3D Point Cloud Target Segmentation Based On RGB-D

With the development of depth scanning equipment,point cloud has became one of the most important representations of three-dimensional objects.Point cloud segmentation is the process of classification of point clouds,which is widely used in robots,reverse engineering,virtual reality,human-computer interaction and other fields,It is a hot topic in recent years.At present,point cloud segmentation is mainly used as preprocessing of large-scale scene cloud,and there are many methods based on triangular mesh,and there are few methods based on point cloud model.In this paper,an independent target point cloud is obtained through the segmentation of the point cloud model obtained by modeling.Aiming at the problem of over segmentation,under segmentation,and the algorithm is not common in the present point cloud segmentation methods,this paper proposes a method to extract the target from the point cloud model using depth information and color information.The registration process of the ICP algorithm is changed and the two frame background clouds are registered respectively.The number of iterations is reduced,the error iteration is reduced,and the accuracy of the point cloud registration is improved effectively.A deep background segmentation method is proposed,and the depth value is compared with the background subtraction method.Because of the occlusion problem caused by the three-dimensional nature of 3D objects,a lot of accurate background cloud data are needed,The acquisition of point clouds is also very important.In this paper,a Kinect camera is used to rotate the 3D object,and the multi view point cloud data and the relative background point cloud data of two frames can be obtained.Image segmentation is proposed to segment the point cloud model through Grab Cut algorithm,the color image is manually selected to get the color information of the target object,then the point cloud model is traversed to get the accurate side information of the target point cloud,and then the final point cloud model is obtained by the fusion of the target point cloud with the geometric attribute of the 3D object.The paper uses VS2013+OpenCV+PCL to verify the effectiveness of the algorithm proposed in this paper.The experimental results show that the two-way bidirectional registration method changes the registration process,reduces the number of iterations of registration,reduces the iteration of transformation error,and improves the registration accuracy of point cloud.The foreground and background cloud can be distinguished effectively by using the depth information of the background cloud.Combined with image segmentation,the target point cloud is extracted from color information,and the target point cloud is fused in the limited range to get the final complete target point cloud.Using Grab Cut algorithm to select targets artificially,the algorithm is flexible and effective.