A Novel Real-time Semantic-Assisted Lidar Odometry and Mapping System

Recently, rich semantic information has proven to be an enabling factor for a wide variety of applications in mobile robots. In this paper, we explore the integration of semantics into lidar odometry and mapping approaches and present a novel real-time semantic-assisted system. To this end, a sparse 3D-CNN model is designed to perform per-frame semantic segmentation of lidar points. Transformations are then estimated by jointly minimizing the geometric and semantic distances between correspondences. At last, new points are transformed into the world coordinate system and used to update predicted labels in the global semantic map. Experiments show that our system has a better performance in pose error compared with the geometry-based method.