Review and Comparison of Vision- and LiDAR-based Methods for Autonomous Vehicle Localization
A robust and accurate vehicle localization is a crucial element in the field of autonomous vehicles. It is needed for trajectory and path planning, tracking, and trajectory prediction. Normally this is achieved by relying on the global navigation satellite systems, but it is not always reliable. Here vehicles sensing modalities step in and take over the task of localization. This can be achieved by detecting features in the surroundings of the vehicle and localizing the vehicle either relatively or globally. Provided, an accurate map exists consisting of expected features, it is possible to match features extracted with sensing modalities to the map and thus the vehicle is able to make sense of the surrounding environment. Another possibility is to localize the vehicle relatively, for example, lane-level localization. Usually, a camera is used as the sensing modality. It is a relatively affordable, simple, and popular sensor. But in recent decades the Light Detection and Ranging (LiDAR) is offering heavy competition to cameras because it has produced more accurate results and can be used to extract more useful information from the environment. The aim of this work is to offer a review and comparison of vehicle localization using cameras and LiDARs as sensing modalities. In most cases, map-based localization is brought out, but in some cases also relative localization is discussed. The focus of the review is on the sensors, how they are used and what are their advantages or disadvantages. The categorization in the comparison chapter is based on features with which the localization is achieved – these include lane markings, road marks, curb detection, traffic signs and landmarks. In the end, the author concludes the work with suggestions for choosing the right sensor for localization of autonomous vehicle in desired operating conditions, based on reviewed papers.
Graduation Thesis language
Graduation Thesis type
Bachelor - Computer Science
Debasis Kumar, Naveed Muhammad