Robust Visual Localization System With HD Map Based on Joint Probabilistic Data Association

Localization based on a high-definition (HD) map is a pivotal technology for autonomous driving. Nonetheless, establishing precise data association (DA) between detected landmarks and map landmarks presents a formidable challenge when leveraging prior information on maps. Traditional DA algorithms relying on nearest-neighbor methods only partially mitigate the ambiguity in DA caused by missed or false detections from the perception module, especially in complex and challenging environments. In this letter, we propose a novel joint probability data association (JPDA) algorithm. By integrating joint probability encompassing semantic likelihood, local spatial likelihood, and global structural likelihood of landmarks, alongside incorporating inter-frame temporal continuity of DA, the proposed algorithm can effectively rectify the erroneous DA. Additionally, we also introduce a max-mixture factor graph optimization framework, which couples the measurements of landmarks and odometry for pose estimation. Building upon these methods, a high-precision and robust visual semantic localization system employing consumer-level sensors has been developed. Experiments conducted on public datasets and real urban roads validate the efficacy of the proposed system in providing more robust and accurate localization results for autonomous driving vehicles.