Lane-level Vehicle Self-localization by Integrating Inertial Sensors and Stereo Camera for Under-bridge Scenario

摘要

Vehicle self-localization plays an important role in autonomous driving. Sophisticated methods have been proposed for positioning in open sky field and even urban scenario. However, there is few discussion for a specific but essential case, driving under the viaduct, which is named as under-bridge scenario in this paper. The challenge of the localization in this environment includes insufficient number of satellites and repeated scene, which is possibly difficult for GNSS and sensing. It is important to local vehicle in lane level and estimate the distance to the next intersection in driving event. This paper presents a self-localizing method with inertial sensor and stereo camera. Stereo camera is used in drivable space detection and road mark detection. From index of the lane, lateral position of vehicle is provided. Stop-line detection and inertial sensors provide the longitudinal position. By knowing lateral and longitudinal distance, the vehicle position can be estimated. In this research, the experiment was conducted in an under-bridge area in Tokyo with the distance of 2km. The experiment results show that high accurate self-localization is achieved.