Research on End-to-end Binocular Stereo Matching Algorithm Based on Image Segmentation
Aiming at the ill conditioned areas of low texture and repeated texture in stereo matching algorithm due to uneven il-lumination and occlusion,resulting in large amount of parameters and low accuracy,an end-to-end binocular stereo matching opti-mization algorithm based on image segmentation auxiliary guidance is proposed.The algorithm takes the image area obtained by Gaussian filter preprocessing and super-pixel segmentation algorithm as the input of stereonet network.After the process of cost cal-culation,cost aggregation,parallax calculation and parallax optimization,it is effectively adapted to the binocular stereo matching algorithm.At the same time,the parallax estimation process is optimized to realize the efficient stereo parallax estimation of the tar-get image under the characteristics of low resolution.The standard data of Middlebury test platform are used to verify.The verifica-tion results show that the algorithm in this paper can reduce memory consumption,improve running speed and improve the identifi-cation of parallax edge region in ill conditioned region.The stereo matching algorithm has certain engineering guiding significance.
国家科技期刊平台
NSTL
万方数据
