Pole-based Localization for Autonomous Vehicles in Urban Scenarios Using Local Grid Map-based Method

Self-localization is a key component of autonomous vehicles in urban scenarios. In this work, we proposed a localization system which is based on pole-like objects such as trees and street lamps. Pole-like objects are extracted from 3D LiDAR point cloud using a cluster-based method. Based on the pole detection results, we propose a new map representation which consists of numerous local grid maps. In order to tackle the data association problem caused by the ambiguity of pole-like landmarks, the detected poles are directly transformed to the local grid map to define a cost function without pole-to-pole matching. The subsequent non-linear optimization method is utilized to minimize the cost function and generate the vehicle pose. We evaluate our localization system on our self-collected dataset. And the proposed system achieves a root mean square error of less than 18 cm for position and less than 0.52 ° for yaw.