【作者】 陈冠华；

【导师】 许华荣； 王宇；

【作者基本信息】 厦门理工学院 ， 电气工程（专业学位）， 2021， 硕士

【摘要】 随着无人机技术的迅速发展,无人机的稳定性、续航能力和载重能力得到了较大的提升,利用无人机进行电力巡检成为了研究热点之一。本文针对四旋翼无人机在电力线巡检过程的自主导航和绝缘子故障检测问题,研究无人机基于双目视觉的障碍物检测与避让、自主着陆技术以及提高绝缘子掉串故障准确率的方法。在无人机避障方面,利用GPS或北斗导航控制无人机自主飞行的途中,通过双目相机检测飞行路径上的障碍物,利用障碍物特征点匹配视差图获得无人机与障碍物的距离;以最近距离的障碍物边缘检测点为圆心,在无人机飞行的水平平面或与飞行方向平行的竖直平面上,构建包围障碍物并预留安全距离的碰撞圆;控制无人机沿着机体中心点到碰撞圆的外切线飞行,实现无人机顺利避开障碍物的飞行策略。在无人机自主着陆方面,通过利用双目相机重建地面的三维点云信息,根据深度以及与地面的夹角信息,首先拟合出较为平坦的平面,转化为二维信息;其次对可着陆地面的特征进行学习,获得可着陆的平面特性;最后将二维信息输入随机森林分类器中,判断是否为可着陆的平面。通过无人机在不同高度对地面的重建操作并稳定着陆的实验,证明了该方法能够有效判断着陆地面,实现无人机的自主着陆功能。在提高绝缘子掉串故障准确率方面,本文以Retina Net网络为基础,结合自适应训练样本选择和泛化的焦点损失函数算法加强有效正样本选择。然后采用平衡特征金字塔结构对特征提取进行加强,将精度更高的样本选择算法与平衡金字塔特征增强结构结合到Retina Net网络中。实验证明,改进的Retina Net网络能够有效提升绝缘子掉串故障的检测准确率。本文通过对无人机自主导航技术以及在电力线巡检过程中对绝缘子故障检测技术的研究,达到利用无人机实现自动电力线巡线的目的,降低电力线巡检的工作风险,提高其工作效率。更多还原

【Abstract】 With the rapid development of UAV technology,the stability,endurance and load capacity of UAVs have been greatly improved.So,the use of UAVs for power inspection has become one of the research hotspots.Aiming at the autonomous navigation and insulator fault detection of quadrotor UAVs during the power line inspection,this paper studies vision-based obstacle detection and avoidance,autonomous landing,and methods to improve the accuracy of insulator string failures.In terms of UAV obstacle avoidance,GPS or Beidou navigation is used to control the UAV’s autonomous flight,and the obstacles on the flight path are detected by binocular cameras.The disparity map is generated through the matched feature point of obstacles,so as to obtain the distance between the UAV and the obstacle.Taking the edge detection point closest to the obstacle by the UAV as the center,construct a collision circle that surrounds the obstacle and reserves a safe distance on the horizontal plane where the UAV is flying or the vertical plane parallel to the flying direction.This flight strategy controls the UAV to fly along the outer tangent line from itself to the collision circle to realize the obstacles avoidance of the UAV.In terms of autonomous landing of UAVs,this paper uses binocular cameras to reconstruct the three-dimensional point cloud information on the ground.According to the depth and the angle between the UAV and the ground,a relatively flat plane is first fitted and converted into two-dimensional information.Secondly,learn features according to the characteristics of the landable ground and obtain the landable plane characteristics.Finally,input the twodimensional information into the trained random forest classifier to determine whether it is a landable plane.Experiments prove that the UAV can judge the landability of the ground at different heights and land stably.This proves that the method can effectively judge the landing ground and realize the autonomous landing function of the UAV.In terms of improving the accuracy of insulator string drop failures,this paper uses the Retina Net with adaptive training sample selection and generalized focus loss function to strengthen effective positive sample selection.Then use the balanced feature pyramid to enhance the feature extraction.The more accurate sample selection algorithm and the balanced pyramid feature pyramid are combined into the Retina Net.Experiments show that the improved Retina Net can effectively improve the detection accuracy of the insulator drop failures.In this paper,through the research on the autonomous navigation technology of UAV and the insulator fault detection technology in the inspection process,the UAV is used to realize the automatic power line inspection,which reduces the work risk of power line inspection and improves work efficiency.更多还原