Tracking and Classifying Moving Objects from Video

This paper presents a method for tracking moving objects in an outdoor environment, classifying them into three categories: single person, people group or vehicle. The proposed method integrates motion, spatial position, shape and color information to track object blobs. The blobs are segmented using the pixel-wise difference with the background, which is automatically updated by a median filter. Our method focuses on the establishment of a correspondence between objects and the templates as the objects pass into view. Principal Component Analysis (PCA) is applied to extract color features to reduce dependence on light change. The variations of motion direction and of compactness are used as the classification metric. The efficiency of our method is illustrated and confirmed by our experimental videos.