Fast and robust semi-local stereo matching using possibility distributions

Global stereo matching methods aim to reduce the sensibility of stereo correspondence to ambiguities caused by occlusions, poor local texture or fluctuation of illumination. However, when facing the problem of real-time stereo matching, as in robotic vision, local algorithms are known to be the best. In this paper, we propose a semi-local stereo matching algorithm (SLSM algorithm); an area-based method that embodies global matching constraints in the matching score. Our approach uses a fuzzy formularisation of the similarity assumption in order to define a matching possibility distribution. An unmatching possibility distribution is defined by applying global constraints to the matching possibility distribution. The final matching cost is computed using the two possibility distributions. Experimental results and comparison with other existing algorithms are presented to demonstrate the performance and effectiveness of our approach.