The invention provides an image surface feature point center quick positioning method based on a GPU. The method comprises the following steps that an original image is obtained by a CPU and transmitted to a global memory; the GPU obtains the transmitted original image from the global memory, Gaussian low-pass filtering is carried out on the original image, and the original image is sent back to the global memory; storage space is distributed on a shared memory; histogram statistic is carried out by the GPU on the basis of the filtered image, and statistic information is transmitted to the global memory; maximum between-cluster variance calculation is carried out on the histogram statistic information obtained by the CPU, and threshold value obtained through the calculation is transmitted to the global memory; binarization processing is carried out on the image by the GPU, and the binarized image is transmitted to the global memory and the CPU; a region of interest is extracted by the CPU and then transmitted to the global memory; the GPU carries out region constraint judgment on the basis of information of the region of interest, and an obtained effective object region is transmitted to the global memory; point center calculation of the region is carried out according to the obtained effective object region, and results are transmitted to the global memory and the CPU.