Robust 3D vision for robots using dynamic programming

In this paper a new stereo vision method is presented that combines the use of a lightness-invariant pixel dissimilarity measure within a dynamic programming depth estimation framework. This method uses concepts such as the proper projection of the HSL colorspace for lightness tolerance, as well as the Gestalt-based adaptive support weight aggregation and a dynamic programming optimization scheme. The robust behavior of this method is suitable for the working environments of outdoor robots, where non ideal lighting conditions often occur. Such problematic conditions heavily affect the efficiency of robot vision algorithms in exploration, military and security applications. The proposed algorithm is presented and applied to standard image sets.