Research On Simultaneous Localization And Mapping System Of Mobile Robot Based On Multi-sensor Fusion

Driven by social development and technological progress,intelligent mobile robots are gradually being used in families,hospitals,factories,nursing homes and other places to provide convenience for human beings.Localization and mapping are the key technical problems for autonomous movement and operation of mobile robots.With the diversification of sensors and the continuous improvement of sensor technology,the simultaneous localization and mapping technology of robots with multi-sensor fusion has become a research hotspot.However,the existing multi-sensor fusion technology still has the problem of insufficient accuracy or poor real-time performance,and there are many challenges in the stability,accuracy and real-time performance of robots in complex outdoor large-scale environments.Therefore,this thesis deeply studies the simultaneous localization and mapping technology of multi-sensor fusion guided by improving the stability,accuracy and real-time performance of mobile robot localization and mapping,and a multi-sensor fusion method is proposed that takes into account high precision and high real-time performance,and realizes the stable localization and mapping of robots in complex outdoor large-scale environments.The main research contents and results are as follows:(1)In order to ensure the effective fusion of sensor data and the stable running of the experiment.First,the pinhole camera model,distortion model,stereo camera model,accelerometer and gyroscope working principle and the measurement model of the lidar are analyzed.Then,the internal parameters of the monocular camera,the stereo camera,the IMU and the lidar are calibrated,the external parameters of the camera and IMU,and the camera and lidar are jointly calibrated.Finally,the effectiveness of the calibration results are verified through experiments.(2)In order to improve the accuracy of the IMU preprocessing process,the IMU pre-integration model and the IMU initialization model in the IMU preprocessing process are studied,the IMU pre-integration process based on fourth-order Runge-Kutta numerical integration is designed and the propagation process of error noise at great length is analyzed.Through comparative analysis with the IMU pre-integration method of median integration,its effectiveness is verified;IMU initialization model based on the maximum a posteriori probability estimation is constructed as an inertial optimization factor and optimized using a factor graph method.Comparative experimental results demonstrate that the proposed IMU preprocessing method has higher localization accuracy.(3)In order to ensure that the visual-inertial system has good real-time performance while ensuring accuracy,the fusion method of vision and IMU is studied.A new fusion optimization method based on multi-constrained state Kalman filtering and factor graph optimization is proposed.The keyframe insert mechanism is proposed,multi-state constraint Kalman filter is used to fuse vision and IMU data at the ordinary frame between two keyframes,and the pose information of each frame is propagated and updated,the factor graph approach based on a sliding window to optimize keyframe poses at the keyframe level is used.Experimental results demonstrate that the proposed method has both high real-time and localization accuracy.(4)The localization and mapping method based on the fusion of vision,IMU and lidar is constructed.The association method and failure detection mechanism of the visual-inertial system and the lidar-inertial system are set up to realize the coupled working mode of the two subsystems,and localization and mapping is realized stably in large-scale outdoor environments.Experimental results demonstrate that the constructed system has high localization and mapping accuracy and stability in large-scale outdoor environments.