The Plane Constraint Method And Fast Algorithm Implementation For Stereo Vision Depth Computation

Stereo vision depth calculation is one of the important research directions of computer vision,which first matches the image features in calibrated and rectified binocular camera image pairs and then estimates the depth between the object and the imaging equipment in the three-dimensional scene by using the triangulation principle.As a basic module of computer vision,stereo vision depth calculation has been widely applied to three dimensional scene reconstruction,autonomous navigation of mobile robots,medical imaging,industrial inspection and many other fields.The main work of this paper is as follows:1)This paper first analyzes the limitations of several traditional disparity estimation methods in weak-texture-region disparity estimation,that are based on point correspondence matching.The comparison is conducted following two lines:matching cost(such as region-based local cost calculation method,Sobel operator,and Census transform)and cost aggregation(including region-based local cost aggregation algorithm,minimum spanning tree based non-local cost aggregation algorithm and the Semiglobal Matching algorithm).Aiming at the problem of disparity estimation in weak texture regions,this paper proposes a triangulization depth estimation algorithm based on plane constraints.The algorithm uses Sobel filter operator to detect support points,and calculates and compares the disparity value between support point pairs.At the same time,the reference image is segmented twice.The segmentation result is taken as the minimum calculation unit,and the pixel-by-pixel calculation is replaced by the the plane calculation.Using the four sets of test images on the Middlebury standard dataset,the disparity estimation algorithm is analyzed experimentally.The results show that:compared with the LIBELAS(Library for Efficient LArge-scale Stereo)algorithm,the proposed algorithm reduces the error rate of disparity estimation in weak texture region by 1.95%on average;and that in terms of algorithm speed,the proposed algorithm reduces the average time consumption by 6.923s compared with the classical SGM(Semiglobal Matching)algorithm,and 3.39s compared with the global algorithm based on minimum spanning tree.2)In this paper,the idea of plane constraint is applied to the post-processing stage.After the left and right consistency check,the pixels of the disparity value to be filled are assigned to the nearest plane to which it belongs.Then,by computing the plane equation of the plane in which it is located,the disparity value of the pixel can be reasonably induced.Due to the limitation of plane constraint,a more accurate dense disparity map can be obtained finally.Compared with other post-processing methods,the proposed method reduces the error rate of disparity estimation by 0.64%.