Towards Efficient and Robust LiDAR-based 3D Mapping in Urban Environments

For autonomous driving, 3D mapping is an important task for accurate and robust navigation. This paper presents an efficient method for 3D mapping in urban scenarios. To improve the mapping robustness, a degeneracy-aware factor graph is constructed, which considers the degradation of scan matching constraints and prior observations. To alleviate the side effects caused by dynamic vehicles, an efficient likelihood-field-model-based multi-object detection and tracking algorithm is applied to filter dynamic objects. Extensive tests in real-world datasets show that our approach works well in dynamic urban environments.