Traversable Region Detection Based On Vision For Intelligent Vehicle

Traversable region detection is key one of intelligent vehicle vision technologies,which is widely used in intelligent transportation system,driver assistance system and other fields.Based on a single image,this paper researches the traversable region detection method of unstructured road for intelligent vehicle from three aspects:vanishing point,obstacle and road segmentation.Firstly,in order to resolve the problem that the vanishing point detection of curved road is not accurate,this paper proposes a novel vanishing point detection method based on weighted voting.The vertical projection rectification method is used to rectify the dominant Gabor texture orientation of a voter in advance,and then a voter increases weights of other voters above it on the same edge line,weights is used to vote for the vanishing point.The vanishing point detection experiments show that the proposed method has the best detection effect and can meet the real-time requirements of the system.Secondly,based on the fact that trees,railings and other obstacles all have a certain vertical structure,this paper proposes a novel vertical obstacle detection method based on Haar-like features and dominant texture orientation.The method firstly uses linear rectangle features to distinguish obstacles and background,and then determines whether the rectangle contains obstacles through the probability of vertical texture in rectangular region.The later obstacle detection experiments show that the method can detect vertical obstacles accurately,and voters in the rectangular obstacles can be used in reverse vote to eliminate the interference of vanishing point detection from obstacles.Finally,in order to improve the accuracy of curved road segmentation,a novel partitioned road segmentation method based on vanishing points is proposed.Firstly,the image region below the vanishing point is divided into two parts according to the height ratio,and the boundary is calculated one by one based on the real road shape in the image.Road segmentation experiments show that this method meets the real-time requirements of the system and improves the accuracy of road segmentation.