GPU Based Design And Realization Of Video Information Parallel Processing System

With the popularization of video hardware and the high demand of HD video, the amounts for people who want to obtain the video information have rised more and more quickly. In the industry, the demand of high performance parallel solutation has also increased quickly. In the face of doubling data quantity, the combined model of CPU and serial video processing algorithms could not well meet the challenge of the high-speed computing of large-scale data. The increasing programmability and high performance computational power of GPU in modern graphics hardware provide a great scope for acceleration of video processing algorithms which can be parallelized. Therefore, general-purpose computing based on GPU has become a very hot research topic in the field of high-performance computing.To develop the parallel algorithm of recognition and tracking of the target object, in this paper, an effective solution was proposed, and an integrated system was developed. The main work is as follows:(1) In video information processing, a heterogeneous architecture consisted of CPU and GPU schemes for moving target recognition and tracking algorithms was given by studying the field of template matching target recognition algorithm and MeanShift target tracking algorithm in parallel processing on the GPU. GPU-based massively parallel computing design patterns were explored in image proeessing. For the difference between GPU and CPU, the acceleration principles of GPU were analyzed, and the general-purpose computing model of current mature framework CUDA and its characteristics were discussed.(2) The mode of design and optimization were explored. In the algorithms design, function partition and reasonable mode of thread assigning were used. In the algorithms optimization, memory access optimization and communication optimization were used for parallel algorithm of template matching target recognition. Shared memory and reduction tree algorithm were used as performance optimization methods in parallel algorithm of MeanShift target tracking.(3) According to the research from the above, the video information parallel processing system based on GPU was designed and achieved. The system searched the target positioning by the automatic identification, and then tracking. The results showed that the GPU-based parallel Mean Shift target tracking algorithm could get a higher speed-up ratio.