基于相似性概率约束的立体视觉稠密匹配算法

Similarity probability based stereo dense matching algorithm computer engineering and applications

本文提出一种面向非结构化环境的快速立体视觉稠密匹配算法,能有效解决弱纹理、岩石局部遮挡条件下的精确立体视觉稠密匹配问题。该算法利用外极线约束缩小搜索区域的候选匹配点,有效解决遮挡和深度不连续的影响,利用Mean-Shift算法构建一个自适应阈值函数来分割图像,再对稠密视差图进行初值优化,提高了高相似度与弱纹理区域内的视差估计精度。利用本算法对实际环境图像和标准测试图像进行处理,实验结果表明对环境中的遮挡和弱纹理问题具有良好的鲁棒性,在室内外环境下均能够获取高质量稠密视差图。

A similarity probability based stereo dense matching algorithm for unstructured environment is proposed to build high quality dense disparity map weak texture and occlusion condition. The searching area of candidate matching points is greatly reduced by epipolar constraint. A gray similarity probability based belief propagation algorithm is proposed to solve the depth discontinuous problem of occlusion. An adaptive threshold function is built and Mean-Shift segmentation is utilized to optimize initial dense disparity map in segment regions respectively. The precision of disparity estimations of weak and high similarity texture region can be effectively improved. Experimental results of standard and real images are shown that our algorithm is robust to occlusion and weak texture problem, and can obtain high quality stereo dense disparity map in both indoor and outdoor environment.