Research On Intelligent Visual Measurement Technology Of Distribution Line Maintenance Robot

With the rapid development of robot technology,many special robots have been successfully developed,which can replace people to complete tasks with high risk factors and complex operations.This paper focuses on the tasks of maintenance robots in power distribution lines,and studies the target pose estimation,hand-eye calibration,and the calibration method of the coordinate system of the double manipulator base.The thesis mainly completes the following work:(1)The overall structure of the distribution line maintenance robot is given by analyzing the working environment and tasks of maintaining the distribution line.Based on the demand for the vision system in the maintenance work of the power distribution line,a hand-eye cooperative operation scheme is designed and a hardware system composition scheme is given.(2)Aiming at the problem of hand-eye coordination system calibration for distribution line maintenance robots,Based on Eye-in-Hand,a calibration method of dual robotic arm ’ s base coordinate system is proposed.By analyzing the conversion relationship between the hand-eye cooperative coordinate system,the hand-eye relationship equation and the dual-arm arm base coordinate relationship equation are established.Tsai two-step method and Zhang’s calibration method are used to complete the binocular camera calibration,hand-eye relationship calibration,and dual robot arm base relationship calibration.The reverse verification of the calibration results shows that the calibration accuracy can meet the needs of hand-eye cooperative operation.(3)A method of cross arm position measurement based on a fully convolutional neural network is proposed,because of the problem that the maintenance robot of the power distribution line cannot accurately grab the arrester due to the tilt of the cross arm during the operation of the arrester replacement.The pose model is established by analyzing the 3D model of the cross arm.A segmentation method based on the VGG-19 full convolutional neural network is used to extract the cross arm region from the image,and the segmentation result is used as a mask to perform edge detection on the original image to effectively eliminate environmental noise.Based on the voting method and Huff space constraint,the straight-line extraction of the cross arm’s edges is carried out,and the method of calculating cross arm’s pose is given.This method can accurately measure the pose of the cross arm,and guide the end of the robotic arm to adjust the pose to grab the arrester correctly.(4)Aiming at the problem that the distribution line maintenance robot is not able to accurately grab the clamp suspended on the power line due to the environmental impact during the lead wire operation,a fast clamp pose tracking method based on kernel correlation filtering and 3D model is designed.The 3D model of the clamp is modeled by CAD software and rendered from different distances and angles.The database containing the clamp’s geometric edge information is trained.The extracted edges in the image are matched with the database to obtain the pose of the clamp.In order to shorten the pose matching time,a kernel correlation filtering method is used to track the position of the clamp in the image,and the target’s bounding box area is used as a mask to perform edge extraction and pose matching.This method can quickly and accurately match the pose of the clamp.