LEVERAGING DYNAMIC OBJECTS FOR RELATIVE LOCALIZATION CORRECTION IN A CONNECTED AUTONOMOUS VEHICLE NETWORK

Abstract. High-accurate localization is crucial for the safety and reliability of autonomous driving, especially information fusion collective perception that aims to further improve road by sharing in a communication network Connected Autonomous Vehicles (CAV). In this scenario, small errors can impose additional difficulty on fusing from different CAVs. paper, we propose RANSAC-based (RANdom SAmple Consensus) method correct relative between two CAVs order ease among Different previous LiDARbased algorithms only take static environmental into consideration, also leverages dynamic objects thanks real-time data Specifically, addition like poles, fences, facades, object centers detected vehicles are used as keypoints matching point sets. The experiments synthetic dataset COMAP show proposed greatly decrease error less than 20cm far there enough poles correctly both Besides, our highly efficient runtime be scenarios driving.