Point Cloud Reconstruction And Point Cloud Registration Technology Based On Binocular Structured Light

With the continuous improvement of people’s living standards,newer and higher demands have been put forward for industrial intelligence.Computer vision is one of the key technologies to solve industrial intelligence.As an important branch of computer vision technology,three-dimensional measurement technology is widely used in the fields of automated production,industrial quality inspection,virtual reality and medical assistance.Structured light technology stands out from other 3D measurement techniques because of its non-destructive,non-contact,high accuracy and structural simplicity.This thesis mainly focuses on the point cloud reconstruction and point cloud registration technology of binocular structured light.The main research work is as follows:1.A set of binocular structured light system was designed independently.Specific parameters for projectors,cameras,and lenses were determined.The parameter calibration of the binocular camera is realized by using the chessboard calibration template and Zhengyou Zhang calibration method,and the reprojection error of the calibration is 0.059 pixels.2.A binocular structured light point cloud reconstruction based on the three-frequency four-step phase-shifting method is realized.The fringe frequencies of the three-frequency four-step phase-shifting method are determined.The principle of the phase-shifting method is used to solve for the phase principal,and the wrapped phase principal value is expanded by the heterodyne principle.Then,the left and right absolute phase maps are corrected by polar line.Finally,stereo matching is carried out through limit constraints to obtain the disparity map and depth map of the measured object.3.A method of binocular structured light absolute phase recovery based on depth learning is proposed.This method combines compound fringe coding and stair phase-coding,and builds a U-Net network model based on attention mechanism,which can train to get the wrapped phase and fringe order,and then get the absolute phase.Three-dimensional point cloud reconstruction of objects is realized by stereo matching of left and right absolute phases.4.Through binocular structured light system,point cloud data under different positions and orientations are obtained.And point cloud normal vector and feature degree are calculated,and feature matching point pairs are established.Then,the random sample consensus algorithm is used to screen the feature point pairs of the point cloud,and the coarse registration of the point cloud is realized based on the quaternion method.Finally,the rotation matrix and translation vector obtained from the coarse registration are used as the initial values of the iterative nearest point algorithm to complete the precision registration of the point cloud.The results show that the algorithm can be applied to the field of binocular structured light,and can achieve high registration accuracy.