Abstract
KEYWORDS – Precise Vehicle Localization, Lane Markings Matching, Stereo Camera, Digital Map, Autonomous Vehicle
This work, which is part of Hitachi’s autonomous driving project, aims to develop a robust and inexpensive localization system with lane level accuracy on highway. The main idea is to improve both lateral and longitudinal localization accuracy by aligning lane markings from visual sensors and a precise digital map. First, a forward view stereo camera and 4 surround view monocular cameras are applied to detect lane markings. Then, the currently and previously observed lane markings are utilized to build a local lane markings map based on the vehicle’s trajectory. Next, the local lane markings map is matched with the digital map to estimate the vehicle’s global position. Finally, a Kalman filter model was applied to handle the matching error, GPS localization error, and vehicle motion constrains. Validation experiments were carried out on a test bed autonomous driving car at one of our company’s test site. Experimental results indicate that both lateral and longitudinal localization accuracy are improved significantly, resulting in a solid foundation for autonomous driving on highway.