Research On Moving Human Body Recognition System Based On DSP

The human body recognition system based on video processing platform has obtained more and more attentions nowadays, for the video processing system can achieve the goal of it more effectively, and the technology of digital signal processors develops rapidly. At present, the human body recognition system is being used widely by many applications such us: human activities analysis, virtual real-life, security surveillance, etc. In this paper, a human body recognition algorithm for this system is proposed. And we have implemented and optimized this algorithm in DSP platform for real-time requirements. This algorithm consists of foreground segmentation, human head location, human body parts identification, human model construction etc. In contour analysis and human head location, a two step view-based head location algorithm acting on foreground region is proposed, which can also successfully locate the head under the ambiguous conditions. Assuming the people detected are all stand-up, a subjective-partition algorithm based on NCMs (Negative Curvature Minima) is also presented to accurately identify the human body parts (arms, legs) with a low cost of computation. Meantime, we propose an embedded video processing system which adopts the digital video decoder(SAA7111A)/encoder(SAA7121H) for conversion between the standard analog video signal and the digital video data. This structure makes the system capable of changing image sensors and monitors freely according to the requirements of practical applications. At lastly, the performance and results of the system are given. The experimental results show that the whole proposed systems are able to effectively locate the human body parts and complete the human model.