S} the slumpas frame where the concretexy slump is located. In order CP isCused inSthis paper. to represent the points P(X,Y,Z) in the camera frame {C}, the expression of whose Z-axis is the opposite direction of gravity. In order to transform P to P, the relative rotation Here, the leading superscript indicates the coordinate system to which the points P are referenced. matrix C translation vector S PSORG are required. In order to calculate the rotation matrix C S R and For S R, the data processing of the measured point cloud, it is convenient to use a description of the points SP. In order to change the description from frame {C} to {S}, the coordinate in frame {S}, that is, it is basically necessary to obtain the plane equation of the ground surface where the slump test can be adopted as a mathematical tool. The theory of coordinate transformation is is performed. transformation well established in the field of robot kinematics to express the position and orientation of a robot’s In this paper, the RANSAC (RANdom algorithm [30] isstep, utilized to obtain the end effector, which is serially connectedSAmple by several Consensus) links and joint angles [29]. As the first the slump frame {S} should be mathematically represented in the camera frame {C}. ground plane equation. This algorithm is an iterative method to estimate parameters of a mathematical In Figure 4, the slump frame {S} is defined as a frame whose xy plane is the ground plane and model from a set ofZ-axis observed datadirection containing outliers. it is used torelative estimate CP to SP, the whose is the opposite of gravity. In order toHere, transform rotationthe parameters 𝐶 matrix 𝑅 and translation vector SPSORG are required. to calculate the rotation 𝑆𝑅 , it plane around (a, b, c, and d) of the 𝐶𝑆following equation from a set In oforder selected data on thematrix ground is basically necessary to obtain the plane equation of the ground surface where the slump test is the slump. performed. In this paper, the RANSAC (RANdom SAmple Consensus) algorithm [30] is utilized to obtain the ground plane equation. This algorithm is an iterative method to estimate parameters of a aX + bY + cZ + d = 0 mathematical model from a set of observed data containing outliers. Here, it is used to estimate the parameters (a, b, c, and d) of the following equation from a set of selected data on the ground plane vector ofslump. the Z-axis in frame {S} is the orthonormal vector of the ground around the

(5)

The unit plane. It is defined as v in this paper and is expressed as the parameters of the ground plane equation as follows: (5) 𝑎𝑿 + 𝑏𝒀 + 𝑐𝒁 + 𝑑 = 0 The unit vector of the Z-axis in frame {S} is the orthonormal vector of the ground plane. It is

a b c ˆ as follows: k √ v defined = v x iˆas+v in vythis jˆ +paper vz kˆ and = is√expressed as the parameters iˆ + √ of the ground plane jˆ + equation 2 + b2 + c2 2 + b2 + c2 2 + b2 + c2 a a a 𝑎 𝑏 𝑐 ̂ ̂ 𝑣 = 𝑣𝑥 𝑖̂ + 𝑣𝑦 𝑗̂ + 𝑣𝑧 𝑘 =

√𝑎2 + 𝑏 2 + 𝑐 2

𝑖̂ +

√𝑎2 + 𝑏 2 + 𝑐 2

𝑗̂ +

√𝑎2 + 𝑏 2 + 𝑐 2

𝑘

(6)

Figure 4. Kinect camera frame {C} and slump frame {S}. v is the unit vector on the Z-axis of slump frame {S}.

Figure 4. Kinect camera frame {C} and slump frame {S}. v is the unit vector on the Z-axis of slump It is reasonable to use the RANSAC algorithm since it detects outlier data of Kinect and does not frame {S}. use them in fitting the plane equation. That is, unlike the least square method, it is possible to remove the effects of the extreme values resulting from erroneous measurement or environmental conditions such as light intensity or reflectivity. Figure 5a shows the depth image captured in the Kinect during the slump test. If four rectangular areas around the concrete slump are selected, the ground plane is generated by the RANSAC algorithm using the ground points at four rectangular areas, as shown in Figure 5b. The origin point of {S} can be selected as the center of four rectangular areas. The unit

(6)

Sensors 2018, 18, 771

6 of 16

It is reasonable to use the RANSAC algorithm since it detects outlier data of Kinect and does not use them in fitting the plane equation. That is, unlike the least square method, it is possible to remove the effects of the extreme values resulting from erroneous measurement or environmental conditions such as light intensity or reflectivity. Figure 5a shows the depth image captured in the Kinect during the slump test. If four rectangular areas around the concrete slump are selected, the ground plane is generated byxthe RANSAC algorithm using the ground points at four rectangular areas, as shown in Sensors 2018, 18, 6 of 15 Figure 5b. The origin point of {S} can be selected as the center of four rectangular areas. The unit vector vector the Z-axis the frame {S}, v isto used to calculate the relative orientation between the camera on theon Z-axis of the of frame {S}, v is used calculate the relative orientation between the camera frame frame {C} and the slump {S} in thesection. next section. {C} and the slump frame frame {S} in the next

(a)

(b)

Figure 5. Generation of the ground plane using the RANSAC algorithm: (a) Depth image captured by Figure 5. Generation of the ground plane using the RANSAC algorithm: (a) Depth image captured Kinect during a slump test and the selection of four rectangular areas; (b) Ground plane generated by Kinect during a slump test and the selection of four rectangular areas; (b) Ground plane generated from the RANSAC algorithm using the selected data. from the RANSAC algorithm using the selected data.

