注:武汉市科技局项目(NO. 2017010201010124)
作者:葛迪,邱程,侯群
单位:江汉大学,湖北武汉 430056
中图分类号:TP274+.3
文献标识码:A
文章编号:1006-883X(2018)11-0013-05
收稿日期:2018-10-09
摘要:在现代工业领域中,需要处理的图像数据越来越庞大,单平台单线程的计算效率已经不能满足于需求。从多核CPU,GPU,DSP再到FPGA,以及它们之间的多种搭配组合,虽然实现了并行运算的操作,但却不能解决软硬件协同设计的难题,且其功耗较大,受限较多。针对这类问题,本文通过结合FPGA平台与OpenCL标准来将OpenCV中一些关键的识别算法转化为硬件模块,从而提高图像数据处理速度。将此系统应用于传送带测试,可以将传统的传送带速度由0.5m/s提升至2.5m/s,并在此速度下进行高速识别,进而在工厂、物流等领域代替人工操作,实现商品的自动分拣和分类统计功能,提升工作效率。
关键词:并行计算;OpenCL;FPGA;高速识别;硬件加速
Moving Object Recognition System Based on FPGA and OpenCL Hardware Acceleration
GE Di, QIU Cheng, HOU Qun
Jianghan University, Wuhan 430056, China
Abstract: In the field of modern industry, the image data needed to be processed is becoming more and more huge, and the computing efficiency of single platform and single thread can not meet the requiements. Multi-core CPU, GPU, DSP and FPGA, as well as a variety of combinations among them, can achieve parallel operation, but not solve the problem of hardware/software codesign, and have large power consumption and more constraints. To solve this problem, the research is completed in this paper to transform some key recognition algorithms in OpenCV into hardware modules based on FPGA platform and OpenCL standard so as to improve the processing speed of image data. The tests show that the system can increase the speed of traditional conveyor belts from 0.5m/s to 2.5m/s, at which the high-speed recognition can still be carried out. And then the system can be used in the factory, logistics and other fields instead of manual operation to achieve automatic sorting and statistics of goods and improve work efficiency.
Key words: parallel computing; OpenCL; FPGA; high speed identification; hardware acceleration
阅读全文
备注:2018年 第24卷 第11期