Detection of road obstacles using dynamic programming for remapped stereo images to a top-view

This paper proposes an artificial vision algorithm capable of extracting road vehicles in front of a stereo system as well as determining the distance between the stereo system and the detected vehicles. The algorithm transforms stereo perspective images into a virtual top-view. Using edge information of the transformed stereo images, polar accumulation functions (PAFs) are formed. Two processes called spatial and temporal peak joining are introduced to search for corresponding peaks between PAFs based on dynamic programming, which solves the typical stereo correspondence problem without a disparity computation. The proposed algorithm has been proven to be successful through a lot of experiments in various environments.