基于GPU的快速影像匹配

Fast image matching based on GPU

随着图形处理器(GPU)的飞速发展和计算同一设备架构(CUDA)的推出,GPU的并行性和可编程性不断增强,本文提出了一种基于CUDA的Harris算子影像匹配并行处理方法,在GPU中完成对影像的灰度化、Harris角点提取、重采样、灰度相关匹配,并从线程分配、内存使用、共享存储器等方面进行优化。实验结果表明,该方法与CPU串行处理方法相比,其速度得到了明显提升。

With the rapid advance of the GPU （Graphics Processing Unit） and the advent of CUDA（ Computer Unified Device Architecture）, the parallelism and programmability of GPU increase over time. The paper trot forward a parallel processing method based on CUDA for Harris operator image matching, and by using the method it accomplished some image processing in GPU, such as image gray processing, Harris corner extraction, image resampling and image gray correlation matching. Furthermore, it optimized the thread allocation, mcmory usage and the shared memory to fully utilize the huge parallel computing capability of GPU. The experimental result indicated that this processing method would be faster than the serial processing of CPU.