Feature Matching Algorithm of Moving Human Bodies

Many video surveillance systems are demanding to monitor moving human behaviors, especially for such environments as jails, warehouses, supermarkets, secret rooms, offices, auditoriums and so on. It has universal significance for image capture of surveillance system to recognize moving human bodies, accordingly, it is an imperative to recognize moving human bodies from other moving objects rather than distinguish differences amongst human bodies. In this paper, the frame difference method is employed for image segmentation of moving objects, matching algorithms such as human body shapes, skin colors and postures are studied. By synthesizing information features of human body shape data and skin color data, a matching algorithm for moving human bodies is established. This algorithm is computationally efficient, and can be applied to resource-restricted systems e.g. embedded systems. The experimental results show that this algorithm can accurately segment moving human bodies from complex background, and can greatly reduce the calculation workload, in addition, it needs less storage spaces and has good robustness and accuracy. Therefore this algorithm can be pragmatically applied to video surveillance via automatically controlling lenses towards moving human bodies.