Leveraging Object Recognition In Reliable Vehicle Localization From Monocular Images

We present the processing pipeline of a monocular vision system that successfully performs the task of detecting, identifying and localizing a city bus electric charger station. This task is essential to the operation of an advanced driver assistance system that helps the driver to dock the long vehicle at the charging station. The focus is on the role of machine learning techniques in developing a robust detection and classification procedure that allows our system to localize the camera with respect to the charger even from long distances. We demonstrate that the learned detection procedure improves robustness of the vision techniques for monocular localization, while the geometric relations estimated by our system can be used to improve the learning results.