In-Lane Localization in Road Networks Using Curbs Detected in Omnidirectional Height Images

For autonomous driving on predefined GPS routes, the use of differential GPS alone is not sufficient. Autonomous cars must perceive the street in order to drive safely. In this paper, we propose to estimate in-lane positioning offsets with regard to a high-level road network model by sensing street curbs with a 64-beam laser-range-finder. The laser measurements are transformed into height images represented in sensor-intrinsic yaw and pitch coordinates. In this dense image-like representation, curbs can be detected with simple oriented edge filters. The curb measurements are matched to expected curbs derived from the road network model. This scan-matching procedure yields measurements of offsets between road model and detected road, which are filtered to obtain a correction of the in-lane position.