基于双目视觉的固定翼无人机自主着陆技术研究  

Research on Autonomous Landing of Fixed Wing UAV Based on Binocular Vision

作　　者：马晓东 吕昊 张杰[1] 高帅华 MA Xiaodong;LYU Hao;ZHANG Jie;GAO Shuaihua(Chinese Flight Test Establishment,Xi’an 710089,China)

机构地区：[1]中国飞行试验研究院

出　　处：《兵器装备工程学报》2019年第11期193-198,共6页Journal of Ordnance Equipment Engineering

摘　　要：针对固定翼无人机自主着陆需求,结合旋翼机自主着陆方法优势,提出一种基于双目视觉的固定翼无人机自主着陆方法。分析无人机着陆过程,设计地标形状及其摆放方式;基于轮廓几何特征及SVM分类器完成目标识别。为保证算法实时性,对目标图像进行网格采样,以基于前后双向跟踪误差与相邻帧局部区域图像相似性约束改进的金字塔L-K光流法为核心对采样点进行跟踪;设计目标再搜索算法,进行跟踪失败再搜索,保证跟踪稳定性。最后,提取亚像素级特征角点,基于透视投影理论完成相对位姿解算,并进行双目均值融合。实验结果表明:算法实时性、稳定性较好,有较高估计精度,可满足固定翼无人机自主着陆需求。Aiming at the requirement of autonomous landing of fixed-wing UAV,it proposed a method of autonomous landing of UAV based on binocular vision,which combined the advantages of autonomous landing of rotorcraft.The UAV’s landing process,and design the shape of landmark and its placement mode were analyzed.It recognized the target via contour geometric features and SVM classifier.To ensure the real-time performance,initialize the target by grid sampling,and track the sampling points via pyramid L-K optical flow was improved by forward-backward tracking error and the similarity constraints of adjacent frames.And it designed target research algorithm for failure tracking to ensure tracking stability.Finally,it extracted sub-pixel corner points,calculated relative position and attitude based on perspective projection theory and performed the binocular mean fusion.The experimental results show that the algorithm has high estimation accuracy,good real-time performance and stability,and can meet the requirements of autonomous landing of fixed-wing UAV.

关 键 词：SVM分类器 金字塔L-K光流法 前后跟踪误差 位姿解算 双目均值融合 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]