基于多传感器融合的无人机可降落区域识别方法研究  

Research on the Identification Method of UAV Landing Area Based on Multi-sensor Fusion

作　　者：刘飞[1] 单佳瑶 熊彬宇 方正[1] 杨正权[2] Liu Fei;Shan Jiayao;Xiong Binyu;Fang Zheng;Yang Zhengquan(Northeastern University,Shenyang 110819,China;China Aircraft Strength Institute,Xi’an 710065,China)

机构地区：[1]东北大学,辽宁沈阳110819 [2]中国飞机强度研究所,陕西西安710065

出　　处：《航空科学技术》2022年第4期19-27,共9页Aeronautical Science & Technology

基　　金：航空科学基金(20184123011,201941050001)。

摘　　要：当无人机遇到电量低、丢失遥控信号、失去GPS信号、天气突变等需要迅速降落的紧急情况时,依靠机载的传感器实现无人机自主降落到安全区域显得非常重要。为保证无人机在遇到紧急情况或者收到降落指令后,能够自动识别安全的降落区域,实现安全自主降落,本文提出一种基于多传感器融合和深度学习网络框架的无人机可降落区域识别方法。首先,使用基于无人机机载图像信息搜索安全降落区域;然后利用孪生网络对安全降落区域进行跟踪。当无人机降落到一定高度时,利用机载激光雷达进行近地面环境实时建模与语义分割,确定安全的可降落区域;最后,通过实时建立的可降落区域点云模型,计算出精确的可降落区域位姿信息,供飞控系统实时着陆控制使用。在仿真环境和实际环境中的试验研究表明,基于多传感器融合的方法对可降落区域的识别准确率达到90%,位置识别的误差为5cm,着陆过程地形高程估计误差为2cm,能够满足无人机自主安全着陆的要求。通过对可降落区域识别方法的研究,实现无人机对下方可降落区域的识别,进而引导无人机实现安全自主着陆。When the UAV encounters the emergency situation of low power,loss of remote control signal,loss of GPS signal,sudden weather change and so on,it is very important to rely on the airborne sensor to achieve the autonomous landing of the UAV to the safe area.In order to ensure that the UAV can automatically identify the safe landing area and realize the safe and autonomous landing in an emergency situation or after receiving landing instructions,we propose a method for the UAV landing area identification based on multi-sensor fusion and deep learning network framework.Firstly,the safe landing area is searched based on the airborne image information of the UAV.Then the safe landing area is tracked using twin networks.When the UAV lands to a certain altitude,airborne liDAR is used to conduct real-time modeling and semantic segmentation of the near-surface environment to determine the safe landing area.Finally,the accurate position and pose information of the landing area is calculated by the real-time landing area point cloud model,which can be used for real-time landing control of the flight control system.Experimental studies in simulation environment and actual environment show that the identification accuracy of landing area based on multi-sensor fusion method reaches 90%,the error of position identification is 5cm,and the estimation error of terrain elevation during landing is 2cm,which can meet the requirements of autonomous safe landing of the UAV.Through the research on the identification method of landing area,the UAV can recognize the landing area below,and then guide the UAV to achieve safe autonomous landing.

关 键 词：自主泊降 深度学习 多传感器融合 三维激光雷达 语义分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]