2.3. Coordinate Transformation from Camera Frame to Slump Frame 2.3. Coordinate Transformation from Camera Frame to Slump Frame In order to describe the measured points of fresh concrete in slump frame {S}, the transformation In order to describe the measured points of fresh concrete in slump frame {S}, the transformation matrix between {C} and {S} should be obtained. The relationship between {S} and {C} is characterized matrix between {C} and 𝐶{S} should be obtained. The relationshipC between {S} and {C} is characterized by the rotation matrix, C𝑆𝑅, as well as the translational vector,C PSORG. CCPSORG is the vector from the by the rotation matrix, S R, as well as the translational vector, PSORG . PSORG is the vector from the camera frame {C} to the slump frame {S}. camera frame {C} to the slump frame {S}. When an orthonormal vector to the ground plane is represented as 𝑣̂ = 𝑣𝑥 𝑖̂ + 𝑣 𝑗̂ + 𝑣 𝑘̂ in {C}, When an orthonormal vector to the ground plane is represented as vˆ = v x iˆ + 𝑦vy jˆ + 𝑧vz kˆ in {C}, it is possible to find the rotation matrix of {C} relative to {S} based on the Z-Y-X Euler angles [29]. Any it is possible to find the rotation matrix of {C} relative to {S} based on the Z-Y-X Euler angles [29]. orientation can be achieved by three rotations about the axes of a moving frame in Euler angles. Any orientation can be achieved by three rotations about the axes of a moving frame in Euler angles. Figure 6 shows the axes of {S} after consecutive Euler angle rotations are applied. When the origin of Figure 6 shows the axes of {S} after consecutive Euler angle rotations are applied. When the origin frame {S} is initially coincident with that of frame {C} in Figure 6, rotation α about 𝑍̂𝑐 ˆ causes 𝑋̂𝑐ˆ to of frame {S}̂is initially coincident witĥthat of frame {C} in Figure 6, rotation α about Zc causeŝXc to rotate into 𝑋′ , and 𝑌̂ to rotate into 𝑌′𝑠 . Here, the angle α is determined when the vector 𝑋′𝑠 is rotate into Xˆ 0s𝑠, and Yˆc𝑐to rotate into Yˆ 0s . Here, the angle α is determined when the vector Xˆ 0s is located ̂𝑐 and 𝑣̂ . The next rotation is performed about an axis located in the plane formed by 𝑍 of the ˆ The next rotation is performed about an axis of the intermediate in the plane formed by Zˆ c and v. ̂ ̂ intermediate moving frame {S’}. It is possible to find β when rotation β about 𝑌′𝑠 causes 𝑍′𝑠 ˆto rotate moving frame {S’}. It is possible to find β when rotation β about Yˆ 0s causes Zˆ 0s to rotate into Zs , that is, into 𝑍̂𝑠 , that is, 𝑣̂ (orthonormal vector to the ground plane). As a result of two consecutive rotations vˆ (orthonormal vector to the ground plane). As a result of two consecutive rotations about the axes of about the axes of the moving frame, the final orientation of {S} is given relative to {C} as: the moving frame, the final orientation of {S} is given relative to {C} as: 𝐶 𝐶 𝑆𝑅 = 𝑆𝑅𝑍 ′ 𝑌 ′ 𝑋 ′ (𝛼, 𝛽, 0) = 𝑅𝑍 (𝛼)𝑅𝑌 (𝛽)𝑅𝑋 (0) C R = C R 0 0 0 ( α, β, 0) = R ( α ) R ( β ) R (0) Z Y ZYX S   S 𝑐𝛽   X −𝑠𝛼 𝑐𝛼𝑠𝛽  𝑐𝛼𝑐𝛽 0 𝑠𝛽1 0 0  𝑐𝛼 −𝑠𝛼 0 (7) cβ0 01 sβ0 1 0 0 𝑠𝛼𝑠𝛽 cαcβ − sα cαsβ cα −sα 0 𝑐𝛼 𝑠𝛼𝑠𝛽 ] [0 1 0] =[  ] 0] [ (7)  = [𝑠𝛼 𝑐𝛼    =  sα 0cα 00  0 10 0𝑐𝛽0 0 0 1 1 0  =  sαsβ −𝑠𝛽 0 cα𝑐𝛽 sαsβ  1 −𝑠𝛽 0 0 1 −sβ 0 cβ 0 0 1 −sβ 0 cβ The angles α and β can be easily expressed in terms of the components of the orthonormal vector v in frame {C}, as illustrated in Figure 7. α represents the angle between 𝑣𝑥 𝑖̂ + 𝑣𝑦 𝑗̂ and the X-axis, and β refers to the angle between 𝑣𝑥 𝑖̂ + 𝑣𝑦 𝑗̂ + 𝑣𝑧 𝑘̂ and the Z-axis. They can be expressed as follows: 𝛼 = 𝑡𝑎𝑛−1 (𝑣𝑦 , 𝑣𝑥 ) 𝛽 = 𝑎𝑐𝑜𝑠(𝑣𝑧 )

