计算机工程与应用 ›› 2023, Vol. 59 ›› Issue (12): 301-308.DOI: 10.3778/j.issn.1002-8331.2203-0336

• 工程与应用 • 上一篇    下一篇

基于无人机视觉的风机桨叶叶尖定位方法

焦嵩鸣,白健鹏,首云锋,陈雨溪,钟宇飞   

  1. 华北电力大学 自动化系,河北 保定 071003
  • 出版日期:2023-06-15 发布日期:2023-06-15

Positioning Method of Wind Turbine Blade Tips Based on UAV Vision

JIAO Songming, BAI Jianpeng, SHOU Yunfeng, CHEN Yuxi, ZHONG Yufei   

  1. Department of Automation, North China Electric Power University, Baoding, Hebei 071003, China
  • Online:2023-06-15 Published:2023-06-15

摘要: 应用无人机对风力发电机进行自主巡检时,需对其桨叶叶尖做出精准定位。为此提出一种基于无人机视觉的叶尖定位方法。无人机利用风机粗略偏航信息到达指定位置后,并利用Yolov5模型对风机桨叶、轮毂等部位进行特征提取;根据提取到的特征点计算单应性矩阵并将其分解,得到无人机偏移量;基于混合式视觉伺服设计修正器对无人机位姿进行修正,从而使无人机正对风机并进行全景图采集,同时对其进行了稳定性分析;利用全景图以及无人机的位姿对桨叶叶尖进行三维空间定位。仿真实验表明,设计的控制器可以使无人机在15?s内自主完成正对风机操作并对其进行全景图采集,提出的方法可对风机桨叶叶尖进行精准定位,定位误差仅在±5?m内。

关键词: 风力发电机, 无人机, Yolov5, 视觉伺服, 单应性矩阵, 叶尖定位

Abstract: When applying UAVs for autonomous inspection of wind turbines, the blade tips need to be accurately located. A method for locating the blade tips based on UAV vision is proposed. Firstly, the UAV uses the rough yaw information of the wind turbine to reach the designated position, and uses the Yolov5 model to extract features of the wind turbine blades, hubs and other parts. Secondly, the homography matrix is calculated and decomposed according to the extracted feature points to obtain the UAV offset. Then, a hybrid vision servo-based design corrector is used to correct the position and attitude of UAV, so that the UAV is squared up to the wind turbine and a panoramic image is acquired, while its stability is analyzed. Finally, the blade tips are positioned in 3D space using the panorama and the position and attitude of the UAV. Simulation tests show that the designed controller can enable the UAV to autonomously complete the operation of squaring up to wind turbine and collect a panoramic view within 15 seconds. The proposed method can accurately locate the blade tip of the wind turbine with a positioning error of only within ±5 meters.

Key words: wind turbine, UAV, Yolov5, visual servo, homography matrix, blade tip positioning