Monocular visual odometry with road probability distribution factor for lane-level vehicle localization

摘要

Towards achieving lane-level localization, precision and accuracy plays an important role in vehicle localization efficiency. While Global Positioning System (GPS) is usually used for localization, it has low accuracy caused by signal degradation due to several reasons such as lack of well-positioned satellites, signal obstruction or multipath error. Thus, multi-sensor data fusion has been widely studied to improve vehicle localization. By utilizing the existing techniques for monocular visual odometry and particle filter localization, this paper presents how road information available in OpenStreetMap contributes to accurate and precise vehicle localization by exploiting road probability distribution factor in particle filter implementation. This approach was verified in two datasets with different road features and it has shown better performance compared with the established particle filter localization. As our results indicate, this approach is feasible for lane-level localization for intelligent vehicles.