Study On Algorithm Of Moving Target Detection In Video Sequences

Abstract:With the increasing demands of the personal property security and social stability, video surveillance system has been widely used. Since an explosive growth of the storage of video data and the development of the independent intelligent monitoring system, the moving target detection in video sequences has been a core research in intelligent video surveillance system for a long time.The accuracy and real-time performance of the foreground target detection are the two important criteria to evaluate such algorithms. Although the traditional three-frame difference algorithm is simple, the current test results are always dependent on the input of the next time. So the drawback of that algorithm is its real-time performance. This paper proposes a three-frame difference algorithm based on the forward prediction. The algorithm proposed in the paper makes an improvement of the real-time performance without increasing the computational complexity.For some complex scenes, the detection effect of the mixture Gaussian model is better than the other background modeling methods, but the introduction of multiple Gaussian models also increases the computational overhead of the system. This paper combines Gaussian modeling with mean modeling, and presents a mean modeling method based on the directional weighting method. Comparing with the Gaussian modeling, the time complexity of the improved algorithm has been greatly refined, especially in the detection of some high-resolution images. The correctness of detection results are basically consistent with the Gaussian model. Although there is a gap in the image quality, this gap can be narrowed by post-processing of the image.Therefore, the improved algorithm proposed in this paper can decrease the time complexity without sacrificing detection accuracy of the results. And it provides an effective compromise between the accuracy and the real-time performance of the moving target detection.