Parallel Optimization For Video Moving Object Detection And Tracking Algorithm Based On GPU

With the development of economy, society, and software and hardware technology, kinds of research and exploitation on image recognition is becoming a hot spot in recent years. Intelligent video analysis synthesizes subjects of mode recognition, machine vision, image processing, artificial intelligence, etc. Therefore it becomes a research task with great application value. Moving target detection and tracking, which provides a foundation for a higher level analysis,is an important research branch of the intelligent visual surveillance. The result of detection and tracking has great impact on following step. people have high requirements on detection accuracy and operational efficiency of the related algorithm. In recent years, with the rapid development of graphic hardware and the promotion of cost-effection, using graphic cards to accelerate general purpose computing is becoming more and more popular. GPGPU computing provides an effective method to accelerate the speed of digital image processing.This thesis analyze the advantage of CPU-GPU heterogeneous system by comparing the CPU and GPU computing characteristics, and gives an introduction to NVIDIA’s CUDA parallel computing architecture and standard library-OpenCL which can do parallel processing across CPU and GPU. CUDA and OpenCL makes GPU has a better programmability, and has a good effect of acceleration to the image processing algorithm. For all of the background modeling methods the Gaussian mixture modeling is considered as a good method width high performance in both detection capability and adaptability. Nevertheless it is difficult to implement in real time for its huge computational complexity. Based on CUDA programming environment on GPU, the parallel improvement is applied to a Gaussian mixture background modeling algorithm. Compared with serial algorithm can achieve nearly five times accelerate effects, faster than the OpenCV CUDA implementation at the same time, it is proved the effectiveness of the parallel optimization algorithm. The experimental results show that the parallel optimization is effective. Using a single detection and tracking algorithm has been unable to meet the practical requirements of target tracking in a complex environment. Based on the optical flow and compressive tracking algorithm, this paper proposes a tracking method combine with tracking, online learning and detecting technology. The experimental results show that the proposed method effectively improves the accuracy and robustness of tracking. Optical flow is a kind of excellent tracking algorithm which has huge computational complexity. Considering the portability on different platforms, so this paper studies the parallel optimization of optical flow algorithm based on OpenCL, effective use of hardware resources and speed up the execution efficiency of the algorithm. Experiments compared with the traditional serial algorithm can achieve near about three times accelerated effects, even faster than the OpenCL version in OpenCV.