Research On Tracked Vehicle Localization Based On Visual-Inertial SLAM

With the rapid development of artificial intelligence technology,agriculture driverless vehicles are widely researched at home and abroad.Unmanned driving technology is one of the important development directions of intelligent agricultural machinery.Simultaneous localization and map building(SLAM)technology is the base support for the industry of driverless,which is a technology to realize localization and map building in an unknown environment by sensors carried on mobile robots,and its focus is on dynamically updating maps and acquiring location information of mobile robots in the environmental.A localization approach using a stereo camera and IMU fusion to take full advantage of the complementary strengths between the stereo camera and IMU.This paper is based on an improved VINS-Fusion framework,which is studied as follows:Firstly,this paper fully analyzes the current research hotspots by using Citespace software,and visualizes the literature distribution,countries,research institutions,journals,and research hotspots respectively,and concludes that the current research directions of interest to researchers include sensor fusion,SLAM in dynamic scenes,semantic SLAM and visual SLAM,and introduces the technology development in the field of SLAM.Secondly,this paper introduces the binocular camera model and IMU error model,respectively,and calibrates stereo,IMU,and stereo + IMU using the ROS calibrating tool and Kalibr calibrating tool,respectively,and applies the calibrated parameters to the subsequent calculations.In the visual-inertial odometry,the visual-inertial odometry framework,the corner point extraction algorithm with faster extraction speed,the visual inertial joint initialization,and the back-end nonlinear optimization are presented respectively.In the front-end odometry,this paper combines the sub-pixel algorithm for corner point extractions,at the same while,the experimental comparison with the FAST algorithm,Harris algorithm,and Shi-Tomasi algorithm,by slightly higher time-consuming in exchange for the improvement in corner point extraction accuracy.In the back-end optimization,the back-end marginalization operation is optimized by dividing the camera pose and the state variables other than the camera pose into two threads for the marginalization operation,which aims to improve the calculation speed of marginalization.Finally,the algorithm of this paper is compared with VINS-Fusion in Ubuntu 20.04 system using the KITTI dataset,Eu Ro C dataset,and selfpicked farmland dataset,and the two schemes were evaluated using the evo tool,and the experimental results show that the error of the algorithm of this paper is lower than that of the VINS-Fusion algorithm in all three datasets.