Strategies for large scale elastic and semantic LiDAR reconstruction

This paper presents novel strategies for spawning and fusing submaps within an elastic dense 3D reconstruction system. The proposed system uses spatial understanding of the scanned environment to control memory usage growth by overlapping in different ways. allows number consumption scale with size rather than duration exploration. By analysing overlap semantic information, our segments distinct spaces on-the-fly during exploration, such as rooms, stairwells, indoor–outdoor transitions. associates semantically labelled poses SLAM pose graph enable global elasticity. A probabilistic model merge voxel labels is incorporated ensure correct submap fusion when loop closures occur. Additionally, we present a new mathematical formulation relative uncertainty between improve consistency reconstruction. Performance demonstrated using experiments exploring multi-floor multi-room indoor environments, transitions large-scale outdoor experiments. Relative baseline, presented approach demonstrates improved scalability accuracy. • Strategies are based on understanding. presented, which guides fusion. LiDAR annotated fly information fused consistency. integration core accounts motion distortion. Multi-floor, multi-room, validate approach.