The Research On Moving Target Tracking Technology In Intelligent Video Surveillance

Intelligent video surveillance is a hot industry in recent years, the detection and tracking algorithm based on moving target is the key and foundation for the intelligent video surveillance, but in practical applications, the results of detection and tracking based on moving target are still facing with great challenge.In order to overcome the problem of detection and tracking in practice, studied and summarized the detection and tracking algorithm based moving target in intelligent video surveillance, this paper designed and proposed a series of detection and tracking algorithm based on moving target. There are the main works finished as follow:Section 1, the algorithm of moving target detection. This paper introduced the moving target detection algorithm in intelligent video surveillance, and introduced the briefly advantages and disadvantages of these algorithms. In order to overcome the failing in full foreground extracting of the traditional Gaussian Mixture Model, a developed algorithm is designed.The developed algorithm combines Gaussian Mixture Model with three frame difference method together. It tracks well in full foreground detection, and the problem of the target’s fault is solved.Section 2, the algorithm of moving target tracking. This paper introduced the moving target tracking algorithm in intelligent video surveillance, and introduced the briefly advantages and disadvantages of these algorithms. It focused on Mean-shift tracking algorithm. The mean-shift tracking algorithm based on block is proposed to solve the problem that tracking performance of traditional mean-shift algorithm decline under the occlusion and variation of target.. Firstly, the target is divided into some similarly sized blocks which are tracked unobstructed part of target by traditional Mean-shift algorithm. Secondly, the effectiveness of target blocks are estimated by tracking detector and the invalid blocks are screened from the target, so the problem that the reduction of tracking performance which is caused by target blocks is solved. Thirdly, much more spatial information can be detected by normalized cross correlation detector and neighborhood homogeneity detector, so the limitation of mean-shift algorithm is made up and therobustness of tracking is increased. Experiment results indicate that the algorithm is proposed can track under the occlusion and variation of target efficiently and accurately.Section 3, the tracking algorithm of TLD. This paper focused on basic principle and of TLD tracking algorithm. Aiming at the problem of detector with slowly processing speed, the algorithm introduced kalman filter in the detector, and the kalman filter estimate the location of the target to reduce the scanning region of the detector, it improved the speed of the detector.Comparing with tracking effect and time of the original TLD algorithm and TLD algorithm based on regional prediction. The experimental results show that under the premise of tracking effect,the TLD algorithm based on regional prediction has faster tracking speed than original TLD algorithm.