Robust visual surveillance based traffic information analysis and forewarning in urban dynamic scenes

Forewarning to avoid potential traffic accidents is of great importance for Intelligent Transportation Systems (ITS). Under pedestrian and vehicle mixed traffic conditions like urban road intersections, traffic monitoring and forewarning have especially important values. Therefore in this paper a novel urban traffic information analysis and forewarning system is presented. Our system contains modules including object detection based on background subtraction; object tracking based on Multiple Hypotheses Tracking; and object status judgment based on forewarning logic for abnormality detection. Different from other approaches, we improve object tracking by fusing object's position, size, velocity and its multi-part color histogram for data association. Through fusion we can better handle foreground object missing, merging and splitting problems during the tracking process. To enhance the practicality of our system, forewarning logic is designed according to different use cases for traffic abnormality detection, which is defined based on our extensive study on traffic status monitoring. Experiments with short and long video sequences show robust and accurate results of abnormality detection and forewarning under conditions of varying view angles, zoom depths, backgrounds, and frame rates. All the experimental results run at real-time frame rates (≥ 25 fps) on standard hardware, which is suitable for actual ITS applications.