基于Census变换的双目立体匹配算法

摘要

Improved feature matching method based on census transform was proposed to solve the problem of overlapping occlusion, repeated texture, low texture and projective scaling.The feature points were extracted using Harris corner detection algorithm and descriptors of feature points were generated by Census transform.The Hamming distance was used to judge the similarity of the descriptor.The left and right consistency was used to complete the rough matching, overlapping occlusion points and low textures were eliminated to provide more accurate location information.The location information of the matching points in the coarse matching set was selected to complete the matching.Experimental results show that the proposed method has good matching effects for the high repetitive texture checkerboard images in binocular camera calibration, and it can be used in realtime system.%针对双目视觉系统特征匹配过程中出现的重叠遮挡、重复纹理、低纹理、投影缩放等问题, 提出基于Census变换改进的特征匹配方法.利用Harris角点检测算法提取特征点, 使用Census变换生成特征点的描述子;用汉明距离对描述子进行相似度判断, 利用左右一致性进行双向匹配完成粗匹配, 消除其重叠遮挡点及低纹理现象, 为再匹配提供更加准确的位置信息;利用粗匹配集合中匹配点的位置信息进行筛选, 完成再匹配, 得到匹配的特征点对.实验结果表明, 该方法对于双目摄像机标定中常用的高重复纹理的棋盘图具有很好的匹配效果, 算法运算时间短, 可用于实时性系统中.