자율주행을 위한 계층적 정보융합기반 정밀 위치추정 및 지도생성 연구

Abstract

Mapping represents a technique to build up a map contained many types of feature, such as road geometry, traffic environments, and landmarks. Localization means a process for determining position of a vehicle on the map generated by the mapping technique. The processed results of localization and mapping should be accurate, reliable, and continuous in order to apply for autonomous driving system. If the localization and mapping algorithm offers inaccurate or invalid information to the autonomous driving system, autonomous cars can result in fatal accidents such as road departures or collisions with obstacles or other vehicles. This dissertation presents localization and mapping algorithms that satisfy the requirements of autonomous driving system (accuracy, reliability, and continuity). There are various types of information sources for localization and mapping of autonomous cars, including a satellites-based global positioning system (GPS), vehicle motion sensors, vehicle dynamic models, digital maps, and perception systems. Each information source can be used for localization and mapping of conventional automotive applications; however, it is not suitable to apply for the autonomous driving application because stand-alone information source is difficult to fulfill the accuracy, reliability, and continuity requirements of autonomous car’s localization and mapping system. To satisfy the requirements of autonomous cars, information fusion of the various types of sensors and information sources is inevitable. The integration of several information can improve the accuracy, reliability, and continuity of localization and mapping algorithm. There are many types of combination of information source for the information fusion: (GPS + Vehicle motion), (GPS + Map), (Vehicle motion + Map), (GPS + Vehicle Motion + Map), and many others. Also, there are various types of information fusion algorithms to integrate the information sources, including Bayesian filtering, Fuzzy logic, Neural networks and many others. The selection of information sources and fusion methods is determined by the purpose and performance of the localization and mapping. This dissertation proposed information fusion algorithm based on a hierarchical structure in order to improve the scalability, reusability, and maintainability of the localization and mapping algorithm. The hierarchical information fusion structure consists of three types of information fusion component that have different purposes and performances. First information fusion component integrates the GPS and vehicle motion information for vehicle state estimation and positioning of autonomous cars. Second fusion component combines the GPS ,vehicle motion and perception information in order to generate the precise digital map for autonomous driving system. Final fusion component integrates the GPS, vehicle motion, perception, and map information for the precise localization of autonomous car. Each information fusion algorithm are evaluated and verified through intensive experimental studies.