Research On Localization Method Of Indoor Unmanned Vehicle Based On Visual-inertial Odometry

When using the unmanned vehicle to work in the unknown indoor environment,it is the basis and guarantee for the follow-up research to obtain its accurate position and attitude in real time.Simultaneous Location and Mapping(SLAM)technology has been considered by many scholars as the key research direction to realize autonomous mobile unmanned vehicle is in uniform motion or pure rotation state in an indoor environment,the monocular vision inertial odometer becomes unobservable due to lack of acceleration excitation,resulting in poor positioning accuracy.The fusion of wheel odometer can solve the above problem caused by motion mode,the localization accuracy and robustness of the system under abnormal conditions can be refined.Based on the classical framework of visual-inertial odometry system,this paper analysis it,and a camera-IMU-wheel odometry fusion localization algorithm which employs nonlinear optimization method is proposed.The certain research contents are as follows:In order to solve the initialization instability of monocular VINS due to the lack of motion excitation,a positioning method based on vision inertial odometer is proposed.The paper studies and analyzes the pre-integration and error propagation process of IMU and wheel odometer.In the initialization module,the loose coupling method is used to fuse the measured values of wheel odometer.By proposing an effective method to monitor the quality of external parameters,the calculation burden of recalibrating external parameters due to fixed time is reduced under the condition of ensuring accuracy,so as to improve the stability of system initialization.In order to solve the influence of indoor weak texture environment on positioning accuracy,this paper uses hierarchical clustering method based on RANSAC algorithm to quickly and real-time extract plane geometry information in the environment.The ground plane is constructed based on the plane geometry detection method,and the influence on the positioning accuracy caused by the accumulated error is reduced by adding the ground constraint.A back-end optimization model based on the position and attitude map optimization method is constructed to further improve the positioning accuracy of the unmanned vehicle in the indoor weak texture and light environment.Finally,an experimental system is built to do comparative experiments under the scene of the public dataset and the actual indoor environment which contains weak texture.The algorithm proposed in this paper has improved the localization accuracy and robustness compared with monocular visual-inertial odometry and traditional wheel odometer algorithm.