Research On Multi Sensor Fusion And Localization Based On Kalman Filter

In the driverless vehicle,positioning is one of its core technologies.Through the analysis of the unmanned driving technology framework,we can see that perception,decision-making and control are the three key factors.The sensing layer is a collection of electronic map and sensor information.The electronic map is composed of traditional navigation and various map data,and the map data includes navigation electronic map data and high-precision electronic map data.The decision-making layer mainly completes path planning,environment understanding,vehicle behavior prediction and other functions.Instead of the driver,the control layer controls the vehicle through electronic drive,so as to gradually realize unmanned driving.Accurate positioning is the basis and core of driverless.Without accurate positioning,driverless vehicles may make various mistakes.According to the positioning process of driverless vehicles,three technologies are usually used.The first is signal positioning technology,such as 5G,GNSS and UWB.The second is dead reckoning technology,which is a process of information accumulation by adding displacement vector to the initial position to calculate the current position.For example,the Inertial Measurement Unit(IMU)technology calculates the current position and direction of the vehicle after knowing the position and direction of the vehicle.The third is the environment feature matching technology.For example,using lidar and visual sensor,the calculation of current position is a process of matching with known data,so as to accurately know the position of unmanned vehicles in the environment.As one of the core technologies of unmanned driving,the development of the technology has experienced several iterations.First,it is Global Navigation Satellite System(GNSS),which is based on satellite positioning,which can provide positioning capability with an accuracy of about 10 meters.The second generation is inertial navigation and positioning technology.The third generation is high-precision positioning based on lidar,millimeter wave radar and vision sensors,which can provide about sub meter level to centimeter level positioning capability.The main contents of this paper are as follows:(1)according to the ICP algorithm in PCL library,the ICP strategy is rewritten and renamed ICP_DYL algorithm is verified on Kitti data set through specific experiments,and the experimental results show that ICP is effective_The maximum error of DYL odometer segment trajectory is 6.2m,and the average error is 1.9m;the average error of each segment trajectory is about 1.9m,but the error is 6.2m in 25 to 30 segments.ICP_The maximum trajectory error of DYL odometer is 60.6m,and the average error is 26.5m;at the beginning of the trajectory,the error increases slowly,but decreases after 2000.It can be seen that the accuracy of segmented statistics from high to low is:NDT>ICP_For the overall trajectory error,the accuracy from high to low is:ICP_DYL>NDT>ICP。(2)In the framework of error state based filtering(eskf),the fusion localization based on point cloud map is realized.(3)Various filtering models of integrated navigation are discussed in detail,including GPS+IMU filtering model,GPS+IMU+odometer filtering model and GPS+IMU+magnetometer filtering model.The observability and observability of different motion states such as uniform speed,static,acceleration and deceleration,and steering are analyzed.Using simulation software GNSS_INS_SIM,the corresponding motion state data is generated and Kalman filtering is performed.The convergence accuracy and convergence rate of each state variable in Kalman filter are calculated and compared with the results of observability analysis.