Evaluation of the Approach for the Identification of Trajectory Anomalies on CCTV Video from Road Intersections

Abstract

The approach for the detection of vehicle trajectory abnormalities on CCTV video from road intersections was proposed and evaluated. We mainly focused on the trajectory analysis method rather than objects detection and tracking. Two basic challenges have been overcome in the suggested approach—spatial perspective on the image and performance. We used trajectory approximation by polynomials as well as the Ramer-Douglas-Peucker N thinning technique to increase the performance of the trajectory comparison method. Special modification of trajectory similarity metric LCSS was suggested to consider the spatial perspective. We used clustering to discover two types of classes—with normal and abnormal trajectories. The framework, which implements the suggested approach, was developed. A series of experiments were carried out for testing the approach and defining recommendations for using different techniques in the scope of it.