Study On Three-dimensional Reconstruction Based On Binocular Structured Light

In the field of optical 3D measurement technology,stereo vision and structured light technology are widely used.In the process of stereo matching,matching mistakes are easily occurred in uniform area and repetitive texture area.As for structured light,projector usually gets a relatively low calibration accuracy and structured light is severely influenced by ambient light.In the recent years,people proposed binocular structured light which take advantage of both stereo vision and structured light.Binocular stereo vision solved above problems but it also brings some new problems which remains to be solved.Same as conventional stereo vision,binocular structured light faces occlusion problem.From the perspective of stereo matching,occlusion means there's no matching points in the target absolute phase matrix for those points at occlusion area.Another problem is that binocular structured light faces more severe contradiction between efficiency and accuracy than stereo vison.The main reason is that binocular structured light's matching algorithms are based on absolute phase matrix,which is of DOUBLE data type rather than INT data type in conventional stereo vision.This paper focuses on occlusion problem and phase-based stereo matching algorithm's contradiction between efficiency and accuracy.In this paper,camera model?camera and projector's parameters calibration are firstly introduced.Then this paper discusses principles of binocular vision?structured light and binocular structured light.Multi-camera fusion method is used in stereo vision to solve occlusion problem.For binocular structured light,projector can be seen as a camera.Therefore this paper combine monocular structured light and binocular structured light to solve occlusion problem.This combinational way needs to project many additional encoding pictures to calibration projector's parameters compared with binocular structured light.Based on this,this paper proposed a method to calibrate projector's parameters using reconstructed points by binocular structured light.This method only need to project one group of encoding pictures to calibration projector parameters and reconstruct object surface.To prove the efficacy of prosed projector's calibration method,this paper compared projector's parameters using proposed projector calibration method and conventional projector calibration method.Then projector's parameters calculated by above two method are used to reconstruct flat calibration board's center respectively.By comparing two methods' error of center distance,proposed projector's calibration proves to be feasible.SAD and AD-Census algorithms are applied into phase-based stereo matching.Absolute phase image and intensity image's features are studied in this paper.According to the above study,absolute phase image has smaller numerical change rate in both depth continuous and discontinuous are.Besides,absolute phase image does not contain textures that are caused by brightness or color discontinuous.Based on this,this paper improves AD-Census algorithm and proposes FWA-AD-Census algorithm,which is specifically used for phase-based stereo matching.FWA-AD-Census aggregates matching cost in fixed support window rather than conventional adaptive window.The reconstruction results of leaf and backside of flat calibration board prove that proposed algorithm can reduce run time and reach similar accuracy compared with AD-Census.These proves the efficacy of proposed algorithm.