(8)

Sensors 2018, 18, 771

7 of 16

The angles α and β can be easily expressed in terms of the components of the orthonormal vector v in frame {C}, as illustrated in Figure 7. α represents the angle between v x iˆ + vy jˆ and the X-axis, and β refers to the angle between v x iˆ + vy jˆ + vz kˆ and the Z-axis. They can be expressed as follows: α = tan−1 vy , v x β = acos(vz )


(8)

Sensors 2018, 18, x Sensors 2018, 18, x

7 of 15 7 of 15

̂ (c) ̂ (b) ˆ̂ Figure 6. Z-Y-X Euler angles: (a) rotation around 𝑍Z rotation around 𝑌′ rotation around 𝑋 Z-Y-XEuler Eulerangles: angles:(a) (a)rotation rotationaround around 𝑍 (b)rotation rotationaround around𝑌′ (c)rotation rotationaround around X ̂Yˆ𝑆𝑆0S;; ;(c) ̂ˆ𝐶𝐶C;; ;(b) Figure 6. Z-Y-X 𝑋̂S𝑆𝑆..

Figure 7. An orthonormal vector v perpendicular to the ground plane described in frame {C}. Figure Figure 7. 7. An An orthonormal orthonormal vector vector vv perpendicular perpendicular to to the the ground ground plane plane described described in in frame frame {C}. {C}.

The homogeneous homogeneous transformation transformation matrix matrix 𝐶𝐶𝑆𝑇 𝑇 that that maps maps SSP P to to CCP P is is represented represented by by the the The 𝑆 C S C The homogeneous transformation matrix T that maps P to P is represented by the orientation orientation and position information as follows [29]: S [29]: orientation and position information as follows and position information as follows [29]: 𝐶 𝐶 𝐶 𝑃𝑆𝑂𝑅𝐺 𝐶𝑆𝑅 𝑃𝑆𝑂𝑅𝐺  𝑆𝑅 𝐶  ] (9) 𝐶𝑆𝑇 = [ C C 𝑇 = [ (9) P SORG] 𝑆 S R0 0 0  0 0 0  C 11 (9)  ST =  𝐶 CPSORG The columns of the directions of the principal axes of {S} and 𝐶𝑆𝑅 are unit vectors defining 0 0 0 1 The columns of 𝑆𝑅 are unit vectors defining the directions of the principal axes of {S} and CPSORG represents the position position vector vector of of the the origin origin of of {S} {S} in in frame frame {C}. {C}. The The description description of of points points with with respect respect represents the 𝐶 to the the slump slump frame frame {S} {S} is is calculated calculated by by the the inverse inverse of of 𝐶𝑆𝑇 𝑇 as as follows: follows: to 𝑆 𝑆 𝐶 = 𝑆𝐶𝑇 𝑇 𝐶𝑃 𝑃 𝑃= 𝐶 𝐶 𝑅))𝑇𝑇 ((𝐶𝑆𝑆𝑅

𝑆 𝑆𝑃

where 𝑆𝐶𝑆𝑇 𝑇= [ where 𝐶 = [

00

00

00

−(𝐶𝐶𝑆𝑅 𝑅)𝑇𝑇 𝐶𝐶𝑃 𝑃𝑆𝑂𝑅𝐺 −( 𝑆𝑂𝑅𝐺 𝑆 ) 11

]]

(10) (10)

Sensors 2018, 18, 771

8 of 16

C The columns of C S R are unit vectors defining the directions of the principal axes of {S} and PSORG represents the position vector of the origin of {S} in frame {C}. The description of points with respect to the slump frame {S} is calculated by the inverse of C S T as follows: SP

= CS T C P
CR T − S

 ST =  where C 


CR TCP SORG S

  

0

0

0

(10)

1

Figure 8 shows the result of the transformation of points in the camera frame to the slump frame using Sensors Equation 2018, 18, x (10). 8 of 15

Figure 8. 8. Concrete Concrete slump slump described described in in frame frame {S}. {S}. Figure

2.4. Reconstruction Reconstruction of of 3D 3D Surface Surface for for Concrete Concrete Slump Slump 2.4. Triangulated irregular irregular network network (TIN) (TIN) is is the the digital digital data data structure structure generally generally used used in in geographic geographic Triangulated information systems (GIS) to represent the topographical surface. TIN is composed of contiguous, information systems (GIS) to represent the topographical surface. TIN is composed of contiguous, non-overlapping triangles. triangles. It It can can be be used used to to convert convert the the measured measured scattered scattered data data points points into into aa model model non-overlapping of the the 3D 3D surface. surface. The The surface surface of of slump slump flow flow generated generated by by TIN TIN is is shown shown in in Figure Figure 9a. 9a. Large Large triangles triangles of on the left part of Figure 9a are the shadow areas where shooting is restricted by the angle of the on the left part of Figure 9a are the shadow areas where shooting is restricted by the angle of the Kinect Kinect sensor. The vertexes are located at irregularly i) in the plane, that is, sensor. The vertexes of TIN of areTIN located at irregularly spacedspaced pointspoints (xi , yi )(xini, ythe plane, that is, TIN TIN itself does not have equal intervals in the Xand Y-axes. For ease of data analysis, as crossitself does not have equal intervals in the X- and Y-axes. For ease of data analysis, such assuch cross-section section extraction, it is desirable to construct a network composed of regular grids, as shown in Figure extraction, it is desirable to construct a network composed of regular grids, as shown in Figure 9b. 9b. Triangulation-based linear interpolation is applied to generate data at the grid points [31]. Each Triangulation-based linear interpolation is applied to generate data at the grid points [31]. Each triangle triangle plane isby defined by three points in thez equation = ax by +point c, and any point located grid at a plane is defined three points in the equation = ax + by +z c, and+ any located at a specific specific grid (x,y) within this triangle can be linearly interpolated. In this grid coordinate, it is easy to (x,y) within this triangle can be linearly interpolated. In this grid coordinate, it is easy to extract the extractheight the slump height atlocation a specific location of the ground plane. slump at a specific of the ground plane.

