Multi-moving Objects Adaptive Tracking System Based On DSPs

This paper mainly discuss the dynamic vision system which track multi-moving objects in the complex background in real-time. We propose a simple and fast scheme to detect moving object in complex background which is the combination of Inter-frame difference, morphologi- cal filter and optical flow field. It's simple and fast to detect moving object using inter-frame difference. However, it's lack of precision and moving information. Optical flow field contains not only the moving information of the object, but also three-dimensional information. This paper integrates the inter-frame difference and optical flow methods. The method of inter-frame difference is used to separate moving area from the whole image. Afterwards, to decrease noise which is brought by complex background, the method of morphological filter is used. Planar velocity field in the moving area can be estimated through optical flow analysis. The improved differential technique is adopted when estimating optical flow field. Then the precise value of moving parameters is achieved.Through the significant test data we certificate the feasibility and robust of the techniques of multi-moving objects detection, feature extraction and selection and moving state detection. Base on these reliable theory, we build a system which track multi-moving objects. DSPs is used to accurately detect the moving area of multi-moving objects and information of their geometrical position in the system, at the same time, it pick up the parameters of geometrical feature, gray feature and planar velocity of each area, at last, we integrate the techniques of multi-moving objects detection and construct a self-adapted system to track multi-moving objects. The camera is controlled to keep the object in its field of view. Therefore we achieve a real time system to track the multi-moving objects in the complex scenes. This system consists of an image capture card, a personal computer, pan/tilt and DSPs board. The result of this research shows that this method can make the real-time tracking for fleetly moving objects in the complex background come true.