Local Stereo Matching with Adaptive and Rapid Cost Aggregation

Abstract

This paper presents a new local binocular stereo matching algorithm. A local window with adaptive support weights is used to aggregate matching costs. A more robust matching score is set up combining sum of absolute intensity differences and a gradient based measure. Each pixel has a different disparity search range according to disparity gradient analysis and computational complexity decreases rapidly. Experimental results using the Middlebury stereo test bed demonstrate the validity of our presented approach.