The Intel® RealSense™ SDK has been discontinued. No ongoing support or updates will be available.
The 3DScan module available in the Intel® RealSense™ SDK (Windows) can be used to scan stationary objects including the face and body. The capture is stored as mesh data (as OBJ, PLY, or STL output) which can then be rendered, edited, and even printed.
While the Intel® RealSense™ F200 and R200 cameras each have 3D scanning capability, they should be used differently and have different "scopes". The user facing F200 camera is best used for face and small object scanning while the R200 world facing camera should be used for full head, body, and any larger object scans. When scanning with the user facing camera, the TARGET is rotated clockwise while the camera doesn't move. (In the case of face scan, the head is centered, turned right, then left, then centered again). For the world facing camera, the target does NOT move, instead the camera should be walked slowly, counterclockwise around the target.
To aid with the most common types of handheld scans, there are presets available by calling DF_3DScan.exe -<preset> where <preset> can be
-face, -head, -body, or -object.
Additionally you can enable
texture mode (-texture) for sharper color features. To the left see with and without the -texture option.
solidification mode (-solid) to produce a close mesh for printing or simulation. (Works by extending color and surface curvature to close holes in areas not visible to the camera scan.) See bananas without and with the -solid option.
Tips and Tricks
- During the first few seconds (scanning initialization), be sure the scan area is centered in the square green scene indicators. Also provide a horizon for the camera to set scan angle tracking.
- Objects smaller than a few centimeters, or larger than half a meter, may not work as the target.
Front facing camera - the target must be rotated counter clockwise .
Rear facing camera - move the camera clockwise around the non-moving object
Face scanning (F200) Adjust sitting position to put the face and shoulder tops in the green box before the scan starts (during init). Then turn the head right and then left (30 degrees) and then return to center and fix the target's gaze for a few seconds to improve eye color sharpness
Head scanning (R200) - Use a second person to move the camera counterclockwise around the non moving person at a distance of ~1 meter distance after checking head is fully within the green box (during init). Include shoulders for tracking. If the scan appears to have too much hair, tie the hair back or wear a tight fitting cap during scan.
Body scanning (R200) - Start with head scan but then continue with another circle to capture the torso and a third circle to capture the lower legs and feet. (counterclockwise at ~1 meter). If the feet get cut off at an angle, make sure to aim the camera at a horion during init.
Small object scanning (both cameras). Use the object detection techniques to tightly fit the scanning volume 'around' an object sitting on a flat surface, such as a table. The table (flat surface) will automatically be removed from the resulting scan data,
Surfaces that are shiny, reflective, or translucent will likely cause issues. If possible cover or "mute" them.
- IR from sunlight can interfere with the depth imaging. Use diffuse indoor lighting.
- Reconstruct can only be called once per scan The record and playback system can be used to process the same scan.
- Color quality requires the -texture option and may be degraded with -solid option.
- Issue: Holes in recreated surface: Fixes: Object must be totally inside the green box at init. Limit outdoor illumination (even inside) and use the -solid option.
- Issue: Shape detail is missing. Fixes: Try getting closer to the target. Note the minimum size to get a visible feature is .5 cm.
- Issue: Color corrupt or too dark. Fixes: Try using multiple diffuse indoor lights and cover shiny (reflective, translucent) surfaces.
- Issue: Color is not sharp. Fixes: Try getting closer. Be sure to add -texture option.
- Issue: Surface is corrupted. Fixes: Make sure target does not move. Try a different angle. Cover shiny surfaces.
- Issue: Scanning didn't start (init failed). Fixes: Include more surface around the object. Cover shiny surfaces and use multiple lights.
- Issue: 3D capture mesh is not generated or is corrupt. Fixes: Ensure target is stationery (with R200). Include a horizon at beginning of shot (or shoulders during head scan).
- Issue: The first preview image after scanning starts is unexpectedly black. Fix: Restart
- Four part series on 3D scanning and Unity*