TIN itself does not have equal intervals in the X- and Y-axes. For ease of data analysis, such as crosssection extraction, it is desirable to construct a network composed of regular grids, as shown in Figure 9b. Triangulation-based linear interpolation is applied to generate data at the grid points [31]. Each triangle plane is defined by three points in the equation z = ax + by + c, and any point located at a Sensors 2018, 18,(x,y) 771 within this triangle can be linearly interpolated. In this grid coordinate, it is easy 9 of to 16 specific grid extract the slump height at a specific location of the ground plane.

Sensors 2018, 18, x

(a)

9 of 15

(b) Figure 9. Reconstruction of the 3D surface: (a) Surface of the concrete slump generated by the Figure 9. Reconstruction of the 3D surface: (a) Surface of the concrete slump generated by the triangulated irregular network (TIN); (b) Surface of the concrete slump in grid coordinate (Result of triangulated irregular network (TIN); (b) Surface of the concrete slump in grid coordinate (Result of triangulation-based linear interpolation for TIN). triangulation-based linear interpolation for TIN).

2.5. Cross-Section Extraction and Construction of a 4D Slump Image 2.5. Cross-Section Extraction and Construction of a 4D Slump Image Through the above-mentioned data processing algorithm, the shape information obtained from Through the above-mentioned data processing algorithm, the shape information obtained from the depth sensor camera can be expressed in a 3D surface where the height data are available at any the depth sensor camera can be expressed in a 3D surface where the height data are available at grid position of the slump frame {S}. The dynamically changing shapes of the concrete slump over any grid position of the slump frame {S}. The dynamically changing shapes of the concrete slump time are the four-dimensional spatiotemporal data, as shown in Figure 10. Here, every image of the over time are the four-dimensional spatiotemporal data, as shown in Figure 10. Here, every image dynamically changing slump flow shape is the result of applying the series of data processing of the dynamically changing slump flow shape is the result of applying the series of data processing procedures described above. procedures described above. By selectively extracting the information from each slump in Figure 10, it is possible to By selectively extracting the information from each slump in Figure 10, it is possible to reconstruct reconstruct the data and visualize the workability effectively. For example, the cross-sectional curve the data and visualize the workability effectively. For example, the cross-sectional curve (2D) of a slump (2D) of a slump can be collected over time (1D) and reconstructed into three-dimensional data, which can be collected over time (1D) and reconstructed into three-dimensional data, which again can be again can be compressed into a 2D image, as seen in the top view. In this paper, the compressed 2D compressed into a 2D image, as seen in the top view. In this paper, the compressed 2D image is called image is called a 4D slump image and it will be shown in Section 4 with an explanation of the a 4D slump image and it will be shown in Section 4 with an explanation of the experiment in Section 3. experiment in Section 3.

By selectively extracting the information from each slump in Figure 10, it is possible to reconstruct the data and visualize the workability effectively. For example, the cross-sectional curve (2D) of a slump can be collected over time (1D) and reconstructed into three-dimensional data, which again can be compressed into a 2D image, as seen in the top view. In this paper, the compressed 2D Sensors 2018,is18,called 771 a 4D slump image and it will be shown in Section 4 with an explanation of10the of 16 image experiment in Section 3.

Figure 10. Spreading shape of concrete slump flow with time. Figure 10. Spreading shape of concrete slump flow with time.

