CMV-based design of motion-vector selector

This paper proposes a novel design of motion-vector selector for vision-based robot navigation. In an imaging plane, five angled observing windows are defined for obtaining candidate motion-vectors. Subsequently, those motion vectors with low reliability are detected, and only reliable ones are selected to compute motion parameters of vehicles by using the reliability of the candidate motion vectors base on complementary voting algorithm (CMV). Experimental results with underwater image sequences demonstrated the effectiveness of the selector.