无人机高电压输电线路障碍物检测方法

Obstacle detection method for unmanned aerial vehicles on high-voltage transmission lines

针对电力巡检无人机在高电压输电线路遇到的各种障碍物及其特点,该文提出了一种基于视觉TF-Net障碍物检测算法。该算法通过巡检无人机上的双目摄像头获取图像,并利用障碍物的几何形状和图像特征确定障碍物类型,结合摄像头和障碍物之间的位置关系计算出障碍物到无人机的距离,以判断是否会影响无人机的正常飞行。研究采用了局部敏感提取、Ghost-Group卷积单元等方法,并对神经网络进行了轻量化处理,以最大化降低功耗。实验结果表明,相对于传统的卷积算法,该算法FPS提升了10 frames/s,平均精度(mAP)提升了6%。该算法在检测精度和检测速度上较普通算法框架都有一定提升。

Aiming at the various obstacles and their characteristics encountered by power inspection UAV in high voltage transmission lines,an obstacle detection algorithm based on visual TF-Net is proposed.The algorithm obtains images by inspecting the binocular camera on the drone,and determines the type of obstacle by using the geometric shape and image features of the obstacle.The distance from the obstacle to the drone is calculated by combining the positional relationship between the camera and the obstacle to determine whether it will affect the normal flight of the drone.Local sensitive extraction,Ghost-Group convolution unit and other methods are used in the study,and the neural network is lightweighted to minimize power consumption.After multiple training with different training sets,the distance of obstacles can be estimated.The experimental results show that compared with the traditional convolution algorithm,the FPS of the algorithm is increased by 10 frames/s,and the Average Precision(mAP)is increased by 6%.The algorithm has a certain improvement in detection accuracy and detection speed compared with the ordinary algorithm framework.