| 摘要: |
| 针对无人机空中自主对接时，软式锥套受尾流场和牵引力强位置约束、锥套传递力等因素导致拖曳锥套运动复杂而存在扰动、鞭打等干扰的问题，提出一种基于机器视觉的单目视觉自主实时识别与定位方法，通过特征提取与分类器分类进行目标检测与识别，分阶段对无人机和靶标关键点进行实时相对位置测量。在机载探测与计算平台实验进行系统应用验证，完成了机动锥套自主对接地面实验，结果表明目标检测精度可达90%，平均响应帧率可达41.322 Hz，定位误差可达厘米量级，证明了所提的软式机动锥套空中自主对接方法的有效性。 |
| 关键词: 无人机 自主对接 目标识别 机器学习 |
| DOI：10.20079/j.issn.1001-893x.230321001 |
|
| 基金项目:国家自然科学基金资助项目（U2141248） |
|
| Target Recognition and Positioning of UAV Autonomous Docking Process Combined with Machine Learning |
| YAO Yao,LI Junfu,HU Zhiyong,AI Junqiang |
| (AVIC The First Aircraft Institute,Xi’an 710089,China) |
| Abstract: |
| The soft cone sleeve is constrained by the wake field and strong traction position,and the cone sleeve transfer force and other disturbances during the unmanned aerial vehicle(UAV) autonomous docking in the air,which lead to the complex movement of the towed cone sleeve and the existence of disturbances,whips and other disturbances.In order to solve these problems,a monocular vision based autonomous real-time recognition and positioning method is proposed.This method detects and recognizes targets through feature extraction and classifier classification.The real-time relative position measurement of UAV and target key points is conducted in stages.Through the system application verification on the airborne detection and calculation platform experiment,the ground experiment on the autonomous docking of the mobile cone sleeve is carried out.The results show that the target detection accuracy can reach 90%,the average response frame rate can reach 41.322 Hz,and the positioning error can reach the centimeter level,which proves the effectiveness of the proposed method of autonomous docking of the flexible mobile cone sleeve in the air. |
| Key words: UAV autonomous docking target recognition and positioning machine learning |
