Point Cloud Target Recognition And Pose Estimation Based On Binocular Structured Light Scanning

With the continuous development of computer technology,the traditional 2D plane measurement technology is gradually replaced by 3D measurement technology.Compared with the measured object information in the plane image,the point cloud data of the measured object obtained by 3D measurement not only includes the object depth information,but also more intuitively obtains the attitude and position information of the measured object in space.Therefore,this paper studies the technology of point cloud target recognition and pose estimation based on binocular structured light 3D scanning.Firstly,the design principle of binocular structured light 3D scanning system is analyzed,mainly including camera imaging model and distortion principle.By establishing the mathematical model of the system,this paper expounds the transformation relationship between the four coordinate systems,and deduces the 3D coordinates of a point on the object in the binocular stereo vision system by using the binocular calibration results.In addition,the method of epipolar correction is proposed,which solves the problem that the imaging planes of the left and right cameras are not parallel,makes the corresponding pixels on the left and right images in the same line,and improves the accuracy of stereo matching.Secondly,the methods of light stripe image processing and stripe center extraction are studied.The basic principles of several commonly used stripe image filtering algorithms are analyzed.Aiming at the problems of uneven and reflective interference of light fringe image,an adaptive bilateral filtering algorithm and adaptive threshold binarization method are proposed.A variety of traditional filtering algorithms and fixed threshold binarization methods are compared to verify the effectiveness of the proposed method.The gray center of gravity method is selected as the preliminary stripe center extraction algorithm by comparing various fringe center extraction algorithms,and the sub-pixel extraction of light fringe center is completed by the algorithm of Steger.Thirdly,in view of the problems of long processing time and low registration accuracy caused by more noise points and large point cloud data,statistical filtering and voxel grid method are used to denoise and downsampling the point cloud.At the same time,the algorithm of RANSAC is used to preliminarily segment the field point cloud,extract the point cloud data after removing the plane background.Then,the European clustering algorithm based on KD-Tree nearest neighbor search identifies the target object point cloud and segments it.Finally,based on the identified target point cloud,the pose estimation of the target point cloud is realized by combining SAC-IA algorithm and ICP algorithm.Finally,the construction of binocular structured light 3D scanning system is completed,the calibration accuracy of the system is experimentally analyzed,and the point cloud target recognition and pose estimation of workpieces of different sizes are carried out.The experimental results show that the 3D scanning technology based on binocular structured light and the point cloud target recognition and pose estimation method proposed in this subject can not only accurately complete the 3D reconstruction,target recognition and pose estimation of the target object thermocouple,but also have applicability in the processing of heterosexual workpieces.