3. Experimental Setup for the 4D Slump Test 3. Experimental Setup for the 4D Slump Test 3.1. Devices for the 4D Slump Test 3.1. Devices for the 4D Slump Test As a low-cost 3D depth sensor, Microsoft Kinect was utilized to acquire data during the slump AsKinect a low-cost 3D of depth Microsoft Kinectanwas utilizedand to acquire during thethe slump test. consists an IRsensor, laser pattern projector, IR camera, an RGBdata camera. When IR test. Kinect projects consists aoflaser an IR laser pattern an IRan camera, anpattern RGB camera. When isthe projector speckle pattern projector, onto an object, image and of the on the object the IR camera and thenpattern processed to an reconstruct a 3D mapof of the the pattern object [26]. IR captured projectorby projects a laser speckle onto object, an image on The the Kinect object is provides depth video stream withprocessed a resolution of 640 × 480 apixels at a of maximum of[26]. 30 frames per captured byathe IR camera and then to reconstruct 3D map the object The Kinect second. At a lower frame rate, it can deliver 1280 × 1024 pixels. The default sensing range is 0.8 m provides a depth video stream with a resolution of 640 × 480 pixels at a maximum of 30 framesto per second. At a lower frame rate, it can deliver 1280 × 1024 pixels. The default sensing range is 0.8 m to 6 m but it is configured to near mode, which provides a range of 0.4 m to 3 m for a better quality of resolution [32]. Depth resolution refers to the minimum detectable difference in a certain continuous distance range [33]. Based on research by Smisek et al., the resolution is 0.65 mm at 0.5 m and it changes with the distance in a quadratic manner [19]. The depth resolution is less than 2 mm at 1 m and 25 mm at a 3-m distance [27]. According to the investigation of Khoshelham et al. on the Kinect, the random error of the depth measurement also increases in a quadratic manner as the distance of the object increases [27]. The random error is less than 0.6 cm at a distance of 1 m but it reaches 4 cm at the maximum range of 5 m. During the experiment, the distance was kept to less than 1 m where Kinect had a resolution of 2 mm and a random error of 6 mm. It is not recommended to increase the distance greater than 1 m because the quality of the depth data is degraded by random error and resolution. The Kinect was assembled with an in-house mounting frame, as shown in Figure 11a. The incorporation of several bolt holes on the mounting frame enabled a better shooting angle for measurement by providing additional freedom to connect it to the camera tripod. The coordinate transform algorithm allowed shooting from any angle with Kinect. Kinect drivers such as OpenNI and NiTE were installed on a Windows 7 laptop computer and an in-house MATLAB code was run for measurement and analysis.

3.2. Test Material and Procedure As a flowable concrete, an SSC mixture with a water-to-binder ratio of 39% was used in the experiment. The proportions of concrete mix are shown in Table 1. The diameter of the silica sand used in the mix ranged from 0.1 mm to 0.6 mm. The diameter of the coarse aggregate was less than 13 mm. The slump flow test was performed based on ASTM C1611 [13]. The cone was filled with fresh concrete to the struck volume in the inverted position on a flat, level, non-absorbent surface, as shown in Figure 11b. When the cone was lifted, the concrete flowed out. The slump flow test was recorded in 24 frames per second by a laptop connected to the Kinect sensor while the raw depth data were displayed on the monitor. In ASTM C1611, slump flow and T50 are measured by operators. Slump flow is the average of two perpendicular diameters across the spread of concrete and T50 is the time for the fresh concrete to reach a diameter of 500 mm after lifting the slump cone. In the conventional method, it is hard to start and stop a clock at exactly the correct times while conducting the slump flow test.

concrete to the struck volume in the inverted position on a flat, level, non-absorbent surface, as shown in Figure 11b. When the cone was lifted, the concrete flowed out. The slump flow test was recorded in 24 frames per second by a laptop connected to the Kinect sensor while the raw depth data were displayed on the monitor. In ASTM C1611, slump flow and T50 are measured by operators. Slump flow is2018, the18, average of two perpendicular diameters across the spread of concrete and T50 is the Sensors 771 11time of 16 for the fresh concrete to reach a diameter of 500 mm after lifting the slump cone. In the conventional method, it is hard to start and stop a clock at exactly the correct times while conducting the slump The flow time (T50) in the(T50) inverted coneslump orientation generally gives a higher value than flowslump test. The slump flow time in theslump inverted cone orientation generally gives a higher in the normal orientation [34,35]. value than in the normal orientation [34,35].

(a)

(b)

(c)

Figure 11. Experimental setup: (a) Kinect with camera mounting frame; (b) Slump cone in an inverted Figure 11. Experimental setup: (a) Kinect with camera mounting frame; (b) Slump cone in an inverted position; (c) Kinect and laptop computer running the 4D slump test program. position; (c) Kinect and laptop computer running the 4D slump test program. Table 1. Mixture mass proportions of test material used in the experiment. W/B

Water

C

39%

9.37%

16.77%

FA

S

G

7.18%

35.01%

31.52%

cm2 /g;

SP 0.345%

cm2 /g;

C: Ordinary Portland cement, $ = 3.16, s = 3214 FA: Flay ash, $ = 2.19, s = 3960 S: Silica Sand, $ = 2.65, diameter from 0.1 mm to 0.6 mm; G: Gravel, diameter less than 13 mm; SP: Polycarbonate Superplasticizer.

4. Experimental Results The depth images recorded at 24 Hz sampling frequency during the slump flow test were processed by the 4D slump processing algorithm described in Sections 2.1–2.4. As a result, reconstructed images of the concrete surface in regularly spaced coordinates of the slump frame at every time frame were obtained, as shown in Figure 12. It was then possible to visualize the spreading cross-section of fresh concrete during the slump flow test, as shown in Figure 13. This 3D surface is made of the extracted 2D cross-sections in a given 3D concrete surface at a specific time frame of Figure 12. By representing the height z at a specific location (x, y) of Figure 13 as color information, the 4D slump image is obtained, as presented in Figure 14. This is a compressed image of the slump flow test showing the dynamically spreading diameter of concrete slump flow. It contains the generalized slump flow time Ti, the slump diameter at any time, and the height of slump at any time and location. In Figure 14, the local area in red at the initial time frame is the result of the concrete drop from the slump cone.

