Accelerating Harris Algorithm with GPU for Corner Detection

Harris algorithm is a corner detection method based on gray scale, which detects corner points by calculating the gradient changes of pixels in horizontal and vertical direction. Due to the usage of Gaussian filtering, Harris algorithm performs well on aspect of robustness, accuracy and stability. A number of improved Harris algorithms have been proposed to enhance the accuracy and robustness during recent decades. However, the execution speed is limited by a large quantity of calculations especially for Gaussian smoothing, which has remained to be a bottleneck in real-time vision processing. Graphics processing unit (GPU) has special parallel computing resources in hardware. Single Instruction and Multiple Data (SIMD) architecture of shader processor array is suitable and fast for mass data calculations in parallel. Therefore, it provides us with an alternative implementation on GPU which can solve such a bottleneck problem. The paper proposes an implementation of Harris corner detection algorithm on GPU by using open graphics library (OpenGL) and graphics library shading language (GLSL) for portability. The experiment results show that the full execution performance of the implementation on GPU is over 73 times speedup of that completely on CPU at most, and can meet the requirements of real time, accuracy and robustness.