摘要 无人机实时图像应用（Realtime Image Applications based on Unmanned aerial vehicle,RIAU）在民事和军事领域具有广泛的应用前景，研发这种系统面临着许多挑战.本文提出了“人在环路上”RIAU系统的概念，采用人工智能技术来提升RIAU系统的能力；研究了人在环路上RIAU的典型计算模式，分析了地面计算和机载计算两种模式系统的特点；研究了RIAU系统的关键技术和基于YOLOv3算法的图像目标检测方法；设计实现了RIAU原型系统.试验结果表明，系统的总时延主要取决于计算单元处理时延和通信单元传输时延；在机载计算模型下利用神经网络计算棒，目标识别时间不超过1秒；采用4G技术，通信单元之间的距离可以不受限制.
Abstract：Real-time image applications based on UAV（RIAU）have a wide range of promising applications in the civil and military fields,and the development of such systems faces many technical challenges.This paper firstly proposed the concept of "people on the loop" RIAU system,which enhances the capabilities of the RIAU system by using artificial intelligence technology.Secondly,the typical calculation models of RIAU on the loop were investigated and the features of the ground computing and airborne computing systems were analyzed.In addition,the key technologies of RIAU system and image target detection method based on YOLOv3 algorithm were studied,a RIAU prototype was implemented.Finally,experimental results show that the total delay of the system mainly depends on the processing delay of the computing unit and the transmission delay of the communication unit.Under the airborne computing model,the target recognition time does not exceed 1 second by using the neural network computing stick.With 4G technology,the distance between communication units of UAVs and Ground station maybe unlimited.
窦晓磊,陈鸣,陈兵,徐亚欣. 无人机实时图像应用系统的计算模式研究[J]. 小型微型计算机系统, 2021, 42(6): 1236-1242.
DOU Xiao-lei,CHEN Ming,CHEN Bing,XU Ya-xin. Research on Computing Models of Systems for Realtime Image Applications Based on UAV. Journal of Chinese Computer Systems, 2021, 42(6): 1236-1242.
2021, Vol. 42 
