Research On Localization Algorithm Of Mobile Robot Based On Vision And Inertial Fusion

With the development of science and technology,robot technology has become the focus of scientific research,and robot precise localization is one of the preconditions to realize other robot technology,which has important theoretical significance and application value.This thesis studies the status quo of mobile robot localization technology at home and abroad.On this basis,binocular cameras and IMU(Inertial Measurement Unit,IMU)are selected as the main sensors of the localization system,and a system scheme of vision and IMU fusion localization is proposed.Firstly,this thesis research on the visual localization system.For ORB(Oriented FAST and Rotated BRIEF,ORB)feature points are densely packed and overlapped phenomenon,this thesis adopts the Non-Maximum Suppression method,which calculates the Harris corner response value of the FAST(Features from Accelerated Segment Test,FAST)corner,and only keeps the corner with the largest response value in a specific area.Because the wrong feature matching affects the localization accuracy,in order to improve the localization accuracy of the system,this thesis proposes an improved RANSAC(Random Sample Consensus,RANSAC)algorithm,which uses the cosine similarity algorithm to coarsely screen the matched feature points,reduces the number of iterations of the RANSAC algorithm,and the Directional Epipolar Geometric Constraint is added to the RANSAC algorithm to verify the basic matrix model,to reduce the time consumption caused by unnecessary internal and external point screening,and improving the localization accuracy and real-time performance of the localization system.Secondly,this thesis research on the vision and IMU fusion localization system.In order to reduce the amount of calculation in the back-end optimization,this thesis adopts the sliding window optimization method.Aiming at the cumulative error in the process of optimizing the motion trajectory of the sliding window,this thesis proposes a non-linear factor to restore the sparse distribution of the window marginal prior,and calculates the Kullback-Leibler Divergence of the sparse distribution of the set of factor structure graphs and the prior distribution.Optimize Kullback-Leibler Divergence together with IMU error and visual reprojection error.The experimental results show that the method of increasing the nonlinear factor to restore the marginal prior is effective improve the localization accuracy.Finally,a mobile robot localization system is constructed based on the improved localization algorithm in this thesis.The software design are completed in the Ubuntu18.04 operating system with ROS Melodic,and the construction of the experimental platform are completed.Using the built robot experiment platform conduct experiments in the actual environment.The experimental results show that the mobile robot localization system studied in this thesis is feasible and reliable.