Research And Implementation Of Moving Target Detection Parallel Algorithm In The Complex Environment

As people continue to raise safety awareness and ongoing advances in science and technology, massive high-definition video manual analysis requires a lot of manpower and material resources. Intelligent video analysis can be done without human intervention under the camera to capture video sequences for automatic analysis, among which intelligent video surveillance has attracted wide attention. Moving target detection is most important in intelligent video analysis; it determines the quality of the results of intelligent video surveillance. Traditional moving object detection algorithms are executed on the CPU serially, with limited speed, to shorten the running time of the algorithm, and takes advantage of hardware acceleration processing them.Recently, the successful examples of the use of GPU (Graphics Processing Unit) technology to solve high-performance general-purpose computing and computer vision resulted in more and more problems. Nvidia’s CUDA computing architecture introduced a unified platform that enables programmers to do better programming for the GPU to further optimize parallel algorithms.The main work is as follows:First, the study of the static background of moving target detection parallel algorithms, and implemented on the GPU. In the static background, background modeling and subtraction methods are most commonly used detection methods, in many background modeling algorithms, Gaussian mixture background modeling (Gaussian Mixture Modeling, GMM) algorithm has been proven to be effective and adaptable than high detection algorithm, but the longer running time of the algorithm can not meet the real-time requirement. This is due to the traditional Gaussian mixture model, the number of Gaussian model for each pixel in the initialization value will be set to not change. To solve this problem, this paper presents an adaptive mixture Gaussian model background modeling algorithm which can dynamically change the number of Gaussian model, and improves and optimizes it, so that the video image of each pixel with the GPU threads correspond, greatly reduce the processing time and improve the processing efficiency. In this paper, different resolution video simulation experiment results show that under the same effect, the proposed method can improve the processing efficiency to meet real-time requirements.Second, we study the dynamic context of moving target detection parallel algorithms, and implement it on the GPU. By moving target detection method commonly used dynamic background is the first video image motion compensation, making dynamic background into static background, and then use a static background moving object detection algorithms to detect targets. However, the whole process of the huge amount of calculation, can not meet the needs of practical applications. By GPU optical flow method and accelerated motion compensation processing to obtain a static background, and then use an adaptive Gaussian mixture background modeling proposed parallel algorithm to detect moving objects. Experimental results show that this method can improve the speed and efficiency of detection.