Stereo Matching Algorithm Using Population-Based Incremental Learning on GPU

Abstract

To solve the general problems of genetic algorithms applied in stereo matching, two measures are proposed. Firstly, the strategy of the simplified population-based incremental learning (PBIL) is adopted to decrease the problems in memory consumption and searching inefficiency, as well as a scheme controlling the distance of neighbors for disparity smoothness is inserted to obtain a wide-area consistency of disparities. In addition, an alternative version of the proposed algorithm without using a probability vector is also presented for simpler set-ups. Secondly, to decrease the running time further, a model of the proposed algorithm which can be run on programmable graphics-hardware (GPU) is newly given. The algorithms are implemented on the CPU as well as the GPU and evaluated by experiments. The experimental results show the proposed algorithm has better performance than traditional BMA methods with a deliberate relaxation and its modified version in both running speed and stability. The comparison in computation times for the algorithm both on GPU and CPU shows that the former has more speed-up than the latter, the bigger the image size is.