Study On Intelligent Vehicle Positioning Technology Based On Multi-sensor Information Fusion

The high-precision and robust vehicle positioning information can provide information support for the perception,decision-making and path planning modules of intelligent vehicles,which is the basis for the realization of driverless.Generally,a single sensor can realize the vehicle positioning function,but in the actual driving process in complex environment,it is easy to be affected by the characteristics of the sensor itself and environmental changes and other factors,which makes it difficult to rely on a single sensor to complete the continuous positioning work.Aiming at the problem of low accuracy and poor robustness of vehicle positioning in complex environment,from the point of view of low cost,this paper proposes a monocular and stereo vision fusion scheme based on Kalman filter and a multi-sensor information fusion and positioning scheme based on nonlinear optimization to achieve the long-term,high-precision and strong robust vehicle positioning function.The main contents of this paper include:First of all,a method based on Kalman filter for the fusion of forward and backward cameras is proposed.The fusion of forward and backward visual odometry is realized by loose coupling.This method uses the environment information of multi camera vision to realize a wider range of perception,so as to achieve more accurate positioning.At the same time,for the failure or lack of sensor information,the positioning method can automatically judge and switch the bypass failure module,which improves the robustness of the system.Aiming at the lack of scale information of monocular camera,a real-time online estimation method based on stereo 3D information to recover the scale of monocular visual odometry is proposed.This method estimates the optimal scale information by minimizing the error constraints in the estimation window,and corrects the estimation scale under the influence of the vehicle’s motion.In addition,the estimated scale can recover the actual pose information of the monocular visual odometry,so that the front and rear cameras have the ability of independent positioning.In order to further improve the positioning performance,this paper introduces multi-dimensional information such as IMU and GPS based on the fusion of front and rear cameras,and proposes an intelligent vehicle positioning method based on nonlinear graph optimization technology.According to the characteristics of different sensors,the fusion method is implemented by local fusion and global fusion bilayer architecture.In the local fusion layer,IMU and camera are composed of VIO system to realize pose estimation.Simultaneous interpreting and optimizing the global pose of each VIO factor and the GPS factor with global information are implemented in the global fusion layer.This method has strong robustness and flexibility.When any kind of sensor fails,the system can still work normally.In the hierarchical fusion strategy,the multi-level and multi spatial information of each sensor can be complemented and combined to improve the positioning performance of the system.Finally,the proposed method is tested and verified on the relevant intelligent vehicle dataset,and compared with other advanced positioning methods.The results show that its positioning accuracy and robustness are improved,and it can provide continuous and reliable positioning information for intelligent vehicles,at the same time,it has the characteristics of low cost.