Research On SLAM Technology For Substation Inspection Robots Based On Vision And IMU Fusion

With the rapid development of China’s power grid,the power transmission and distribution have becoming more complex,and traditional manual substations and equipment inspections face increasing pressure and danger.Therefore,introducing robotics technology into power system inspection and replacing manual inspection is crucial.Simultaneous Localization and Mapping(SLAM)is a key technology of substation inspection robots,which has developed rapidly in the past decade,moving towards higher mapping quality,faster speed,higher positioning accuracy,and more complex environmental applications.Therefore,this project aims to design a substation inspection robot SLAM technology that combines Inertial Measurement Unit(IMU)information with vision.Main research work includes the following:1.Introduce of basic theory of vision and IMU information fusion,including the rigid body transformation in 3D space,the working principle of camera and IMU and their corresponding coordinate relations,and to establish the corresponding mathematical model.2.Research on SLAM systems based on depth cameras is carried out.For the complex problem of substation scenes,linear features are introduced and parallel lines in the scene are selected to intersect at the vanishing point in the image species for online calibration of the depth camera.In the visual tracking thread,features extraction is first performed,then image features are tracked and depth camera pose is iteratively calculated according to the sparse optical flow algorithm.In the back-end thread,map points are optimized using beam leveling,and to avoid cumulative errors,a bag-ofwords model is used to describe key frames in loopback detection.Calibration experiments are carried out on the D435 i camera and the experimental algorithm shows that the error is controlled to within ±1.9°3.Propose a visual SLAM scheme that fuses IMU information,derive IMU preintegration and error transfer matrix and state update matrix,and solve the problem of inconsistent IMU and camera frequencies.At the same time,all parameters of the two sensors are jointly calibrated to pave the way for tight coupling of the two sensors.Finally,the least-squares estimation function of marginalization information,IMU residual term and visual residual term is constructed,and the Levenberg-Marquardt(LM)is usd to solve the least squares function,and the sliding window marginalization is used to improve the real-time performance,finally realizing the SLAM system of substation inspection robot built by tightly coupling the depth camera and IMU.4.Evaluation of the algorithm proposed in this paper,the SLAM localization experiments with vision and IMU fusion are carried out with the Eu Ro C datasets and compared with pure vision systems,VINS-Mono and VI-ORB systems.The experimental results show the ability to improve system performance compared to other algorithms.A vision and IMU fusion mobile robot platform is built to verify the practicality and effectiveness of the algorithm in real-life scenarios as well.