Three Dimensional Perception Method Based On Binocular Structured Light

Three-dimensional perception is an important research content in the field of computer vision.At present,there are various ways to obtain three-dimensional information.In the field of vision,the three-dimensional information of scene can be obtained by binocular stereo vision.However,the traditional binocular stereo vision method has a large amount of computation,which relies heavily on image matching,and has a low matching accuracy for images without obvious features.In this thesis,a three-dimensional perception method based on binocular structured light is proposed.The method combines the three-dimensional measurement technology based on structured light with the binocular imaging technology,solves the matching problem caused by the difficulty of feature extraction,and improves the measurement accuracy.Compared with traditional structured light imaging using monocular camera,this method uses binocular camera stereo to solve the difficult problem of projector calibration,reduces the complexity of the system,and improves the operability and flexibility of the system.The specific contents of this thesis are as follows:Firstly,based on the construction of binocular structured light imaging hardware system,a calibration algorithm of binocular structured light imaging system based on plane calibration is studied.The calibration algorithm can not only acquire the internal parameters of left and right cameras in binocular structured light system,but also calculate whether the external parameters can reflect the relationship between left and right cameras.On this basis,the Bouguet polar line correction method is used to correct the polar line of the binocular structured light camera,so that the optical axis of the binocular camera is parallel,so as to ensure that the image points are in the same height between the left and right cameras.Then,a Census-PatchMatch stereo matching algorithm for binocular structured light is proposed.In the matching cost function of the algorithm,the illumination intensity is used instead of the chromatic aberration factor in PatchMatch,and Census transform function is used as the weight function.At the same time,an optimization method based on sparse normalization is proposed in the iteration process.In the iteration,only the pixel data of the interlaced rows and columns are extracted for analysis.After extracting the corners of the image,the number of corners of the left and right images is not necessarily equal.Next,the normalization method is used to optimize the corners of the image,so that the corners of the left and right images are optimized.Point-to-point correspondence improves the speed of the algorithm.After the depth map is obtained by Census-PatchMatch matching algorithm,the experimental results of binocular stereo vision system and binocular structured light system based on PatchMatch matching algorithm are compared.The accuracy of the proposed algorithm is higher than that of the above two comparison algorithms.Finally,the disparity map is acquired by binocular structured light stereo matching algorithm based on ensus-PatchMatch.At the same time,the disparity map is transformed into depth map by triangulation method,and the depth information and color information are fused to form a three-dimensional point cloud image with color information.Then the experimental results of three-dimensional point cloud with color information are given.