Research On UAV’s Auto-tracking Method For Power Line Based On Visual-inertial Navigation System

Unmanned aerial vehicle(UAV)is a popular aerial research platform and has important application in industrial inspection.In the field of power line patrol,due to the adverse conditions in the real environment,the UAV can’t identify the power line stably and autonomously.The main research problem of this paper is to enable the UAV to accurately estimate its own position and pose under adverse environmental conditions and conduct stable autonomous power line tracking.IMU(Inertial measurement unit)is independent of visual information.It can perform pose estimation under severe environmental conditions.However,IMU is susceptible to the influence of noise and bias accumulation.The pose estimation results are not reliable enough.In this paper,the visual information and inertial information are tightly coupled and nonlinear optimized.The advantages of the two sensors are combined to improve the autonomous line-tracking performance of the UAV in the actual environment.A UAV’s auto-tracking method for power line based on visual-inertial navigation system is proposed,which is a tightly coupled nonlinear optimization method for data fusion of visual information and inertial information.It includes front-end initialization method,back-end nonlinear optimization method and sliding window marginalization method to reduce the amount of computation.The paper mainly introduces from the following aspects:This paper first studied and deduced the mathematical model of UAV and camera.The collecting and processing methods of visual information and inertial information are also studied.In visual information,we studied the Harris corner detecting method and KLT sparse optical flow tracking method.Then A pre experiment is designed to verify the performance of Harris corner point detecting method with the power line dataset.In inertial information,the pre-integration processing is proposed.Then,this paper designed the visual-inertial navigation system.We introduce the method of initialization in the front end of the system: first the pure visual Sf M and then the alignment and calibration of the visual information and inertial information.In the back end,we introduced the nonlinear optimization method,the tightly coupled nonlinear optimization based on Bundle adjustment.We studied schur complement method and designed a marginalization sliding window based on schur complement method to reduce system computation.Finally,the ROS robot operating system is implemented in this project.The Eu Ro C UAV dataset is used to test the performance of the system in this paper and other existing systems.The Evo evaluation program is used to evaluate the performance of the system based on the trajectory information collected from the performance test.The advantages and disadvantages of the system are evaluated and summarized after comparing with other existing algorithms.Based on the loose-coupled nonlinear optimization of the visual information and inertial information,a tightly coupled nonlinear optimization method for the visual information and inertial information is designed in this paper.The experiment proved that the UAV can complete the state estimation of itself in the place with severe environmental conditions.From the experimental data,we can conclude that the system reduces the amount of the computation and realizes real-time estimation.Finally,through four kinds of evaluation indicators to analyze the system’s stability of auto-tracking,we proved that our algorithm can improve the stability of the power line tracking.