The article introduces SNI-SLAM, a Semantic Neural Implicit SLAM system that uses neural implicit representation for semantic mapping, surface reconstruction, and robust camera tracking. The system includes a hierarchical semantic representation for multi-level semantic comprehension, enabling top-down structured semantic mapping of the scene. It integrates appearance, geometry, and semantic features for feature collaboration, allowing a multifaceted understanding of the environment. The authors demonstrate that SNI-SLAM outperforms other NeRF-based SLAM methods in terms of mapping and tracking accuracy, while also providing accurate semantic segmentation and real-time semantic mapping.
Publication date: 18 Nov 2023
Project Page: https://arxiv.org/abs/2311.11016
Paper: https://arxiv.org/pdf/2311.11016
