Color vision-based multi-level analysis and fusion for road area detection

This paper presents an integrated approach to robust analysis of road area images in front of the car from a single color camera. In order to get more information from the image source, we build a three-level data fusion based on Dempster-Shafer’s decision theory. During the first level, we separate the input image into perspective view and bird’s view and recognize small road patches from the background using color histogram features. In the second level we fuse the evidences from each recognized patch and re-calculate the road-like probabilities in each view. The third level is to merge the results of the two views into one bird’s-view segmentation map to achieve a high degree of reliability. Our Bagging-like method has been tested in real world environments and the fusion results have been shown to be robust to road textures, trees and moving objects in the scene.