Binocular Stereo Matching Algorithm Based On Adaptive Weighted AD-Dualcensus

As an important branch of computer vision,binocular stereo vision is widely used in biology,medicine,artificial intelligence,ecology,traditional industry and other fields.The regional binocular stereo matching system obtains the depth of the corresponding point by mapping the two lenses of the binocular camera to the same pixel points in the pictures taken in the same scene,thereby restoring the three-dimensional information of the scene from the two-dimensional image.It is an important research content in the field of computer vision research.Cost aggregation and cost calculation are the basis of the regional binocular stereo matching system,which is the research focus of this paper.The aggregation window is the key to cost aggregation.Adaptive cross braced windows have replaced fixed windows and are widely used for cost aggregation.However,the cross-support cost aggregation window generation algorithm uses constant values as the arm length judgment threshold,which is not stable enough for the generation of center points with different characteristics in the image.This paper proposes a nonlinear cross-threshold cross-over cost aggregation window generation algorithm.A non-linear variable threshold that conforms to the characteristics of the image area is used to replace the fixed threshold in the cross-support cost aggregation window generation process,thereby making the cross-support cost aggregation window generation process more reliable and improving the effect of the cost aggregation process.Cost calculation is a link required for both local stereo matching and global stereo matching,and is the basis of both,which directly affects the accuracy of stereo matching.In this paper,an adaptive AD-dual sequence Census cost calculation algorithm is proposed.The minimum arm length of the cross support of the pixel is used to adaptively set the composite weight of the AD cost and the dual sequence Census cost,so that the AD of pixels in different regions of the image The transformation cost and the Census transformation cost have different effects on the matching cost calculation.The synthetic matching cost can reflect the regional characteristics of the image area where the pixels are located,and improve the matching accuracy.Finally,in this paper,based on a nonlinear variable threshold cost aggregation neighborhood generation method and an adaptive AD-dual sequence Census cost calculation algorithm,a regional binocular stereo matching system is implemented.The evaluation results of the third generation binocular stereo matching evaluation platform in Middlebury show that the depth error of the unoccluded pixels of the stereo matching system designed in this paper is 6.18%,and the depth error of all pixels is 8.98%,which is practical.