Lane segmentation in traffic monitoring systems based on probability map

This paper describes an approach for lane segmentation in traffic monitoring systems based on probability map extracted form vehicles movement information. In traffic monitoring system, the region that a vehicle passes through might belong to a certain lane. The more vehicles pass through a certain region, the more likely it belongs to a certain lane, and then we can get probability map of the lane. The regions belong to the same lane must have the same probability. So, we can segment the probability map to extract the segmentation of the lanes. Firstly, we use Gaussian background modeling algorithm to extract the motion regions and background image, then the probability map is built by moving vehicles. The road areas acquired by region growing algorithm based on moving vehicles are used to eliminate the disturbance outside of road areas. The straight-lines about lane boundaries can be extracted by using Hough transform on the edge image of probability map. Finally we divide a road area into several lane regions by fusing the trajectories information of vehicles acquired by KLT method. The experiments in deferent scenes show the efficiency and robustness of the proposed approach.