CUDA Based Moving Object Detection

Moving object detection is one of the basic image processing topics. In the field of intelligent video surveillance, as a key algorithm in the inspection process, moving object detection occupies a very high position. Because its result has great impact on follow-up monitoring and analysis steps, people have put forward high requirements for detection accuracy and operational efficiency of the moving object detection algorithm.For the traditional moving object detection algorithms, there’re many pros and cons. In general, the algorithms enjoying better detection results and better adaption of complex scene have better practical value. But they usually cost more time and have worse efficiency, and cannot meet the real-time requirements. So, it is hoped that some good algorithms can be improved to enjoy faster speed.In recent years, with the rapid development of graphics hardware and the promotion of cost-effection, using graphic cards to accelerate general purpose computing is becoming more and more popular. GPGPU is belonging to the branch of high-performance parallel computing, and can well-meet the requirements of large amount data processing. If optimized properly, the calculation speedup over CPU can be hundred timers or more. The image processing algorithms, heavily involved in matrix operations, are very easily to get parallel acceleration and enjoying good performance. CUDA technology is a GPU GP model introduced by NVIDIA. Later in this article, there will be the introduction of CUDA parallel upgrading and speedup of2algorithms-GMM&HOG.GMM&HOG are two algorithms with good results but long running time, and have high practical value. GMM algotithm has been applied in the field of building monitoring and traffic analysis, but HOG algorithm is still less used because of its complexity. In this article, this2algorothms will be optimized with CUDA technology, and achieve the goal of real-time requirements. Finally, by doing data analysis and comparison, we can draw relevant conclusions.