A High Performance Real-Time Edge Detection System with NEON

This paper proposes a high performance real-time edge detection system based on Xilinx Zynq platform to support multiple resolutions dynamically. The previous systems implemented on FPGAs can achieve better performance than CPU at the cost of hardware resources, and the performance improvement is constrained by the low frequency and limited resources of FPGA. In this paper, we employ the ARM NEON engine to accelerate the edge detection algorithm. Optimization methods are introduced to reduce computation complexity and memory access. We also employ data preload, instruction scheduling and dual-core processing to make full use of NEON engine and deliver a better speedup. In our experiment, the Prewitt operator is used to realize the edge detection. The system can achieve a processing time of 2.95ms for a 1280 x 720 image, and 7.25ms for a 1920 x 1080 image.