Abstract: An improved Harris corner detection algorithm was proposed to solve the problem that the number of detection corners was too few and the accuracy was low when detecting infrared images. Firstly, the Canny operator was used to detect the target edge to determine the target region, and then the cubic B-spline gradient operator was used to filter the target area to obtain the M matrix. Then the convolution of the cubic B-spline function and the Gaussian window function was used to replace the original Gaussian window function for data filtering, and extract the corner points of the target area. Finally, the adaptive threshold was used for non-maximum suppression to remove false corners. The experimental results show that compared with the Harris algorithm, the proposed algorithm not only increases the number of corners points extracted, but also reduces the running time of the algorithm by 3/4.