Pole-Based Real-Time Localization for Autonomous Driving in Congested Urban Scenarios

Abstract

Real-time and robust pose estimation is required by autonomous driving in dynamic urban environment. However, many point cloud based localization methods consume large storage space and computing resource. What's worse, in congested urban scenarios, dynamic objects like vehicles and pedestrians cause serious occlusion, which raises difficulties in map building and leads to wrong map-matching results. This paper proposes a localization approach bases on pole-like feature like tree trunks, telegraph poles and street lamps in urban environment. The feature-based method greatly reduces the amount of map data, increases real-time performance and improves robustness against dynamic objects. Localization experiments have been carried out on a very challenging urban road, and the results showed our proposed method is real-time and robust in congested urban environment.