Figure 12. By representing the height z at a specific location (x, y) of Figure 13 as color information, the 4D slump image is obtained, as presented in Figure 14. This is a compressed image of the slump flow test showing the dynamically spreading diameter of concrete slump flow. It contains the generalized slump flow time Ti, the slump diameter at any time, and the height of slump at any time and location. In Figure 14, the local area in red at the initial time frame is the result of the concrete Sensors 2018, 18, 771 12 of 16 drop from the slump cone.

Figure 12.12. Reconstructed time during duringthe theslump slumpflow flowtest test(top (top view). Figure Reconstructedimages imagesofofthe theconcrete concrete surface surface over over time view).

Sensors 2018, 18, x

12 of 15

Figure 13. 13. Visualization Visualization of of spreading spreading concrete concrete cross-section cross-section during during the the slump slump flow flow test test (3D (3D surface surface Figure representation). representation).

13.771 Visualization of spreading concrete cross-section during the slump flow test (3D surface SensorsFigure 2018, 18, 13 of 16 representation).

Figure 14. 14. 4D image (2D (2D color color image image representation). representation). Figure 4D slump slump image

In the 4D slump image, it is possible to detect the boundary of the slump flow, as presented in In the 4D slump image, it is possible to detect the boundary of the slump flow, as presented Figure 15. By utilizing the characteristics of the 4D slump image where the slump flow diameter in Figure 15. By utilizing the characteristics of the 4D slump image where the slump flow diameter increases from the left to right side, the boundaries of the slump flow were searched along the vertical increases from the left to right side, the boundaries of the slump flow were searched along the vertical axis from the upper and lower border lines to the middle. For an edge detection algorithm, the simple axis from the upper and lower border lines to the middle. For an edge detection algorithm, the simple threshold method or Sobel operator could be applied. In Figure 15, the red line represents the moving threshold method or Sobel operator could be applied. In Figure 15, the red line represents the moving average of the detected edge points plotted with blue color. After detecting the upper and lower average of the detected edge points plotted with blue color. After detecting the upper and lower boundaries of the slump flow in the 4D slump image, the slump flow diameter over time was boundaries of the slump flow in the 4D slump image, the slump flow diameter over time was obtained, obtained, as shown in Figure 16, by calculating the relative distance between the two boundaries in as shown in Figure 16, by calculating the relative distance between the two boundaries in Figure 15. Figure 15. This graph provides useful information for comparing the experimental behavior of This graph provides useful information for comparing the experimental behavior of concrete flow with concrete flow with simulation results through a CFD analysis. simulation results through a CFD analysis. Sensors 2018, 18, x 13 of 15

Figure 15. Detected on the the 4D 4D slump slump image. image. Figure 15. Detected boundaries boundaries of of slump slump flow flow on

Sensors 2018, 18, 771

14 of 16

Figure 15. Detected boundaries of slump flow on the 4D slump image.

Figure Figure 16. 16. Slump Slump flow flow diameter diameter over over time. time.

5. Conclusions 5. Conclusions In this paper, a novel workability test method using a 3D depth sensor was proposed and the In this paper, a novel workability test method using a 3D depth sensor was proposed and the dynamically spreading diameter of concrete slump flow was successfully visualized through a 4D dynamically spreading diameter of concrete slump flow was successfully visualized through a 4D slump processing algorithm. In order to represent the surface of fresh concrete in the slump frame, slump processing algorithm. In order to represent the surface of fresh concrete in the slump frame, we transformed the depth data in the camera frame {C} acquired by Kinect into the data in slump we transformed the depth data in the camera frame {C} acquired by Kinect into the data in slump frame {S} with the calculation of the relative orientation and position between the {C} and {S} frames. frame {S} with the calculation of the relative orientation and position between the {C} and {S} frames. The scattered data in the slump frame were then interpolated in the regularly spaced grid and the The scattered data in the slump frame were then interpolated in the regularly spaced grid and the cross-sections of the slump flow at each time frame were collected. Through the experiment, it was cross-sections of the slump flow at each time frame were collected. Through the experiment, it was confirmed that the information on the spreading slump flow diameter, slump height, and slump flow confirmed that the information on the spreading slump flow diameter, slump height, and slump flow time could be obtained by a single 4D slump image using the proposed 4D slump processing time could be obtained by a single 4D slump image using the proposed 4D slump processing algorithm. algorithm. In the conventional concrete workability test device, it is impossible to obtain these kinds In the conventional concrete workability test device, it is impossible to obtain these kinds of images of images and graphs. and graphs. The proposed 4D slump test method offers several advantages. First, instead of human The proposed 4D slump test method offers several advantages. First, instead of human measurements using a ruler and a stopwatch, which can introduce human error, it can digitize the measurements using a ruler and a stopwatch, which can introduce human error, it can digitize the workability data and quantitatively evaluate the workability at low cost. Second, it affords nonworkability data and quantitatively evaluate the workability at low cost. Second, it affords non-contact, contact, portable digital measurement and provides additional information besides the basic portable digital measurement and provides additional information besides the basic information while information while performing the existing workability test. Third, measurement of concrete flow will performing the existing workability test. Third, measurement of concrete flow will be helpful in the be helpful in the quality monitoring of concrete by providing a digital record that cannot be easily quality monitoring of concrete by providing a digital record that cannot be easily modified. In addition, modified. In addition, it will accelerate the development of flowable concrete based on numerical it will accelerate the development of flowable concrete based on numerical simulations by providing simulations by providing digital data for concrete flow experiments. Beside systematic error, random digital data for concrete flow experiments. Beside systematic error, random error, and resolution error, and resolution discussed in this paper, the depth measurement quality is influenced by the discussed in this paper, the depth measurement quality is influenced by the light condition and target light condition and target reflectivity. At a high light intensity, the IR camera cannot detect laser reflectivity. At a high light intensity, the IR camera cannot detect laser speckles on the object since the relative contrast of the speckles is decreased. That is to say, the signal-to-noise ratio of the reflected pattern is low under sunlight. If the water content of fresh concrete is high or the water in the fresh concrete tends to rise to the surface, speckle patterns projected by an IR projector are not reflected back from the concrete surface. Consequently, the Kinect will create gaps and outliers [27,36]. In terms of its usability, it can be used for a slump test, the most popular workability test, since this test only requires measurement of the static height of the concrete slump. Acknowledgments: This work is supported by the Korea Agency for Infrastructure Technology Advancement (KAIA) grant funded by the Ministry of Land, Infrastructure and Transport (Grant 18AUDP-B121595-03). Author Contributions: Jung-Hoon Kim conceived the idea of utilizing the Kinect for the concrete workability test and provided guidance in developing the software for the Kinect, designing the test, and analyzing the results. Minbeom Park designed and performed the experiment of the concrete slump flow test using the Kinect. He analyzed the data by the 4D slump processing algorithm. All authors contributed to writing the manuscript. Conflicts of Interest: The authors declare no conflict of interest.

