A Comparative Study of Stereovision Algorithms

Stereo vision has been and continues to be one of the most researched domains of computer vision, having many applications, among them, allowing the depth extraction of a scene. This paper provides a comparative study of stereo vision and matching algorithms, used to solve the correspondence problem. The study of matching algorithms was followed by experiments on the Middlebury benchmarks. The tests focused on a comparison of 6 stereovision methods. In order to assess the performance, RMS and some statistics related were computed. In order to emphasize the advantages of each stereo algorithm considered, two-frame methods have been employed, both local and global. The experiments conducted have shown that the best results are obtained by Graph Cuts. Unfortunately, this has a higher computational cost. If high quality is not an issue in applications, local methods provide reasonable results within a much lower time-frame and offer the possibility of parallel implementations.