Research On Workpiece Matching And Localization Problem Based On RGB-D

With the development of Computer Vision and Point Set Recognition, graph matching has been already playing an important role and an important problem in Point Set Recognition field. Commonly used image matching algorithms are mainly based on local feature region matching and point pattern matching. As the basic unit of the matching, the point pattern is divided into two basic algorithms based on transformation and the feature based matching algorithm. In point pattern matching, when the image exist the geometry distortion, and be affected with the lighting intensity and occlusion, the fast matching is still a significant problem. Therefore, this paper based on the research background of robot objects capture of computer vision, and graph matching problem in three-dimension. This paper makes a deep analysis and research on the problem of the robot target matching based on the RGB-D. This has broad application prospects and important scientific significance.The main works of this paper are as follows:(1) The RGB-D data are pre-processed: the preprocessing process includes the gray processing of the image and the image filter using Gauss filter. Then use the Kinect depth information to background segment. Finally, use the Canny edge detection to extract the edge of the image. This paper uses the Framework.Net C# form application to do the image pre-process and edge detection.(2) Using the improved EMD distance to extract the point pattern histogram of the image, the shape context algorithm is calculated by the histogram of the cost function. According to the shape context in the circumstances of not reach the timeliness of histogram construction. This paper proposes the improved EMD distance to reach the timeliness.(3) And then construct the graph model, according to the matching error, this paper propose the way of using the Sequential Monte Carlo algorithm combined with shape context to achieve the continuous distribution of the graph model to achieve the purpose of selecting the best matching. The accuracy and time complexity of the PA, Sift, TBA, and ULA are compared in the MATLAB and.Net FrameWork experimental platform. Experimental results show that the algorithm has good matching accuracy and a small time complexity.(4) The image matching and grasping: in the VS2012 platform, the camera of the calibration experiment is carried out to determine the coordinates of the workpiece in the world coordinate system, use the experiment result and select the location of the camera. When the image is rotated, the position of the robot is selected by the matching results, in order to realize the robot accurate target localization and grasp process.