Feature-based stereo matching via coevolution of epipolar subproblems

This work proposes a new parallel method to calculate accurate disparity maps of edge features in stereo imagery, based on Symbiotic Genetic Algorithms (SGAs). The search is not limited by the Ordering Constraint. For each epipolar line a single GA is allocated, configured to search for the optimum intra-row matching. Every GA evolves in parallel with the others and the credit assignment for its members takes into account the structural compatibility with the currently best solutions of the neighbouring GAs. In this way, the symbiotic optimization enforces figural continuity at the global inter-row level without the need for post-processing. For the cost of matching two edgels we propose a match functional based on the distance transform of the edge maps as well as the similarity of the local topographic profile in the vicinity of the features under consideration.