A C++/Python toolkit for reconstructing 3D objects from multiple camera views by mapping motion detection from 2D image space into 3D voxel space.
- Process multiple camera feeds in real-time using RTSP streams
- Detect motion between consecutive frames
- Cast rays from changed pixels into a 3D voxel grid
- Accumulate brightness in overlapping regions to reconstruct moving objects
- Visualize the 3D reconstruction in real-time
Run the setup script to install all dependencies:
sudo ./setup.shConfigure your cameras in the camera_config.json file:
[
{
"camera_index": 0,
"rtsp_url": "rtsp://admin:[email protected]:554/live",
"camera_position": [0.0, 0.0, 0.0],
"yaw": 0.0,
"pitch": 0.0,
"roll": 0.0,
"fov_degrees": 60.0
},
...
]- Start the RTSP stream processor:
./build/rtsp_processor camera_config.json output_dir- Visualize the voxel grid in real-time:
python3 visualize_voxel_grid.py output_dir- Mouse: Look around
- W/A/S/D: Move camera
- Space/C: Move up/down
- +/-: Adjust threshold
- Q: Quit
The project also supports batch processing of pre-recorded images using the original ray_voxel.cpp implementation:
./build/ray_voxel <metadata.json> <image_folder> <output_voxel_bin>This project is licensed under the MIT License - see the LICENSE file for details.