Stereo matching using hierarchical features for robotic applications

In this paper, an efficient matching method for solving hierarchically the correspondence problem in stereo vision for robotic applications is proposed. Hierarchical features such as line segments, corners, connectivities and regions are used for matching process, and the matching results at the higher level are used to guide and constrain the matching at the lower level. The proposed hierarchical approach makes it possible to maintain a globally consistent matching of features and thus to avoid local mismatches. Experimental results show that the proposed hierarchical algorithm can be effectively used for stereo matching.