A High-Throughout Real-Time Prewitt Operator on Embedded NEON+ARM System

To tackle the problem of inadequate performance in data processing on existing embedded systems, in this paper, we present a well-optimized parallel image processing scheme based on the NEON acceleration engine. It targets at the Prewitt operator, a classical image edge processing method with good properties including high precision and strong anti-noise ability. However, it requires a high computational cost and leads to insufficient real-time processing capability. In this work, by exploring the parallel processing characteristics of the NEON engine, we concatenate R, G, B three-channel data in parallel, optimize the data processing schedule of the Prewitt operator and implement the real-time video processing on Xilinx's Zynq-7010 all programmable system-on-chip (APSOC) development platform. Experimental results demonstrate that the proposed design can achieve approximate 4 times speedup compared to a regular design using ARM C library. It is almost 2 orders of magnitude faster than using ARM Coretex-A9 without NEON.