面向空对空场景的无人机识别与跟踪系统
Aerial-to-Aerial Scene Oriented UAV Recognition and Tracking System
丁逍 1蒋鸿宇 1郭有为 1王坤1
作者信息
- 1. 中国工程物理研究院电子工程研究所,四川 绵阳 621900
- 折叠
摘要
针对空对空无人机识别与跟踪任务,设计了一个基于双目视觉感知和强化学习控制的自主识别与跟踪系统.首先,通过YOLOv8 进行目标检测,并根据双目视觉特点设计了基于语义分割的定位算法,实现了对机动目标的快速准确定位,并运用滤波算法提高了定位信息的可靠性.基于强化学习思想对目标的跟踪过程进行建模,并通过DDPG算法训练跟踪控制器.最后,在高保真的Airsim仿真平台中进行了多组对空中机动目标的识别跟踪实验,验证了所设计的无人机双目视觉识别与智能跟踪系统的有效性.
Abstract
For air-to-air drone identification and tracking task,this paper presents a autonomous recognition and track-ing system based on binocular vision perception and reinforcement learning control.Firstly,target detection is performed using YOLOv8,and a semantic segmentation-based localization algorithm is designed according to the characteristics of binocular vision,achieving rapid and accurate localization of maneuvering targets.Additionally,a filtering algorithm is em-ployed to enhance the reliability of localization information.The tracking process of the target is modeled based on the principles of reinforcement learning,and a tracking controller is trained using the DDPG algorithm.Finally,multiple sets of experiments for identifying and tracking airborne maneuvering targets are conducted on the high-fidelity Airsim simulation platform,validating the effectiveness of the designed drone binocular vision recognition and intelligent tracking system.
关键词
无人机/双目视觉/深度强化学习/空对空跟踪系统Key words
UAV/binocular vision/DRL/aerial-to-aerial tracking system引用本文复制引用
出版年
2024
NETL
NSTL
万方数据