Research On Multi-source Sensor Fusion Positioning Technology Based On Long Short-term Memory Neural Network And Unscented Kalman Filter

Navigation and positioning is one of the foundations of intelligent driving technology research.The precise position of the intelligent driving vehicle can be obtained through the navigation and positioning technology,so as to better realize the intelligent driving functions such as path planning and behavior decision-making.The Global Positioning System(GPS)is widely used in the navigation and positioning technology of intelligent driving vehicles due to its simple structure and low cost of use.However,the positioning accuracy and robustness of the civil GPS are insufficient.The accuracy of the pose calculated by the Inertial Measurement Unit(IMU)is high,and the sensitivity to the environment is low,but the long-term dead reckoning error of the IMU is relatively large.In addition,Lidar has high ranging accuracy and long scanning distance,which cannot perform global positioning.Therefore,this paper comprehensively researches the localization methods of the above sensors,and presents a multi-source sensor fusion localization method based on Long Short-Term Memory neural network and Unscented Kalman Filter.The main research contents are as follows:Research on GPS and IMU fusion positioning method based on Long Short-Term Memory neural network.In order to improve the positioning robustness of the global positioning system when the signal is occluded,the neural network algorithm and the filtered IMU pose are studied.And a fusion positioning method based on Long Short-Term Memory neural network is presented.When the GPS positioning signal is normal,the GPS positioning difference is trained through the filtered IMU pose data.Once the GPS is blocked and cannot be positioned,the trained Long Short-Term Memory neural network model and the filtered IMU pose predicts the GPS positioning.The method realizes the fusion positioning of GPS and IMU.Finally,the tensors of the Long Short-Term Memory neural network model are trained through the real vehicle data collected by the roadside,and it is verified that the method can effectively solve the robustness when the GPS positioning is failure.Research on the fusion positioning method of GPS/IMU and Lidar based on Unscented Kalman Filter.In order to improve the positioning accuracy of the navigation and positioning system,the error source of Lidar in dynamic environment is firstly analyzed,and a dynamic error correction method for Lidar data is presented.Then,the matching and localization algorithm of Lidar based on Normal Distribution Transformation is studied.Then,the kinematics model of the vehicle is built according to the kinematics of the vehicle,and the fusion positioning model based on the Unscented Kalman Filter algorithm is established according to the vehicle kinematics model.This method estimates the optimal trajectory of the GPS/IMU and Lidar fusion positioning model.Finally,the optimal estimated position can be obtained by this method,which verified through the simulation experiment.The accuracy of the navigation and positioning system is improved.Research and build a multi-source sensor fusion positioning system and a real vehicle experimental platform.Firstly,the overall framework of the fusion positioning system is designed,and based on the framework,the corresponding sensors are selected to build a real vehicle experimental platform.Then the process of the real vehicle experiment is studied and analyzed,and the relevant real vehicle experiments are designed according to the above methods.On a straight road,the fusion localization method based on Unscented Kalman Filter designed in this paper is verified,The experimental results of the real vehicle show that the fusion positioning trajectory is smooth and can effectively improve the positioning accuracy of the intelligent driving vehicle.On the closed-loop curved road,the method of partially cutting off the GPS positioning signal is used to simulate the GPS failure.The multi-source sensor fusion positioning model is verified in real vehicles,which is based on Long Short-Term Memory neural network and Unscented Kalman Filter designed in this paper.The experimental results of the real vehicle show that before and after the GPS positioning signal is cut off,the error of the fusion positioning system remains basically unchanged.The output trajectory is smooth,and the positioning effect is significantly improved compared with the sensor.It is verified that the method in this paper can effectively improve the positioning robustness and positioning accuracy while ensuring the smooth running of the intelligent driving vehicle.