Sensors 2018, 18, 771

15 of 16

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

11. 12. 13. 14. 15. 16. 17. 18. 19.

20.

21. 22.

23.

Koehler, E.P.; Fowler, D.W. Summary of Concrete Workability Test Methods; Research Report ICAR-105-1; International Center for Aggregates Research, University of Texas at Austin: Austin, TX, USA, 2003. Popovics, S. Fundamentals of Portland Cement Concrete: Fresh Concrete v. 1: Quantitative Approach; John Wiley & Sons Inc.: New York, NY, USA, 1982. Kwon, S.H.; Jang, K.P.; Kim, J.H.; Shah, S.P. State of the Art on Prediction of Concrete Pumping. Int. J. Concr. Struct. Mater. 2016, 10, 75–85. [CrossRef] Nehdi, M.L. Only Tall Things Cast Shadows: Opportunities, Challenges and Research Needs of SelfConsolidating Concrete in Super-Tall Buildings. Constr. Build. Mater. 2013, 48, 80–90. [CrossRef] Shin, S.W.; Yun, C.B.; Popovics, J.S.; Kim, J.H. Improved Rayleigh Wave Velocity Measurement for Nondestructive Early-Age Concrete Monitoring. Res. Nondestruct. Eval. 2007, 18, 45–68. [CrossRef] ˙ Assessment of Concrete Compressive Strength by Ba¸syigit, ˘ C.; Çomak, B.; Kılınçarslan, S¸ .; Serkan Üncü, I. Image Processing Technique. Constr. Build. Mater. 2012, 37, 526–532. [CrossRef] Rizzo, P.; Ni, X.; Nassiri, S.; Vandenbossche, J. A Solitary Wave-Based Sensor to Monitor the Setting of Fresh Concrete. Sensors 2014, 14, 12568. [CrossRef] [PubMed] Lilliman, M. Control of Mortar Rheology for 3D Concrete Printing. Ph.D. Thesis, Department of Architecture, Building and Civil Engineering, Loughborough University, Loughborough, UK, 2017. Perrot, A.; Rangeard, D.; Pierre, A. Structural Built-Up of Cement-Based Materials Used for 3D-Printing Extrusion Techniques. Mater. Struct. 2016, 49, 1213–1220. [CrossRef] Jeon, K.-H.; Park, M.-B.; Kang, M.-K.; Kim, J.-H. Development of an Automated Freeform Construction System and its Construction Materials. In Proceedings of the 30th International Symposium on Automation and Robotics in Construction, Montréal, QC, Canada, 11–15 August 2013; pp. 1359–1365. [CrossRef] Khayat, K.H. Workability, Testing, and Performance of Self-Consolidating Concrete. ACI Mater. J. 1999, 96, 346–353. [CrossRef] Bartos, P.J.M.; Sonebi, M.; Tamimi, A.K. Workability and Rheology of Fresh Concrete: Compendium of Tests—Report of RILEM TC 145-WSM; RILEM Publications: Paris, France, 2002. ASTM C1611/C1611M-14. In Standard Test Method for Slump Flow of Self-Consolidating Concrete; ASTM International: West Conshohocken, PA, USA, 2014. Kurokawa, Y.; Tanigawa, Y.; Mori, H.; Komura, R. A Study on the Slump Test and Slump-Flow Test of Fresh Concrete. Trans. Jpn. Concr. Inst. 1994, 16, 25–32. ASTM C143/C143M-15a. In Standard Test Method for Slump of Hydraulic-Cement Concrete; ASTM International: West Conshohocken, PA, USA, 2015. Koehler, E.P.; Fowler, D.W.; Ferraris, C.F.; Amziane, S. A New, Portable Rheometer for Fresh SelfConsolidating Concrete. ACI Specif. Publ. 2006, 233, 97–116. [CrossRef] Kim, J.H.; Jang, H.R.; Yim, H.J. Sensitivity and Accuracy for Rheological Simulation of Cement-Based Materials. Comput. Concr. 2015, 15, 903–919. [CrossRef] Gao, J.; Fourie, A. Spread is Better: An Investigation of the Mini-Slump Test. Miner. Eng. 2015, 71, 120–132. [CrossRef] Smisek, J.; Jancosek, M.; Pajdla, T. 3D with Kinect. In Consumer Depth Cameras for Computer Vision: Research Topics and Applications; Fossati, A., Gall, J., Grabner, H., Ren, X., Konolige, K., Eds.; Springer: London, UK, 2013; pp. 3–25. El-laithy, R.A.; Huang, J.; Yeh, M. Study on the Use of Microsoft Kinect for Robotics Applications. In Proceedings of the 2012 IEEE/ION Position, Location and Navigation Symposium, Myrtle Beach, SC, USA, 23–26 April 2012; pp. 1280–1288. Park, J.-H.; Shin, Y.-D.; Bae, J.-H.; Baeg, M.-H. Spatial Uncertainty Model for Visual Features Using a Kinect™ Sensor. Sensors 2012, 12, 8640–8662. [CrossRef] [PubMed] Tannous, H.; Istrate, D.; Benlarbi-Delai, A.; Sarrazin, J.; Gamet, D.; Ho Ba Tho, M.; Dao, T. A New Multi-Sensor Fusion Scheme to Improve the Accuracy of Knee Flexion Kinematics for Functional Rehabilitation Movements. Sensors 2016, 16, 1914. [CrossRef] [PubMed] Qi, X.; Lichti, D.; El-Badry, M.; Chow, J.; Ang, K. Vertical Dynamic Deflection Measurement in Concrete Beams with the Microsoft Kinect. Sensors 2014, 14, 3293–3307. [CrossRef] [PubMed]

