Real-time person detection method without background model

This paper discusses how to detect the person in natural environments and in real-time without background model. Contrary to conventional method, we propose a system which can detect the person in the complex background without the learning about empty scene. This system employs an adaptive filter, closing method, and multiple cues such as motion, color, shape and size to detect the head and hands part as features for person's tracking in consecutive frames. Since the proposed method does not need to initialize the background model, it works well even if the camera view is changed or large changes occur in the scene.