DGS-SLAM: Gaussian Splatting SLAM in Dynamic Environment

Mangyu Kong,Jaewon Lee,Seongwon Lee,Euntai Kim

2024-11-16

Abstract:We introduce Dynamic Gaussian Splatting SLAM (DGS-SLAM), the first dynamic SLAM framework built on the foundation of Gaussian Splatting. While recent advancements in dense SLAM have leveraged Gaussian Splatting to enhance scene representation, most approaches assume a static environment, making them vulnerable to photometric and geometric inconsistencies caused by dynamic objects. To address these challenges, we integrate Gaussian Splatting SLAM with a robust filtering process to handle dynamic objects throughout the entire pipeline, including Gaussian insertion and keyframe selection. Within this framework, to further improve the accuracy of dynamic object removal, we introduce a robust mask generation method that enforces photometric consistency across keyframes, reducing noise from inaccurate segmentation and artifacts such as shadows. Additionally, we propose the loop-aware window selection mechanism, which utilizes unique keyframe IDs of 3D Gaussians to detect loops between the current and past frames, facilitating joint optimization of the current camera poses and the Gaussian map. DGS-SLAM achieves state-of-the-art performance in both camera tracking and novel view synthesis on various dynamic SLAM benchmarks, proving its effectiveness in handling real-world dynamic scenes.

Robotics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to achieve high - precision simultaneous localization and mapping (SLAM) in dynamic environments. Specifically, although most of the existing dense SLAM methods can use Gaussian Splatting technology to improve the quality of scene representation, these methods usually assume that the environment is static, so they are prone to photometric and geometric inconsistency problems when dealing with dynamic objects. This leads to a decline in the performance of localization and map construction in dynamic environments.

To solve these problems, the paper proposes Dynamic Gaussian Splatting - SLAM (DGS - SLAM), which is the first dynamic SLAM framework based on Gaussian Splatting technology. DGS - SLAM improves SLAM performance in dynamic environments through the following aspects:

1. **Dynamic Object Processing**: DGS - SLAM integrates a robust filtering process to handle dynamic objects throughout the SLAM process, including Gaussian insertion and key - frame selection. This process helps to reduce the noise and inconsistency caused by dynamic objects.

2. **Robust Mask Generation**: To further improve the accuracy of dynamic object removal, DGS - SLAM introduces a robust mask generation method, which reduces the noise from artifacts such as inaccurate segmentation and shadows by enforcing photometric consistency.

3. **Loop - Aware Window Selection Mechanism**: DGS - SLAM proposes a loop - aware window selection mechanism based on unique Gaussian - based key - frame IDs, which helps to detect loops between the current frame and past frames, thereby achieving joint optimization of the current camera pose and the Gaussian map.

Through these innovations, DGS - SLAM achieves state - of - the - art performance in various dynamic SLAM benchmark tests, demonstrating its effectiveness in dealing with real - world dynamic scenes.