Target vehicle detection and tracking in surveillance videos based on exponential forgetting and perceptual hash algorithm

In order to obtain the stable background of a traffic surveillance video, the application scenarios, computational complexity, and results of the Gaussian background model were analyzed. In the background modeling process, many artifacts were generated, making it difficult to reconstruct a highly stable background image. In order to solve this problem, we proposed a background model based on exponential forgetting algorithm. This method satisfies the background reconstruction of complex environment, the computation is small and the background result is pure. To further improve the accuracy of the test, shadows of the vehicle were detected to reduce the error caused on vehicle detection. The perceptual hash algorithm was used to track the target vehicle. The proposed algorithm is verified by different video data. The experimental results show that a good vehicle tracking effect is achieved in a variety of weather conditions.