Towards real-time stereo employing parallel algorithms for edge-based and dense stereo matching

Few problems in computer vision have been investigated more vigorously than stereo. Nevertheless, the main obstacle on the way to their practical application is the excessively long computation time needed to match stereo images. This paper presents parallel algorithms for edge-based stereo that are suitable for depth computation. Edge-based stereo techniques produce only sparse depth maps; thus we present, in addition, an efficient parallel algorithm for dense stereo matching that can be employed in scene reconstruction. Both approaches are implemented on several different computers to measure their performance. We compared single-processor and multiple-processor implementations to evaluate the profit of parallel realizations. We show that both approaches are very suitable for parallel implementations and that the computing time can be considerably reduced with parallel implementations. Furthermore, we present the results that are obtained when employing the different approaches to stereo images.