Sensors 2018, 18, 771

24.

25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36.

16 of 16

Mohamed, A.; Yulu Luke, C.; Mohammad, R.J.; Sami, F.M.; Wei-Men, S.; Uvais, A.Q. 3D Dynamic Displacement-Field Measurement for Structural Health Monitoring Using Inexpensive RGB-D Based Sensor. Smart Mater. Struct. 2017, 26, 125016. [CrossRef] Khan, F.; Bolhassani, M.; Kontsos, A.; Hamid, A.; Bartoli, I. Modeling and experimental implementation of infrared thermography on concrete masonry structures. Infrared Phys. Technol. 2015, 69, 228–237. [CrossRef] Freedman, B.; Shpunt, A.; Machline, M.; Arieli, Y. Depth Mapping Using Projected Patterns. U.S. Patent 8,493,496, 23 July 2013. Khoshelham, K.; Elberink, S.O. Accuracy and Resolution of Kinect Depth Data for Indoor Mapping Applications. Sensors 2012, 12, 1437–1454. [CrossRef] [PubMed] Darwish, W.; Tang, S.; Li, W.; Chen, W. A New Calibration Method for Commercial RGB-D Sensors. Sensors 2017, 17, 1204. [CrossRef] [PubMed] Craig, J.J. Introduction to Robotics: Mechanics and Control, 3rd ed.; Pearson Prentice Hall: Upper Saddle River, NJ, USA, 2004. Fischler, M.A.; Bolles, A.R.C. Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography. Commun. ACM 1981, 24, 381–395. [CrossRef] Amidror, I. Scattered Data Interpolation Methods for Electronic Imaging Systems: A Survey. J. Electron. Imaging 2002, 11, 157–176. [CrossRef] Kinect Sensor. Available online: https://msdn.microsoft.com/en-us/library/hh438998.aspx (accessed on 10 January 2018). Yang, L.; Zhang, L.; Dong, H.; Alelaiwi, A.; Saddik, A.E. Evaluating and Improving the Depth Accuracy of Kinect for Windows v2. IEEE Sens. J. 2015, 15, 4275–4285. [CrossRef] Fares, G. Effect of Slump Cone Orientation on the Slump Flow Time (T50) and Stability of Sustainable Self-Compacting Concrete Containing Limestone Filler. Constr. Build. Mater. 2015, 77, 145–154. [CrossRef] Ramsburg, P. The SCC Test: Inverted or Upright? Concr. Prod. 2003, 21, 34–38. Azzari, G.; Goulden, L.M.; Rusu, B.R. Rapid Characterization of Vegetation Structure with a Microsoft Kinect Sensor. Sensors 2013, 13, 2384–2398. [CrossRef] [PubMed] © 2018 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/).