Design And Implementation Of Autonomous Driving Localization System

With the rapid development of artificial intelligence technology and the encouragement of relevant government policies at all levels,the autonomous driving industry has become the most promising artificial intelligence application industry in the current era.In this context,this paper designs and implements an automatic driving positioning system based on multi-sensor fusion based on actual needs.The purpose of this work is to solve the problem of accurate positioning in the process of automatic driving applications.Firstly,the characteristics of the system are analyzed from the perspectives of three different users: automatic driving positioning engineers,map engineers and planning control engineers.The functional requirements of the system are established from the analysis.Then,the four-layer technical architecture of the system,namely the application layer,algorithm layer,access layer and hardware layer,is established.Finally,the author designs and implements the four functional modules of the system.(1)Online localization module: This is the core module of the system,which is responsible for preprocessing the optical image camera,IMU(Inertial Measurement Unit,inertial measurement unit)and wheel speedometer sensor raw data.The collected raw position data is used for nonlinear graph optimization based on sliding window to complete the preliminary positioning of the autonomous vehicle.(2)Mapping module: This module uses the vehicle’s positioning information and environmental image information given by the online positioning module to build a semantic three-dimensional model of the surrounding environment of the vehicle.The semantic map established by this module contains semantic elements such as lane lines,road indication arrows,and pillars,which are helpful for the safe driving of autonomous vehicles.In terms of map representation,this module uses a vectorization method to represent the dense point cloud in the map,which can reduce the storage space of the map.(3)Re-localization module: This module can perform more precise positioning in the existing map.By using word vectors to represent images,it can precisely search for vehicle positions in existing map databases,and also can optimize the pose of autonomous vehicles and observations of the surrounding environment.(4)Interaction module: This module can display the operation results of the automatic driving positioning system,and can provide a practical map ranging function.The system uses Open LORIS-Scene,a positioning test data set in complex environments released by the Tsinghua research team in 2020,and conducts functional and non-functional tests.The experimental results show that the system can effectively and robustly complete the automatic driving positioning task with little computing resources.