Real-time multi-vehicle tracking based on feature detection and color probability model

As traffic surveillance technology continues to grow worldwide, computer vision-based vehicle tracking is becoming increasing important. One of the key challenges with vehicle tracking is dealing with high density traffic, where occlusion often leads to foreground splitting and merging errors. In order to help solve this problem, global features such as color or local features like corners can be used for tracking. However, tracking based on global features or local features alone does not work well with a high amount of occlusion. In this paper, we propose a real-time multi-vehicle tracking approach, which combines both local feature tracking and a global color probability model. In cases with low occlusion, corner feature detection and tracking algorithm can be used to estimate vehicle positions and trajectories. When there is a high degree of occlusion, corner features can be tracked to provide position estimates of moving objects. Then a color probability can be calculated in the occluded area to determine which object each pixel belongs to. This approach is scalable to both stationary surveillance video and moving camera video. Experimental results from a challenging transportation video clip are presented.