Positioning Method And Application Of Substation Inspection Robot Based On Fusion Of Vision And Inertial Navigation Information

With the rapid development of multi-sensor information fusion technology and mobile robot technology,research on substation inspection robots is becoming more and more mature.Compared with traditional manual inspections,unmanned and intelligent robot inspection method has gradually highlighted its immeasurable advantages.Substation inspection robots need to be able to accurately and stably locate the robots when performing instrument monitoring and fault inspection tasks.This thesis focuses on the stable positioning of inspection robots under certain lighting changes,and develops research work on the positioning of inspection robots in substations that combines vision and inertial navigation information.The specific work contents are as follows:Aiming at the harsh and dangerous inspection environment of the substation,An inspection robot hardware platform that could run stably in the substation environment is designed and developed in this thesis.The four-wheel active differential motion model of the inspection robot is analyzed and verified in detail.In order to solve the problem of large errors in wheel odometer results caused by the structural parameters of the robot,a method for estimating the parameters of the robot combined with the visual odometer is proposed to realize fast and accurate estimation of the structural parameters of the robot.Aiming at the problem that the corner extraction in the process of visual pose estimation is susceptible to changes in light,an image edge extraction corner extraction method is proposed.The edge detection is performed on the grayscale image in advance,and the corner points are further extracted on the edged image to ensure the stability of the extracted corner points under a certain lighting change.In order to solve the problem of repeated integration in the pose process of IMU(Inertial Measurement Unit)sensors,an IMU pre-integration algorithm is proposed to reduce the computational complexity.Finally,in order to ensure the stable and accurate positioning of the inspection robot under a certain light changing environment,The EPnP(Efficient Perspective-n-Point)algorithm is proposed to estimate the pose of the binocular vision information and achieve the positioning of the inspection robot in this thesis.In addition,IMU information constraints are added to the pose optimization objective function to enhance the stability of the positioning system when fast rotation or lack of visual information.It is further proposed to use loop detection and global pose optimization to improve the consistency and smoothness of global pose estimation.Finally,the effectiveness of the proposed positioning method is verified by experiments.First,the method is verified on the EuRoc open source dataset and compared with the VINS-Fusion system and stereo positioning system.The experimental results show that the method in this thesis is superior to the compared positioning systems in terms of positioning accuracy while ensuring the real-time performance of the system.Further,the positioning experiments are performed in an indoor environment with a reflective scene and an outdoor inspection environment with different daytime lighting changes to simulate the lighting changes in the substation environment.The experimental results show that the positioning system in this thesis can still perform stable and reliable positioning under the conditions of light changes and reflection effects in the experimental environment.