基于深度学习的无人机飞行通道辅助系统研究  

Research on UAV Flight Channel Assistance System Based on Deep Learning

作　　者：伍彩云[1] 翁宇 白帆 WU Caiyun;GONG Yu;BAI Fan(Shenyang Ligong University,Shenyang 110159,China)

机构地区：[1]沈阳理工大学装备工程学院,沈阳110159

出　　处：《沈阳理工大学学报》2023年第6期1-9,共9页Journal of Shenyang Ligong University

基　　金：国家自然科学基金青年基金项目(62102272)。

摘　　要：针对复杂环境下高速飞行的穿越机面临碰撞、坠毁等安全威胁的问题,设计一种新的高动态无人机飞行通道辅助系统,将基于景深的人工势场法与深度学习算法相结合,设计一个深度通道轨迹预测网络(DCTN)预测景深信息及其飞行通道,并结合无人机位姿信息和视角图像预测当前位置下可能的避障轨迹,以避免无人机与障碍物的碰撞。使用Jetson TX2作为机载图形处理单元进行验证实验,结果表明,DCTN算法的每个轨迹所需要的生成时间比传统人工势场法有显著的降低,而且能以较低成本达到与碰撞检查和规划算法RAPPIDS相同数量级的响应时间,能够满足无人机在高动态场景下的应用需求。For the problem that collision,crash and other safety threats may happen to a high-speed flying First Person View Droneuses in a complex environment,a new highly dy-namic UAV flight channel assistance system is designed,combining the depth-of-field based artificial potential field method with deep learning algorithms to design a depth channel traj-ectory prediction network(DCTN)to predict the depth-of-field information and its flight channel,and combining the UAV attitude information with viewpoint image to predict the possible obstacle avoidance trajectory under the current position,so as to avoid the collision between the UAV and the obstacles.The results show that the proposed DCTN algorithm needs significantly less time to generate each trajectory than the traditional manual potential field method,and can achieve the same order of magnitude response time as the collision checking and planning algorithm RAPPIDS at a lower cost,which can meet the requirements of UAV applications in highly dynamic scenarios.

关 键 词：人工势场 深度学习 轨迹规划 轨迹预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]