Research On The Target Detection Method Of Codebook Used In Embedded Video Surveillance

Target detection intelligent monitoring system is the most basic and critical tasks. Target detection is the process of dividing the moving prospect object from a video sequence. In real life scenarios, the background is often more complex, in order to meet the needs of people in all aspects of life, the algorithm needs to be applied to embedded monitoring systems. Therefore, to improve the system’s real-time and accuracy in a complex background, target detection algorithm has become a critical issue. In this paper, this algorithm has been further improved based on previous studies, the algorithm main improves the algorithm’s real-time. In this thesis, the main work done in three parts, the first is the study of the previous algorithm, the second is the concrete realization of the improved algorithm, and the third is to use the experimental results to verify the validity of the algorithm. Details are as follows:1. Base on in-depth research of the existing target detection algorithms, an article proposed to resolve target detection algorithm cannot meet the real-time and small complexity of the embedded system in complex background. According to the results summarized the advantages and disadvantages of existing target detection algorithms:frame difference method, optical flow method, background subtraction, and comparing advantages and disadvantages of the average background subtraction method,Gaussian mixture method in the method of background subtraction algorithm.2.According to previous studies,research on the target detection algorithm of local codebook in embedded systems which first step is convert video sequence which RGB color space to YCrCb space. Then in the background modeling using frame difference image drawn into a rough background pixel that can cluster accurate codebook background model.In target detection stage, comparing the video frames with the trained model segment foreground out from the background and updating the background model in real time.3.With the extensive use of linux systems in embedded systems, we realize the moving target detection using open source vision library OpenCV in linux system, and complete the task of transplanting the OpenCV to linux. Finally, this paper demonstrates and analysis detection algorithm to verify the effectiveness of the algorithm. Experimental results show that the method’s background modeling speed, low computational complexity, and can effectively detect the target in the complex scenes such as light changes and waving trees, and can handle shadows, reducing the false detection rate.