An Efficient Video-Based Vehicle Trajectory Processing Approach

Vehicle trajectories provide valuable information of traffic dynamics. The paper proposed a modified video-based trajectory data processing approach to reliably and efficiently collect trajectory data on congested urban roads or at intersections. The approach is on the basis of Knoop's idea of video transformation so that traces of vehicles are visible in a single image. Different from the existing method, a mode-based background subtraction algorithm is applied in the RGB color space to alleviate the influence of a small gap. Vehicle trajectories are then directly detected by searching connected components in the segmented binary vehicle trace image. Test results show that the complete process procedure of a single lane can be finished in 1 hour for a 5-min video. Almost 97% of the trajectories can be accurately detected which is very promising.