On-road Obstacle Detection based on Stereovision Analysis

Object recognition has been playing a key role in the image understanding applications. With the development of information technology, there is an increasingly demand for detecting on-road obstacles from more complicated data sources. Based on the stereo images collected from two parallel video cameras mounted on the top of a vehicle driving in the real scenes, this paper conducts obstacle detection that may be potentially dangerous for the road safety. Inspired by the V-disparity representation technique, the authors design a data classification and segmentation algorithm to separate the foreground from the complicated changing background and greatly reduce the data redundancy. Experimental results show the improvement on the overall performances.