采用GPU的视觉定位系统加速实验  

GPU-based Visual Positioning System Accelerated Experiment

作　　者：王玲玉 董文博[2,3] WANG Lingyu;DONG Wenbo(University of Chinese Academy of Sciences,Beijing 100049;Technology and Engineering Center for Space Utilization,Chinese Academy of Sciences,Beijing 100094;Key Laboratory for Space Utilization,Chinese Academy of Sciences,Beijing 100094)

机构地区：[1]中国科学院大学,北京100049 [2]中国科学院空间应用工程与技术中心,北京100094 [3]中国科学院太空应用重点实验室,北京100094

出　　处：《计算机与数字工程》2020年第1期78-82,共5页Computer & Digital Engineering

摘　　要：计算机视觉在实际应用中的主要瓶颈是系统的准确性和实时性,而实时性的提高又受限于计算机中处理器的处理速度。在一个视觉定位伺服实际系统中,要求有很高的实时性,尝试研究使用GPU建立实时的视觉系统的实际性。视觉定位系统软件分为图像粗定位(目标识别)、图像精定位、运动解算等几部分。当识别图像像素为640×480时,采用单独GPU加速的方式,比CPU加速了2.1444倍,采用CPU和GPU相结合的方式,比CPU加速了4.1548倍。The main bottleneck of computer vision in practical applications is the accuracy and real-time performance,and the improvement of real-time performance is limited by the processing speed of computer processors.A visual positioning servo sys tem requires high real-time performance,so it is tried to study the practicality of using a GPU to establish the real-time vision sys tem.Visual positioning system software is divided into image coarse positioning(target recognition),image accurate positioning,motion resolution and other parts.When the image pixel is identified as 640×480,a nd a separate GPU accelerated approach is used,the processing speed is 2.1444 times faster than the CPU.With a combination of CPU and GPU,the speed is accelerated to 4.1548 times faster than the CPU.

关 键 词：视觉定位系统 实时性 粗定位 精定位 GPU加速 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]