Localization System Modeling And Application For Unmanned Driving

Localization System Modeling and Application for Unmanned DrivingAs the main means to achieve vehicle safety,comfort and economy,driverless technology has become an inevitable trend.Whether adopting vehicle networking scheme based on V2X(Vehicle to Everything)with position information sharing as the core,or a single intelligent body scheme based on multi-sensor combination,vehicle’s precise position is the precondition of intelligent vehicle implementation.As one of the core sensors in the sensing layer,positioning system plays an extremely important role and is an important fundamental issue in the research of unmanned vehicle.At present,there are many problems in the positioning system,such as low sampling frequency,drift of positioning result,failure of satellite signal and interference of driving environment.In order to solve such problems,it is necessary to conduct comprehensive and detailed modeling of the positioning system to reflect the defects,error forms,laws,generation mechanism and environmental correlation of the vehicle positioning system.The model of unmanned positioning system,especially the sensor model with high confidence,is an important component of the simulation process besides vehicle dynamics,high-fidelity scene,driver behavior and traffic flow model.The essence of positioning system modeling is to model a variety of sensors that constitute the positioning system.The research in this paper mainly focuses on the modeling and application of satellite positioning system,inertial measurement unit,camera and other sensors.Due to the complexity of the physical world,differences in hardware schemes and different ways of data fusion,the actual modeling process is limited by practical factors and cannot build a perfect model.In order to reduce the error with the actual positioning system,improve the model confidence,and practical engineering application oriented,this paper established three types of common positioning process-related sensor models,and based on them for the combined positioning system application simulation verification and algorithm testing.Specific research contents are as follows:1)This paper introduces the features and implementation principles of the current mainstream positioning scheme,and separately describes the data fusion technology related to the positioning system.The essence of vehicle positioning technology is clarified and the framework of typical vehicle positioning system is analyzed.For common vehicle positioning methods,the characteristics and implementation principles of positioning technologies realized by satellite positioning,dead reckoning,inertial navigation and visual perception means are analyzed and introduced,and the fusion method in combined positioning state estimation is deduced and explained.2)The GPS physical model and IMU mathematical model were established and their confidence was preliminarily verified.Based on the two independent sensor models,the simulation application of GPS/IMU combined positioning system was conducted.Specifically,according to the error source of the satellite positioning system,the actual positioning mechanism is simulated,and a process model including ephemeris,occlusion,pseudo-distance error and other process factors is established.In addition,by studying the characteristics and randomness of IMU data errors,a mathematical model considering systematic errors and random errors is established.Through system parameter identification,actual simulation and application process,sensor model output conforms to the actual application requirements and can reflect the real data characteristics.3)The vehicle monocular pinhole camera model was established,and a multi-dimensional camera model confidence evaluation method was proposed.This paper introduces the realization principle of a vision-based odometer.The camera model is mainly based on the projection model which describes the imaging process of the camera,and on this basis,imaging error models such as blur,distortion and halo are added.The model confidence evaluation is mainly based on the goodness of fit index of pixel level and feature level between the simulation image and the real image.4)A positioning system framework including visual multi-sensor fusion was proposed to comprehensively simulate and verify the positioning system sensor model,and the positioning system algorithm was tested and analyzed based on the open source data set.Based on the robot operating system platform,the simulation and experimental system distributed the vehicle integrated simulation platform,Simulink model and positioning algorithm,and achieved good simulation results.The effectiveness of the positioning system sensor model established in this paper is verified by theoretical analysis,model establishment and simulation application in the combined positioning system.Through the comprehensive simulation application and real data set test comparison,it reflects the advantages and potentials of the positioning system sensor model simulation in terms of cost,efficiency,safety and parameter controllability.