WildGS-SLAM: Monocular Gaussian Splatting SLAM in Dynamic Environments

2025-04-04

AI Summary

To address tracking drift, map distortion, and rendering artifacts in monocular RGB SLAM caused by dynamic objects in dynamic environments, this paper proposes the first monocular Gaussian splatting SLAM system tailored for dynamic scenes. Our method introduces three key innovations: (1) an uncertainty-aware geometric map enabling robust dynamic object removal guided by per-vertex uncertainty; (2) integration of uncertainty prediction into dense bundle adjustment (BA) and Gaussian map optimization to enhance dynamic reconstruction accuracy; and (3) a pixel-wise uncertainty estimation module combining shallow MLPs with DINOv2 features, driving uncertainty-weighted Gaussian splatting mapping from monocular RGB input. Evaluated on multiple dynamic datasets, our approach achieves artifact-free novel-view synthesis, outperforming state-of-the-art monocular SLAM and neural rendering methods in both tracking and reconstruction accuracy.

Technology Category

• Computer Vision: Motion & Tracking
• Reasoning under Uncertainty: Uncertainty Representations
• Intelligent Robots: State Estimation

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We present WildGS-SLAM, a robust and efficient monocular RGB SLAM system designed to handle dynamic environments by leveraging uncertainty-aware geometric mapping. Unlike traditional SLAM systems, which assume static scenes, our approach integrates depth and uncertainty information to enhance tracking, mapping, and rendering performance in the presence of moving objects. We introduce an uncertainty map, predicted by a shallow multi-layer perceptron and DINOv2 features, to guide dynamic object removal during both tracking and mapping. This uncertainty map enhances dense bundle adjustment and Gaussian map optimization, improving reconstruction accuracy. Our system is evaluated on multiple datasets and demonstrates artifact-free view synthesis. Results showcase WildGS-SLAM's superior performance in dynamic environments compared to state-of-the-art methods.

Problem

• Handles dynamic environments in monocular SLAM
• Integrates depth and uncertainty for moving objects
• Improves reconstruction accuracy with uncertainty maps

Innovation

• Uses uncertainty-aware geometric mapping
• Integrates depth and uncertainty information
• Introduces uncertainty map for dynamic removal