High Performance Circle Detection through a GPU Rasterization Approach

GPUs have recently attracted our attention as accelerators for a wide variety of algorithms, including assorted examples within the pattern recognition field. After proving the effectiveness of the GPU for computing the Circle Hough Transform [Ujaldon et al. (2008)], we present in this paper a radius compensation method which improves the accuracy and speed-up using the GPU rasterizer. Experimental results confirm a higher precision in circles detection and up to 27% additional savings in the GPU computational time on a GeForce 8800 GTX for a 1024x1024 sample image, thus enhancing the execution of the Circle Hough Transform on a low-cost and emerging architecture like the GPU.