Dynamic background modeling using intensity and orientation distribution of video sequence

摘要

Moving object detection in a video sequence is a challenging task in presence of dynamic background. In this paper, we propose a novel approach for background modeling by exploiting orientated patterns present in a video scene. Based on the observation that there exists a difference in directional edge patterns between foreground and background, we use the statistical measures of the orientation of texture via two angle co-occurrence matrices (ACMs). Orientation based features extracted from ACMs are then clubbed with intensity distribution-based features extracted from well-known gray level co-occurrence matrix (GLCM) to model the dynamic background. The model is then used to classify pixels within a video frame into background and foreground. Experimental results on a diverse set of video sequences have shown the effectiveness of the proposed method over competing schemes.