Integration of a stereo matching algorithm on chip for real-time 3-D sensing

Compared to active triangulation systems, the binocular vision ones have the superiority of low cost and small size for real-time 3-D sensing. The guided-filter driven stereo matching is state-of-the-art from the viewpoints of denseness and edge-preserving. However, as most of stereo matching algorithms, the high computation and storage prevent the guided-filter driven one from being integrated on chip. In order to overcome the engineering problem, this paper analyzes how to optimize the guided-filter driven algorithm, and compares the performances among the optimized method, its original and other most important ones. Thorough experimental results validate that the optimized method is most suitable to be integrated on chip as the core of a compact binocular vision product. Besides, this paper describes the design of the algorithm on the chip system briefly.