Abstract: Abnormal event detection in a traffic scenario is significant for accident prevention, on-call solution and other applications.While currently, the detection approaches are neither labor efficient, nor real-timing.A multi-target tracking algorithm combining motion characteristics and apparent characteristic for vehicle trace in highway traffic scenario is derived by introducing target detection in deep learning and extracting targets in a video.An abnormal event detection method which is based on vehicle trajectory feature is proposed.The proposed tracking algorithm declines the dependent on background during the procedure of trajectory extraction.The abnormal event detection algorithm is designated adequately for highway scenario with an additional strategy of sliding window to improve the performance on abnormal detection for remote and complex scenes.Through the experiments on practical video repository, the proposed method is compared with existing methods and shows a highlighted performance in the terms of Precision, Recall rate and F-value index.This method turns out to be solid and efficient in abnormal event detection in highway traffic scenario.

Key words: Abnormal event detection, Target detection, Target tracking, Trajectory feature, Video surveillance