Research On Positioning Method Of Mobile Robot Based On Vision

With the progress of artificial intelligence,mobile robot technology has been increasingly used in the field of life and industry.Its outstanding performance makes it an indispensable part of many fields.Positioning problem is the most important and the basic problem in the field of mobile robots.Positioning refers to the estimation of the position and attitude of the mobile robot.At present,the vision-based mobile robot positioning method is developing rapidly.The positioning method combining the monocular vision and the Inertial Measurement Unit(IMU)can effectively solve the scale uncertainty problem of the monocular and cumulative drift problem of the inertial navigation,which enhances the robustness of the system and meets the high-precision positioning requirements.Therefore,this paper will use the method of monocular vision and inertia fusion to study the mobile robot localization algorithm.Firstly,The article outlines the research background and status based on visual fusion positioning method.The theoretical basis of mono-VINS is also introduced,including the imaging model of the camera,the coordinate transformation relationship during the positioning process,and the basic visual positioning mapping process.Then,for the pose estimation problem of mobile robots,a positioning algorithm of monocular fusion IMU is proposed.The VO uses the feature extraction of Harris corner points and the feature tracking of KLT optical flow method.The front end of inertial unit uses pre-integration method to process the data.After the initialization process,the pose information of the mobile robot is estimated by the non-linear optimization process based on the sliding window at the back end.The effectiveness of the algorithm is verified by experiments.But the shortcomings of the cumulative drift of the algorithm has been found.Aiming at the shortcomings of the fusion location algorithm,an improved algorithm based on loopback detection relocation and global pose optimization is proposed.The pose of the optimized window is relocated using the pose informationof the loopback frame,and the pose estimation of the globally is obtained through the optimization process.Experiment with the improved algorithm to verify the feasibility of the improved algorithm.Then,the accuracy analysis of the non-loop detection positioning algorithm and the improved positioning algorithm with loop detection is carried out.The two positioning algorithms are compared by absolute pose error(APE)and relative pose error(RPE),and the improved algorithm is obtained globally.The pose is estimated to be more accurate.Finally,the inertial camera sensor is used to test the visual fusion algorithm to verify the versatility and portability of the algorithm.