A Video-based Framework for Automated Trajectory Collection at Intersections

Abstract:

A video-based framework was presented to collect trajectories of objects at intersections. Firstly, a modified ViBe algorithm was used to extract the foreground of moving objects. Then, an OPC(ObjectPointContour) matching approach was developed for paring, tracking and generating image trajectories. Finally, real world trajectories were obtained through perspective transformation so that velocity and acceleration could be estimated. The framework was stable to occlusion and stagnation. The accuracy of 88.89%, 86.00%, and 83.33% were obtained for vehicles, bicycles and pedestrians counts and 91.71% for velocity estimation. This automatic system provides a solution for data collection and video analysis for intersection safety management and research.