Research And Application Of Pedestrian Recognition Algorithm Based On DSP In Intelligent Monitoring

In recent years, with the fast development of technology, the video surveillance system was transforming from the third-generation called network video surveillance to the fourth intelligent video surveillance. In the intelligent surveillance, video / image processing has become an important research topic, moving target detection, particularly pedestrian detection has become the focus of the study of intelligent surveillance.This paper studies the video processing to extract moving target and detect pedestrians in full HD intelligent surveillance system. The main work of this paper are:1. Analysis the function of the entire intelligent surveillance system and find out the shortcomings of architecture of intelligent surveillance system.In this paper, smart camera with DSP chips is used to design a framework of intelligent surveillance system which the DSP is used to detect moving objects and the PC is used to detect pedestrian in the moving area.2. Careful analysis of traditional pedestrian detection algorithms with multi-scale sliding window detection and find ont these algorithms are unefficiency. Depending on the application scenario, with the priori information that the proposed pedestrian is always in motion, I design algorithm to detect pedestrian in moving area and the algorithm will reduce the number of detection windows.3. According to the characteristics of DSP processors, I analysis and comparison of a variety of motion detection algorithm.Then I choice algorithm called Vibe to detection moving objects. On the basis of motion detection, the design method based on the movement of the pedestrian area, instead of the conventional multi-scale sliding window.4. Detailed analysis the feature of HOG and HOGLBP. For computationally intensive problems, use principal component analysis(PCA) to reduce the dimension of HOGLBP features. We verify the result in theory and experiment and finger out the principal component analysis reduce the time-consuming and the pedestrian detection effect is guaranteed.5. According to the overall design and motion pedestrian detection algorithm, I realized the prototype system.On careful analysis of the time-consuming problem of the system algorithms, compiler options, and double buffering operation to optimize the code. Based on a careful analysis of the structure of algorithms,I using the SSE instruction set extensions for vector inner product operation carried out targeted optimization, and creatively presented using C64 x + instruction 3×3 median filter algorithm optimized. Experiments show that, under the same conditions, the paper optimized 3×3 median filter algorithm is faster than TI provides library functions in imgLib 5.24 times.This design of intelligent motion detection algorithm monitoring system will be distributed to each DSP smart cameras, PC only suspected sports area for scanning and classification, low system resources; transmission of high bit-rate video streams only in exceptional circumstances, the network load is low. Experiments show that the design of intelligent monitoring system in full HD video can effectively detect the moving pedestrians, speed, and good detection results, fewer missed and false detection.