Moving Object Detection System Research Based On Dsp

Moving object detection system is an important component of the intelligent video analysis system, which has gradually promotion and application. Whereas, The traditional PC-based moving object detection system exists some constraints including high hardware requirements, large volume of power, real-time poor that can't achieve large-scale applic-ation.The combination of video analysis technology and embedded technology offers the solution to these problems. The DSP-based moving object detection system utilized the DSP chips' characteristics containing high-performance of digital signal processing, cheap-ly, facilitating the integration of applications to lay the foundation for constructing a high practicability and product's video processing platform. Therefore, it is a valuable and meaningful topic.In this paper, aiming the design and implement of the DSP-based moving object detec-tion system, we carry out the research into the existing moving object detection technology, analyze some of the most commonly used moving object detection algorithm and contrast these algorithms' advantages and disadvantages. Also, we do in-depth study on Gaussian mixture modeling (GMM) which has been commonly used in complex scene. Compares and analyses the difference between the platform of DSP and other hardware. Comparing with other embedded processing chips, this paper illustrated both hardware performance and software development advantages of video processing using DSP. Then we do a further comparison of the differentia of DSP and PC platform software exploitation.With the comparision and analysis, the paper finally chooses DM642 as the embedded-core processing chips and takes seed-vpm642 as development platform to study the design and implementation methods of moving object detection system. On the algorit-hm, combining the color space's difference between DSP video capture signal and PC, we improve the GMM algorithm, construct Gaussian models in YUV space, make use of Y channel to supply Gray-scale signal, optimize the procedure of model matching, reduce the models updating computation and improve shadow suppression while use the property of shadow uniform attenuation in YUV space. The experimental results indicate that the modified algorithm achieves good effects. On the implementation, using the DSP/BIOS real-time system kernel and the RF reference framework, we do the GMM algorithm's transplantation and optimization. By the procedures, such as driver configuration calling, memory space allocation, simultaneous multi-threading and so on, we accomplish a video capture, processing, output of moving object detection system which can implement a quasi-real time moving object detection in complex environment.