Design And Implementation Of Vision-based Power Inspection Robot Navigation System

For large-scale hydropower stations,due to their large area for inspections,less demand for environmental maintenance,long cycle times and fixed patrol routes,the hydropower stations waste the human resources involved in inspection tasks.Based on the current situation of robot research,this paper puts forward a inspection robot navigation system framework with vision.Using this system,the inspection robot can follow the prescribed route for self-inspection and obstacle avoidance in the work environment,transmit the images and data to the remote monitoring system,monitor the operation of the equipment in the work area in time,and effectively eliminate potential safety hazards.This topic mainly studies and improves the inspection robot navigation system in the following aspects:1.After introducing the working principle of the depth vision sensor,the basic concepts of vision-based location,navigation and mapping is introduced,and the front-end and back-end parts of the RGB-D SLAM system is described.In order to avoid the problem of mis-matching in the image matching step,the cross-matching method based on the k nearest neighbor algorithm is used to reduce the number of mis-matching points.In the motion estimation and optimization part,RANSAC and GICP algorithms are used to avoid the the problem of the computation process falling into the local minimum and other issues.2.For the back-end part of the visual navigation system,the topic adopts the concept of minimum visual changes to improve the selection criteria for key frames,and formulates the criteria for the determination of redundant key frames.Thus,the problem of large mount of computation and high memory occupancy is improved.3.According to the actual situation of large-scale hydraulic power stations,the topic improves the global path planning method.For the original location coordinate information,redundant coordinate normalization operations are performed on the coordinate points,and an available mesh map is constructed.And on the basis of the mesh map,the optimal route is solved by adopting the improved route planning algorithm that combines the necessary points and three routes,so that the planned route can meet the user's needs.4.Vision-based close-range planning algorithm mainly includes four aspects:using depth image information to detect obstacles,filtering out the fake obstacles in the detected obstacles,adopting certain obstacle avoidance strategies,and returning to the original route after the inspection.