基于单目视觉的无人机避障方法研究综述  

A review of UAV obstacle avoidance methods based on monocular vision

作　　者：张东旭 张晶 李娜[3] 洪功名 刘明建 Zhang Dongxu;Zhang Jing;Li Na;Hong Gongming;Liu Mingjian(Army Military Transportation University,Tianjin 300171,China;The Unit 92941 of PLA,Huludao 125000,China;Beijing Electro-mechanical Engineering Institute,Beijing 100074,China)

机构地区：[1]陆军军事交通学院,天津300171 [2]中国人民解放军92941部队,葫芦岛125000 [3]北京机电工程研究所,北京100074

出　　处：《战术导弹技术》2024年第2期83-92,105,共11页Tactical Missile Technology

摘　　要：单目视觉相对传统立体视觉具有设备简单、计算成本低等特点,目前已在视觉避障领域得到了广泛应用。然而,不同的单目视觉避障算法在鲁棒性、准确性和实时性等指标上差异较大,实际使用中需做出权衡。因此,分析了基于光流、特征匹配、机器学习的避障方法的研究进展与现状。在鲁棒性、准确性、实时性等方面将三种方法进行了对比,指出了各自的优缺点,并给出了目前制约无人机单目视觉避障发展的难点与挑战,指出了低成本、高精度、智能化的避障方案是无人机单目视觉避障领域未来的重点研究方向。Compared with traditional stereo vision,monocular vision has the advantages of simple equipment and low computational cost,which has been widely used in the field of visual obstacle avoidance.However,different monocular visual obstacle avoidance algorithms differ greatly in robustness,accuracy,real-time and other indicators,so trade-offs need to be made in practical use.Therefore,research progress and status of obstacle avoidance methods based on optical flow,feature matching and machine learning are analyzed.The proposed three methods are compared in terms of robustness,accuracy and real-time,and the advantages and disadvantages of each method are listed.Moreover,the challenges that currently constrain the development of monocular visual obstacle avoidance for UAVs are given.It is pointed out that the low cost,high precision and intelligent obstacle avoidance scheme are the future key research directions in the field of UA V monocular vision obstacle avoidance.

关 键 词：无人机 单目视觉 避障 光流 特征匹配 机器学习 强化学习 

分 类 号：V279[航空宇航科学与技术—飞行器设计]