.png)
Abstract
Mountainous river beds generally consist of gravel particles that the precise description of suchbed is not only important hydraulically, but also it has great environmental significance. Theaccurate estimation of bed roughness gives us valuable information to make reliable hydraulicmodels of flow in river with rigorous bed form. This study focuses on the accuracy of the Kinectdevice in determining the digital elevation model (DEM) of the gravel-bed. In this regard, theDEMs of two beds include hemispheres and two beds with artificial gravel beds have been usedand their statistical characteristics have been analyzed. The results show that while the error inthe area among the particles is quite high, the method can accurately conduct these in general.The comparison of the bed elevation histograms shows that although the artificial gravel bedshistograms have higher accuracy compared to the histograms of the beds with hemispheres form,the gravel bed with distance elevations histogram shows the best fit among the four explored beds. Furthermore, exploring the statistical characteristics of these four beds show that theKinect device is able to obtain reasonable error rate in statistical parameters except the skewnessquantity which has the highest rate of relative error. The variogram analysis of artificial gravel beds emphasis that the Kinect and scanner variograms reasonably close to each other and the longitudinal and transversal particles length scales are exactly the same. According to the results of this investigation, application of the Kinect device in statistical analysis of the gravel beds can be suggested.
Experimental procedureThe Kinect facility is produced for the first time by Microsoft Company as a game console for Microsoft Xbox. This device could be used as a low-cost short-range high-resolution 3D/4Dcamera to simulate spatial data (X, Y, Z) in order to create a 3D model of surfaces. The Kinect includes infrared camera, IR emitter and, RGB camera and several microphone arrays to receive environmental sounds. Details of the Kinect device have been illustrated in figure (1). The IRemitter is capable to determine 3D surfaces by optical networking structure [21]. Experimental results show that the random error of depth measurement increases when the sensor become far from the object. The range of random error changes from 1 mm (at the distance of 500 mm) and 75 mm (at the distance of 5000mm) [22]. The error limitation has enough accuracy for gravel bed studies, but it can be checked through experiments in this study.
In this study, finding the abilities and accuracy of the Kinect - in generating DEM of gravel beds divided in two main steps. At the first step, in order to verify the method, regular shape objects with defined surface equation like hemisphere has been used. For appropriate method verification, two hemispheres with different size of diameters has been selected.The hemisphere shape has been used for laboratory test because of its similarity to gravel beds particles. Two different diameters and materials with hemispheres shape were built in laboratory which have been shown in figure (2). Smaller hemispheres has 28 mm diameter which is built by cement grout (figure 2-a). Greater hemispheres were created in 60 mm diameter by a 3D printer with FDM technology (figure 2-b). The color of cement grout (gray) is the same as gravels color and therefore we have color reassembly between men make hemisphere and natural gravel. To make the hemispheres surface smoother, they have been polished by a very soft sand paper and then painted by the spray. It causes the reduction in errors of Kinect’s 3D model.
Figure 2.Beds which consist of hemisphere shaped objects: a) Small cement hemispheres, b)large plastic hemispheres
The 3D surface of beds (4 hemisphere) has been detected by Kinect and Skanect software. A scaled paper has been drawn in Auto CAD to set as a background. It helps us to scale the data from Kinect spatial outputs to Cartesian DEM coordinates.In second step, random size natural gravels (which were collected from a natural river bed)have been glued on a plate to simulate the river bed artificially in laboratory. The plate made of wood with 400×400 mm length. Simulated river beds have two major types which is depicted in figure (3). Simulated river bed includes 1) collection of grains with no space and 2) collection of grains with considerable space.
Figure 3. Artificial simulated gravel beds: a) Simulated gravel bed - collection of grains with nospace, b) Simulated gravel bed collection of grains with considerable space.
The simulated river bed has a great degree of irregularity in space. This makes the verifying process of the real bed with generated bed by Kinect so hard. For verifying process, the MATLAB code has been developed to calculate the special error between real bed height and generated bed by Kinect. In case of simulated gravel beds, high resolution DEM is necessary togenerate high density point clouds with the minor errors as a pattern to compare- with DEMs generated by Kinect. 3D scanners can help us to catch the real bed DEM as an accurate and reliable device. In this study, a Solutionix C500 (new generation of 3D scanner) was used wherethis facility is shown in figure (4). The Solutionix C500 is an industrial 3D scanner, which released in 2017 by MEDIT Metrology Engineering Group for the first time, optimized for theautomatic scanning of small to medium objects. Dual 5.0MP cameras -prepares high resolutionfacilitate to capture 3D components, where the accuracy is limited down to a few microns (50micron). The device covers four visibilities with 9, 175, 350 and 500 mm length for differentresolution which is shown in table (1). The C500 is portable and can be placed over an office desk and have very simple usage. Scanning easily and rapidly makes it more user-friendly, and also automatic turntable with 10 Kg weight capacity, convince us to use Solutionix C500 in thisstudy. This scanner generates DEM in STL format using ez Scan 2017 software [23].The Kinect equips Infrared technology, and this system make it inappropriate for utilizing in daylight. Interfering sunlight and Kinect infrared ray causes considerable noises in results infield surveys on day. Because of this, Kinect can be applied in laboratory under controlled condition, to obtain accurate results. The experiment starts by setting up the Kinect - and link itto the Skanect software. We can circulate the kinect around the objects in different angle and simultaneously the point clouds are generating by Skanect software in PLY formant. At the endof data acquisition, the raw data need to be edited or corrected where MeshLab software has been utilized.
