Multi-directional Sobel operator kernel on GPUs

Sobel is one of the most popular edge detection operators used in image processing. To date, users utilize two-directional 3×3 operator as detectors because its low computational cost and reasonable performance. Simultaneously, many studies have been conducted on using large multi-directional to satisfy their needs considering high stability, but at an expense speed. This paper proposes a fast graphics processing unit (GPU) kernel for four-directional 5×5 operator. improve performance, we implement based warp-level primitives, which can significantly reduce number memory accesses. In addition, introduce prefetching mechanism transformation into complexity data transmission latency. Compared with OpenCV-GPU library, our shows performances 6.7x speedup Jetson AGX Xavier GPU 13x GTX 1650Ti GPU.