Research On Structure Light Depth Perception Algorithm Based On Phase-Shifting

The depth perception based on phase-shifting method means that using time-series phase encoding to obtain high-precision and high-resolution depth information.It has a wide range of applications in industrial manufacturing,cultural relics protection.biomedical,media arts and national defense.In phase-shifting depth perception,perceptual accuracy is limited by the speed,accuracy,and robustness of phase unwrapping.The purpose of phase unwrapping is to eliminate the periodic ambiguity of the phase and achieve a consistent phase distribution across the image.Due to the complex illumination of the scene,the color interference of the target surface and other factors,there are still many technical difficulties in high-precision phase unwrapping.Based on the analysis and summary of relevant research results at domestic and foreign.the existing methods are comprehensively evaluated,and the phase-shifting depth perception method is deeply explored in this paper.The main research contents and innovations of this paper are reflected in the following aspects:In order to solve the problem of gamma error caused by the nonlinear response of the projector in the structured light depth perception system,a phase-shifting projection pattern based on the defocusing effect and three auxiliary phase unwrapping projection patterns are designed.In terms of the binary phase projection pattern,in order to obtain the phase-shifting projection pattern of the defocusing effect,a binary phase-shifting projection pattern optimization method is designed based on error diffusion and gray phase correction.In terms of the auxiliary phase unwrapping projection pattern,in order to improve the accuracy and robustness of the phase unwrapping,the CGC projection pattern is designed based on the Gray code.In order to further reduce the number of timing codes and improve the speed of phase unwrapping,the GSC projection pattern is designed based on gray-scale step coding.In order to obtain the exact code value of the spatially discrete coding of the defocusing effect,a random grid projection pattern is designed based on the random speckle and line pattern.In order to solve the problem of low robustness and poor anti-interference ability of phase-shifting depth perception during phase unwrapping,two phase unwrapping methods based on time series coding pattern are proposed.In order to improve the accuracy and robustness of phase unwrapping in complex scenes,a phase unwrapping algorithm based on CGC timing coding pattern is proposed,which can realize high-precision phase unwrapping in complex scenes.In order to further improve the fastness of phase unwrapping,a phase unwrapping algorithm based on GSC timing coding pattern is proposed.This method improves the phase unwrapping accuracy and improves the fastness of phase unwrapping.In order to solve the problem that the phase-shifting depth perception has more auxiliary information and larger boundary error in the phase unwrapping,a phase unwrapping algorithm based on sparse depth is proposed.The method applied the sparse depth perceived by the random grid as the auxiliary information,based on the actual relative phase boundary,used the sparse depth to correct the phase ordinal,and has superior performance in boundary offset,which can achieve fast and high-precision phase unwrapping and improve the efficiency of phase unwrapping.In order to obtain the accurate sparse depth of discrete space coding of defocusing effect quickly,the topological feature matching result of speckle and line pattern in random mesh is taken as the true value,a fast sparse depth sensing algorithm based on machine learning is proposed,which improves the efficiency of the generation of auxiliary information.In order to solve the problem that auxiliary information in phase-shifting depth perception cumbersome,an RP-Net phase unwrapping network based on R-CNN and LSTM is proposed.The segmented irregular phase region block is used as the RP-Net input,and the high-dimensional features of the relative phase region block are obtained by R-CNN,and the correlation feature between the high-dimensional vector sequences is extracted by LSTM.The Softmax classifier is used to classify the high-dimensional feature vectors of the region block to obtain accurate phase ordinal information.This method transforms the phase unwrapping problem into a phase region classification problem.It can realize unassisted accurate phase unwrapping,which has certain advantages compared with the traditional phase unwrapping method.