Study On Vision-based Road Detection

The key for road detection based on vision is how to accurately classified pixels point for road surface and non-road surface, it is a big challenge in different light, shadow and road with heavy traffic flow and complex road conditions. Monocular vision has the characteristics of low cost, difficult to handle, road detection based on monocular vision has become a hot spot of research. Therefore, this paper combined with the practical application of unmanned vehicle to study how to use monocular vision to detect the road complex condition, put forward to have the accuracy and robustness of the road detection algorithm. The main work content and research results are as follows:In order to improve the processing speed of road detection, reduce the interference of irrelevant information, this paper presents a method of acquiring an interest region based on driving speed sense. According to the running speed of the unmanned vehicle, to identify the interest region of image which taken from car camera. The method is simple, real-time effective.In order to improve the robustness of road detection algorithm to the shadow, this paper uses the physical space based on illumination invariant. By use of this feature can affect shadow effective less in the original image, even with the simplest model of road. The light change direction orthogonal projection greatly reduces the influence of shadow, and can achieve realtime in the single sensor cameras, so that it can meet the real-time vision road detection unmanned vehicle requirements. In this paper, the entropy for light independent angle based on multiple images, by removing the illumination invariant angular deviation big, on this basis we mean value is independent of illumination angle, which increases the robustness of the algorithm.In order to improve the accuracy of road detection algorithm, presents a growing algorithm based on seeded region. In the paper by detecting a frame of image based on out of the road as input, and the combination of road edge detection as road seeds and roughly determine the road edge. Finally based on the probability density function and the mean to determine the seed point whether the road, so it can effectively improve the speed of the algorithm and the robustness of the algorithm. Also there is better accuracy for pixel road edge segmentation.Finally, programming in the interested region of the illumination invariant graph algorithm and modified seeded region growing segmentation algorithm based on the road, the experimental results show that, in a complex environment, it can detect the road area effectively, compared with the detection method of the new roads, in running speed as that significantly improve the detection efficiency.