TDGM-SLAM: A Robust Two-Stage Dynamic SLAM with Guided Motion Segmentation

Video Demos

TDGM_demo1.mp4

TDGM_demo2.mp4

Overview

This repository provides supplementary materials showcasing real-world experiments as outlined in Section 4.4 of the Main Text. The demonstration videos, captured using handheld smartphones, highlight the capabilities of TDM-SLAM in challenging environments.

Explanation

Demo Details

Demo 1:
Recorded in an indoor setting—a long corridor with varying lighting and dynamic pedestrians.

Demo 2:
Captured on an outdoor road, featuring complex dynamic scenes with moving vehicles and pedestrians.

Video Analysis

• Input Frames vs. Dynamic Remove:
  The upper part of the demo videos compares key frames ("Input Frames") with frames after dynamic region removal ("Dynamic Remove"). Dynamic regions are identified using TDM-SLAM's motion segmentation, enhancing scene clarity.

• Initial Points 3D and Scene Reconstruction:
  The lower section showcases 3D point cloud maps ("Initial Points 3D") and their rendering into a final 3D Gaussian scene model ("Scene Reconstruction"), visualized using the SIBR viewer.

• Localization:
  On the right side, "Localization" demonstrates TDM-SLAM's real-time camera pose estimation and map point construction, emphasizing accuracy and robustness in dynamic environments.

Throughout the demonstrations, key moments are highlighted with informative prompts, providing insights into TDM-SLAM's operation and performance.