Belval / Scanner3D

Using learned and non-learned algorithms to reconstruct 3D objects with the SR300 camera

3DMeshReconstruction

Using registration algorithms to reconstruct 3D objects with the SR300 camera

*Element implemented in the current project

Interfaces

The main interface uses OpenCV and as such is quite limited. Here are the implemented key bindings:

• s : Take a capture
• r : Start recording (takes multiple capture until R is pressed again)
• q : Stop capture
  • In live mode, starts mesh generation
  • In step mode, starts registration
• c : Delete the last capture

Command line usage

Make sure that your camera is connected otherwise the interface will be unresponsive

usage: run.py [-h] [--mode [MODE]] [--output-file [OUTPUT_FILE]]
              [--mesh-reconstruction-method [MESH_RECONSTRUCTION_METHOD]]
              [--max-triangle-count [MAX_TRIANGLE_COUNT]]
              [--ensure-consistency] [--log-level [LOG_LEVEL]]

Scan a 3D object and save the mesh file

optional arguments:
  -h, --help            show this help message and exit
  --mode [MODE]         Mode, should be one of live or step
  --output-file [OUTPUT_FILE]
                        Name of the output file
  --mesh-reconstruction-method [MESH_RECONSTRUCTION_METHOD]
                        Method that will be used to create a mesh from the
                        point cloud. One of [BPA|Poisson]
  --max-triangle-count [MAX_TRIANGLE_COUNT]
                        Maximum triangle count, default is -1 (None)
  --ensure-consistency  Whether to clean the resulting mesh or not
  --log-level [LOG_LEVEL]
                        Python loglevel to use should be one of (DEBUG, INFO,
                        WARNING, ERROR)

Tests

Most features have some level of coverage. You can run the unit tests with python tests.py. Due to the nature of 3D registration you can expect up to 20 minutes for the whole test suite to complete.