Traffic Event Automatic Detection Based on OGS-DTW Algorithm

Abstract

In order to improve the comprehensive performance of the automatic traffic incident detection, the dynamic background updating and the improved motion estimating algorithm were used to convert the dynamic image sequence into vehicle map sequence to realize the vehicle tracking. Then, the trajectory modeling and encoding methods were constructed to extract the motion trajectory of the vehicle. Self-organizing neural networks were also built up to learn the typical pattern of the motion trajectories. Finally, the OGS-DTW algorithm was used to preprocess the data of motion trajectory and compute the distance function, so that the match between trajectory of testing incident sequence and pattern of typical trajectory data can also be realized. Groups of experiments on U-turn, illegal left-turns and illegal changing-lane were carried out, the success ratio of incident detection was above 80%. The three detection methods were compared in respect of the average time-consuming and the success rate. The average time-consuming of the general DTW algorithm, the improved DTW algorithm and OGS-based improved DTW algorithm are 126.5 s, 62.5 s and 69.8 s respectively, but the success rate of the incidents testing are 84.6%, 68.8% and 88.3% respectively. The experiment results show that the traffic incident detection algorithm based on OGS-DTW is reliable and stable, which has higher matching accuracy when reducing computation. The incident detection algorithm has higher success ratio and good real-time performance.