Motion Segmentation via On-line Gaussian Mixture Model and Texture

Motion segmentation is a fundamental problem in video-based motion analysis. This paper implements a new method to detect moving objects, based on the linear fusion of color and texture features. The on-line Gaussian mixture model is employed to update the background image, and to calculate for every pixel the color difference and its corresponding color weight. Texture information is used on the other hand to describe the local structure information of pixel neighbor so as to enhance the robustness of motion segmentation. The experimental results of different real scenes show that the proposed method is effective and practical.