Research On SLAM Localization Method Based On Multi-Sensor Fusion For Unmanned Vehicles

The visual-inertial SLAM method has been widely used in the positioning of robots and UAVs,but when it is applied to the vehicle platform,the IMU will lack acceleration excitation,which seriously affects the positioning accuracy.In order to solve the problems such as inaccurate initialization scale estimation due to vehicle plane motion,the positioning accuracy of unmanned vehicles is improved.Aiming at the unmanned vehicle platform,this paper makes full use of its sensor equipment and adopts the method of multi-sensor fusion to carry out research on the positioning of unmanned vehicles,and proposes a multi-sensor fusion SLAM algorithm using wheel speed sensors,monocular cameras,and inertial measurement units.The main research contents of this paper are as follows:(1)This paper summarizes the research status and development trends of SLAM technology at home and abroad,analyzes the problems existing in the application of visual-inertial SLAM to vehicle platforms,and expounds the significance of visual-inertial SLAM research with wheel speed fusion.(2)The theory and optimization method of SLAM positioning are studied.For the SLAM system,the relevant basic theories are analyzed,the motion of objects in three-dimensional space is analyzed through four expressions,the modeling process of the SLAM problem is studied,and the related optimization methods are analyzed for the solution process.(3)A visual inertial odometer fusing wheel speed was proposed.The sensors used in the system are analyzed,and the camera model,IMU measurement model and wheel odometer measurement model are established.Based on the sensor model,the corresponding error sources are analyzed in detail,and the calibration method of the internal and external participation error coefficient is studied,and the calibration experiment is carried out.Aiming at the problems existing in the current application of visual-inertial SLAM on vehicle platforms,a visual-inertial odometer fusing wheel speed was proposed.For the front-end of WVIO,the strong corner points are extracted by the Shi-Tomasi algorithm,and the image pyramid LK optical flow is used for feature tracking.The wheel speed-imu pre-integration is constructed based on the wheel speed and IMU data,and the error state equation in discrete time is deduced through the median integration.Using the methods of visual initialization and joint initialization,the initialization of scale factor and speed is completed through the wheel speed-imu pre-integration constraint.For the back-end optimization part of the system,each component of the objective function is analyzed,and the sliding window is used to perform nonlinear optimization of wheel speed,vision,and inertia to solve the optimal state estimation.(4)A multi-sensor data collection and verification platform for unmanned vehicles has been built.An unmanned vehicle sensor platform is built based on Seres SF5 and sensor equipment,and the data collection of cameras,millimeter wave radar,wheel speed sensors,IMU and other sensors is realized through ROS.The KAIST dataset experiment and the real car experiment are carried out,and the accuracy of the proposed algorithm is verified.One set of datasets shows that when the car travels 690 meters,the average error is 0.986 meters,and the RMSE value is 1.153 meters.Verified by many experiments,the algorithm proposed in this paper has more accurate location results than VINS-Mono.