Research On Localization And Mapping Of Inspection Robot For Parking Based On Visual-IMU Fusion

In order to maintain the damaged vehicles and maintain the normal state of vehicles and equipment,the army forces will implement the depot day system,through the implementation of the Parking Day System,to achieve the combat preparation task.In view of the problems of "dealing with errands,insufficient attention,improper methods and low quality" in the current Parking Day System,this paper proposes to use unmanned and intelligent inspection robot to check the implementation of the Parking Day System,the first task is that the robot can achieve accurate positioning.In this paper,a Vision-IMU Fusion Localization and Mapping algorithm based on sliding window and key frame is designed.The research work is carried out in the following three aspects:(1)In this paper,a fast corner distribution algorithm based on region threshold in view of the weak texture and repeated texture environment in the Parking.Firstly,the fast corner threshold is calculated based on the global information of the image,and the image pyramid is constructed to provide scale invariance.The image is divided into blocks,and the corner points are processed by quadtree to improve the distribution of image features.The lower the evaluation standard,the higher the degree of uniformity.The experimental results show that in public data EUROC,MH＿01 、V1＿01 sequence and DEST were reduced.(2)In this paper,a fusion system of vision and IMU based on sliding window and key frame is proposed in view of the application of inspection robot in the parking environment to improve the accuracy and stability of positioning.Design a joint initialization scheme of vision and IMU to realize system self-calibration and improve stability;The key frame filtering strategy is improved to filter the repeated and redundant image frames to ensure the stable and accurate operation of the system in the yard environment.The experimental results show that on the EUROC public data set,The trajectory estimated by the proposed system is superior to the classical fusion scheme VINS in Mean,Median,Min,RMSE and STD.For MH＿04 sequence,The mean value error,median value error,root mean square error and standard deviation were reduced.(3)In this paper,according to the engineering requirements of inspection robot,combined with the cost,performance and other considerations,an experimental platform is built by using the Turnlebot 2 robot,Indemind binocular camera module and notebook.The Sensor,Slam and Robot Control System are integrated to form a complete inspection robot system.When testing the real environment,V120: Trio motion capture system is used to collect the True Value.For data set 1 and 2,the error is reduced.