Video of reconstruction attempt: https://www.youtube.com/watch?v=2fL4kwdZyag

Important: Make sure you are are using a USB 3 port.

RealSense-Viewer (Optional) Link to heading

Run this to verify that the camera works.

2.54.2 SDK version is needed in order to use L515 with it. Newer version dropped support for L515: https://github.com/IntelRealSense/librealsense/issues/13194

Correct version for L515: https://github.com/IntelRealSense/librealsense/releases/tag/v2.54.2

Build guide: https://dev.intelrealsense.com/docs/compiling-librealsense-for-linux-ubuntu-guide

RTAB-Map Link to heading

ROS2 not needed, just compile/download standalone rtabmap: https://github.com/introlab/rtabmap/releases

For Linux, compiling from source is needed for RealSense support: https://github.com/introlab/rtabmap/issues/694

Difficult to reconstruct environment, slight movement causes the feature not to match. can’t open the .ply pointcloud file in Blender, open it in MeshLab Un-textured mesh, looks bad. Scale seems accurate

Open3D Link to heading

Correct version is important, use this version: pip install pyrealsense2 v 2.54.2.5684

Resolved one of the error with:

pip uninstall opencv-python-headless -y
pip install opencv-python --upgrade

Capture dataset: https://www.open3d.org/docs/latest/tutorial/reconstruction_system/capture_your_own_dataset.html

No textured mesh: https://github.com/isl-org/Open3D/discussions/4947

Tried to generate textured mesh by running color_map_optimization_for_reconstruction_system.py as shown in this example: https://www.open3d.org/docs/latest/tutorial/sensor/realsense.html but failed (likely due to VRAM limitation)

Error in the documentation, use this instead (note the –config): python3 run_system.py --config config/realsense.json --make

  • Huge file size when recording.
  • Not to scale (~98 cm instead of 32cm when measuring the width of the laptop)
  • No texture, running texture generation code failed.

Troubleshoot Link to heading

Realsense-viewer not working all of a sudden, stuck at loading screen: plug into a USB2 slot and run the viewer, then plug it back to a USB3 slot

Conclusion Link to heading

  • Absolutely unusable for 3D reconstruction purpose (Dot3D untested though).
  • SLAM-based software have a hard time registering features.