A stereo matching algorithm based on top-view transformation and dynamic programming for road-vehicle detection

摘要

This paper presents an algorithm to detect vehicles on roads and to measure inter-vehicle distance. Hypothesis generation and verification are the bases of the algorithm. The hypothesis generation is composed of 1) transformation of a pair of perspective stereo images into a pair of virtual stereo top-view images; 2) construction of polar accumulation functions (PAFs) for the stereo top-view images; and 3) stereo matching of PAFs by dynamic programming. The verification is comprised of 1) temporal matching of PAFs by dynamic programming and 2) verifying the potential positions of objects. Through multiple experiments in various environments, the proposed algorithm has been proven to be effective in detecting vehicles on roads and for measuring inter-vehicle distance.