A real-time multiresolutional stereo matching algorithm

Depth from stereo is one of the most active research topics in computer vision. In this work, we present an efficient depth estimation algorithm suitable for real-time implementation on commodity PCs. The proposed algorithm is based on an MML correlation-based algorithm [1]. The original MML algorithm is not reliable to produce accurate depth estimation around areas of low texture complexity and depth discontinuities. We devise an effective and efficient method for background / foreground separation. By identifying background and foreground, we successfully improve the accuracy of the MML algorithm and obtain satisfactory experimental results.