Adaptive binocular vision system for robots

The authors describe two conditions for binocular fusion, and propose a method of adaptively detecting disparity in each portion of images. Since this method automatically selects the minimum size of neighboring region in which reliable disparities can be detected, fast computation of disparities is realizable. Furthermore, they propose a more powerful method to reduce ambiguity of disparity than the neural network approach, and show that it leads to fast and reliable computation of ambiguous disparity. It is shown with several experimental results that the algorithm with these features is effective especially for robot vision