Research On SLAM Algorithm Based On Fusion Of Binocular Vision And IMU

Simultaneous Localization and Mapping(SLAM)generally refers to the process that the carrier uses its own sensors to estimate the pose and get its own trajectory.It is an indispensable key technology in the field of mobile robot research.At present,the vision slam technology with camera as sensor is the focus of research.However,when the controlled object moves fast or the illumination of the environment changes greatly,the SLAM algorithm relying only on visual information is easy to lose the motion state of the target,and it is difficult to achieve accurate positioning.Therefore,this paper improves the current mainstream monocular vision inertial navigation slam system,and designs a pose estimation and optimization algorithm based on information fusion of binocular vision and inertial measurement unit(IMU).The purpose is to improve the accuracy of indoor positioning.The main contents of this project include the following parts(1)Firstly,the mathematical tools and basic concepts related to visual slam are introduced.By establishing motion equation and observation equation,slam is defined as the problem of state estimation from the perspective of probability.The mathematical model of sensor in visual inertial odometry(vio)is studied.The theoretical basis involved includes:camera model and distortion processing,IMU measurement model and pre integration theory,binocular stereo vision ranging,nonlinear optimization method and so on.(2)Based on the framework of vins,the whole flow of the algorithm is designed into four modules: information preprocessing,system initialization,back-end nonlinear optimization,loop detection and relocation.In the preprocessing stage,the orb feature points in the image are extracted and matched,the depth information is calculated by binocular stereo vision,and the IMU measurement data is pre integrated.After the system initialization process,the visual inertia information is provided to the back-end.In the back-end nonlinear optimization process,the vision and inertial measurement information are tightly coupled.According to the constraint relationship among vision,IMU and landmark,the objective function and incremental equation to be optimized are solved.In order to reduce the information redundancy,a key frame screening mechanism and an edge strategy based on the idea of sliding window are proposed.Firstly,the temporal consistency of candidate keyframes is checked,then the geometric consistency is checked.The problem of tracking loss is solved by relocation and closed-loop optimization to improve the positioning accuracy.(3)The experimental design of the proposed binocular vision inertial SLAM algorithm verifies the effectiveness of the method.Through the comparative experiment on the public data set and the test on the embedded development platform,the results show that the improved binocular vision inertial slam system can meet the requirements of real-time positioning,has high accuracy,and can realize the trajectory tracking of mobile robot in indoor environment.