Improving mixture Gaussian background model by integrating trace information obtained from Kalman filter

Intelligent video analysis, which analyzes behaviors of moving objects in the scene, determines their trajectories, morphological changes and detects abnormal behaviors by setting certain rules, is a combination of techniques such as image processing, computer vision, artificial intelligence, and so on. The very algorithm system mainly includes four parts. They are foreground extraction, object recognition, tracking and high-level processing named behavior understanding. Among them, foreground extraction is the most crucial and basic part which has great impact on the follow-up operations. Our work in this paper improves mixture Gaussian background model, a popular foreground extraction algorithm, by integrating trace information obtained from Kalman filter. This helps remove large blocks of noise caused by suddenly illumination change or non-periodic sway of branches and get a more accurate mask image.