Research On Simultaneous Localization And Mapping Technology Of Intelligent Vehicle Based On Multi-source Information Fusion

Accurate localization and environment perception are essential for intelligent vehicles in autonomous driving mode,and simultaneous localization and mapping technology(SLAM)is one of the key technologies to solve this problem.SLAM technology based on monocular vision sensors and inertial measurement units(IMU)has become a hot research direction because of low sensor cost,easy installation and the potential to complement each other’s advantages.However,there are still some problems when it is applied to intelligent vehicles moving on the ground:when environment features are missing,the vision sensors approximate a failure state,degrading the system to rely on the dead reckoning of IMU,and errors will grow rapidly;When the vehicle is in uniform circular motion or uniform linear motion,the scale observability of the visual inertial SLAM system will change.In addition,the initialization of the system provides a good initial for the nonlinear optimization module which will also have a large impact on the accuracy of the entire localization and mapping results.During the starting phase of the vehicle,the acceleration is usually small,so it is also a challenge to initialize the system robustly under inadequate incentives;To solve the above problems,combined with the characteristics of the vehicle movement,a new sensor data based on the vehicle kinematic model measurement information is introduced to improve the performance of the system in terms of initialization robustness,scale observability and the accuracyof localization and mapping.The specific research and results of this paper are as follows:1.A framework of intelligent vehicle SLAM algorithm based on multi-source information fusion is proposed,and monocular camera-IMU-vehicle kinematic model is used as the observation input.To improve the adaptability to external features,in the vision front-end,point-line features are extracted for camera motion matching estimation;IMU and vehicle motion-based measurements of front wheel angle and longitudinal velocity are pre-integrated to construct constraints between the adjacent key frames,and considering the planar motion of the vehicle,motion constraints based on the SE(2)plane are introduced.Finally,a combination of loop-closure detection and a sliding window algorithm is introduced to optimize the estimation of the vehicle’s own positioning state and environmental information.2.A modeling and error online calibration method considering the error parameters of vehicle kinematic model is proposed.In this paper,the effects of vehicle velocity measurement deviation,front wheel rotation angle measurement deviation,front wheel rotation angle bias,and vehicle wheelbase bias on the model accuracy are considered in the vehicle kinematic model.At the same time,the kinematic error state propagation is used as the prediction and the GNSS-INS position,velocity and angular velocity measurement is used as the observation,a model error online estimation method based on the error-state Kalman filter is constructed,which can estimate the above model parameters during the vehicle motion,and the vehicle odometer based on the model of this paper is proved to have higher accuracy in the experiment.3.For the case of gyroscope and accelerometer bias estimation scattering and scale information difficulty convergence under the inadequate incentives condition,the vehicle odometry information is aligned with IMU-vision and the scale factor,velocity,bias and gravity of the system are robustly initialized.At the same time,the multi-sensor involves coordinate transformations between multiple coordinate systems,i.e.,external parameters,and the external parameters between the vehicle coordinate system and IMU coordinate system are also considered as variables to be estimated online to improve the overall system performance.Finally,the performance of the proposed algorithm is verified under different vehicle driving conditions through real-world experiments,and compared with the current mainstream SLAM framework.The results show that the intelligent vehicle SLAM algorithm based on multi-source information fusion designed in this study can meet the vehicle positioning accuracy and mapping requirements.