Design And Implementation Of Multi-scene Combined Localization Algorithm

In recent years,with the explosive development of information industry and Internet applications,new automotive businesses such as autonomous driving,online car-hailing and car sharing have emerged one after another,and automotive intelligence has also become a development trend.So how to continuously and accurately perform vehicle positioning in multiple scenarios in cities has become a research hotspot.The traditional vehicle positioning method based on Global Navigation Satellite System(GNSS)is only suitable for outdoor open scene.Moreover,the positioning accuracy is not high(about meter level),so it is generally impossible to accurately distinguish the primary and secondary roads or multi-lane information to which the vehicle belongs.Facing the common urban interchanges and underground parking lots,there will be serious deviation or even impossible positioning.Therefore,in order to continuously and accurately locate vehicle positions in multiple scenes,this thesis designs a multi-scene combination positioning algorithm.Typical urban vehicle scenes recognition is based on features of collected satellite signals and other radio signals.A multi-source combined positioning algorithm with dynamic parameters is designed.According to the above-mentioned recognition results,different parameters and positioning information sources are dynamically used for combined positioning.And continuous and consistent high-precision positioning results can be obtained in typical urban vehicle-mounted multi-scene situations.Firstly,this thesis proposes a dynamic parameter combination localization algorithm based on Extended Kalman Filter(EKF)to solve the problem that the traditional vehicle positioning scene is single and cannot be accurately positioned in the actual city multiple scenes.For several typical scenarios,different positioning methods and parameters are switched dynamically.For open outdoor scenes such as main and side roads or multi-lane scenes,Real-Time Kinematic(RTK)positioning is mainly used with high-precision.Inertial navigation is integrate using EKF algorithm and the EKF measurement noise matrix is filled with RTK position error covariance.Fusion inertial navigation can improve the positioning output frequency and real-time positioning while achieving lane level positioning accuracy.Inertial navigation dead reckoning is mainly used for semi-closed urban overpasses and urban canyons surrounded by high-rise buildings.Combined with the poor satellite positioning results at this time,the noise matrix of EKF measurement is filled with RTK positioning error and appropriate error coefficient parameters of semi-occluded scene,so that the positioning accuracy can still be maintained even if the short-term satellite positioning is poor.For long tunnels and underground parking lots,there is no satellite signals for long time.The self-built Ultra Wide Band(UWB)base station was used for positioning using UWB technology.Inertial navigation was integrated to improve stability and the measurement noise matrix of EKF was filled with UWB positioning error parameters fitted by measurement.Therefore,various typical vehicle-mounted platforms can be positioned accurately,continuously and at high frequency.And it greatly expands the positioning scene of traditional vehicle positioning with GNSS.Then,aiming at the continuity of vehicle positioning between different scenes,a scene recognition algorithm based on the features of radio signals(satellite signals,ultra-wideband signals,etc.)is proposed.The current scene is identified by analyzing the radio signal characteristics collected by the sensor.And the radio signals mainly include satellite positioning signals and UWB signals.The characteristics of satellite signals mainly include the number of effective common-view satellites,the average signal-to-noise ratio(SNR)and Q value of RTK positioning state recognition;while the UWB signals mainly judge the existence of identifiable base station signals and SNR strength.When switching between the open multi-lane scene and semi-closed inter-city scene,the average SNR and number of satellites usually fluctuate abnormally.This thesis designs an improved standard score outlier detection algorithm,which eliminates the influence of outliers and improves the robustness and recognition accuracy of the algorithm in practical application.On the one hand,the next value is predicted by using the mean and variance information of the historical data of satellite number and average SNR.The abnormal change is judged if it exceeds the threshold,and the abnormal points are smoothed by different influencing factors.On the other hand,through actual observation,the average SNR and number of satellites are set with extreme abnormal ignored values respectively.If they exceed,they will not be included in the calculation of the next mean variance of historical data,which makes the recognition results more in line with the actual scene changes reflected by satellite signal fluctuations.Finally,based on the positioning algorithm proposed above,a multi-scene combined positioning system is designed and implemented.The system hardware uses the NEO-M8T satellite positioning module of U-Blox to collect the original satellite signals,the OPENIMU300 series module of Aceinna to collect the inertial navigation signals,and the DW1000 module of DecaWave for UWB positioning.The system software was implemented using the open source RTKLIB and module driver APIs provided by Aceinna and DecaWave.Moreover,the system code of C language is developed in Linux system,and has been tested in the actual vehicle environment.A large number of test results show that the algorithm system can identify three kinds of vehicle-mounted scenes stably.Then,by setting the dynamic parameters,continuous and accurate positioning can be carried out in the multi-scene,and the accuracy can reach decimeter level,thus achieving the design goal of the algorithm.