Table of Contents
-
Welcome to DAVID (help_welcome.htm)
-
Camera Calibration (help_camera_calibration.htm)
-
3D Laser Scanning (help_3d_scanning.htm)
-
Shape Fusion (help_shape_fusion.htm)
-
Frequently Asked Questions (help_faq.htm)
-
General Hints (help_hints.htm)
Frequently Asked Questions
General Questions about DAVID
-
What is the meaning of "DAVID"?
You have the choice:
a) Definitely Affordable Vision Device
b) It's just a nice name for a small but powerful software between all those Goliaths... ;-)
-
How does DAVID work?
The concept has been published as a Paper at the German Association for Pattern Recognition
(DAGM, Deutsche Arbeitsgemeinschaft für Mustererkennung) and can be downloaded
here.
-
Is there a forum about DAVID?
Yes! Click here.
Please do not hesitate to ask questions, answer questions or post
comments.
Performance
-
How accurate are the scanning results?
They are surprisingly good! Of course this depends on the hardware you
use and the setup (camera distance, object size, light conditions,
triangulation angle, quality of the calibration target...). In our setup with a not-low-cost grayscale camera, a
rather bright laser, an object size of about 50 mm, a camera distance of about
600 mm, and a reasonable triangulation angle of about 30°, we obtained errors
of less than 0.4 mm. Generally, the error is not more than 1% of the
camera distance.
Remark: If you execute the camera calibration correctly, the 3d output will not only
contain the shape of the object, but will have a correct absolute scale!
There are many things that affects the scan quality.
Regarding the camera:
- as little noise as possible
- high photosensitivity
- grayscale camera instead of color camera (color pixels are interpolated)
- image resolution (affects the primarily the mesh resolution)
- quality of the lens (glass is better than plastic)
Regarding the laser:
- laser line as thin as possible
- as bright as possible
- perfect straight line
- a red or green laser if you use a color camera (because DAVID can use the red
or green channel)
Regarding the calibration corner:
- the corner must have an precise angle of 90 degree
- the paper with the markers must not be wavy (the best way is to glue it on the panels)
-
What mesh resolution (number of polygons) can be achieved with webcams?
You get one mesh vertex per foreground/object pixel. If you use a 1.3
megapixel camera and if the depth data of your object (that you have scanned)
covers 1/3 of your camera image, than you get approximately 450 000 vertices.
But remember: mesh resolution and mesh accuracy are two different things.
Hardware
-
What kind of camera / laser do you recommend?
Before you buy expensive hardware, we suggest that you make your first
experiments with cheap hardware and see whether it’s sufficient for you or
which problems you have. First try to solve them as much as possible by
changing your setup (camera settings, room brightness, background color, object
color, background corner precision).
LASER: You only need to use an expensive (high power) laser if
you cannot make your room dark enough, or if you have to scan really dark
objects. Or if you have too much noise (e.g. Speckle effect,
http://en.wikipedia.org/wiki/Speckle_pattern,
http://de.wikipedia.org/wiki/Speckle) which you cannot avoid by improving
camera settings (exposure time, aperture size...). You also should avoid a
large brightness difference between object and background; so if you cannot
make your object brighter, use a darker background.
We use a 5 mW laser with a 30° or 60° line module. We also use a cheap 10-Euro-Laser
from a home improvement store. It is really sufficient in not-too-bright rooms
on not-too-dark objects!
CAMERA: High resolution is not the most important property!
(Low noise may be more important.) The necessary camera resolution depends on
the object’s shape. Only very small details in the object require high
resolution. Remember how we create the 3d data: For each image pixel, we create
one 3d point. With a 640*480 camera, set up so that the object maybe covers
some 300*350 pixels, you get approx. 100,000 3d points just for the visible side of
the object! I think in most cases this is more than enough, you might even want
to scale it down by factor 4 or 9 or so (thus decreasing the noise).
We use Firewire cameras. For our test objects, we are
very satisfied with a 640*480 grayscale camera. We also tried a webcam (640*480)
for 20 Euro. The results are worse, of course, but they may be sufficient in
many cases.
LENS: Although our camera calibration does compensate linear
and nonlinear lens distortion, a good lens will increase precision. We have not
made any experiments comparing different lenses.
GENERAL: Do not forget that the components depend on each
other. For example, a high resolution camera will not be advantageous if your
laser line is rather thick! And a very good lens will not help when your
calibration corner has not been constructed very precisely!
Troubleshooting
-
Calibration works, but I get no 3d data while scanning.
Instead, I get the error message "LASER LINE NOT DETECTED"
What is wrong?
DAVID is not able to detect the laser line in the image. There may be several reasons for that:
IMAGE SETTINGS:
The brightness settings during scanning need to be much darker than during calibration.
In the Scanning dialog, you can select "Show camera image". This shows you the image that
DAVID has to work on.
Please compare your camera image with the ones in our online help. You might have to
decrease exposure time, or aperture size, or the general brightness in your room.
- The laser line must be clearly visible: on the Calibration Corner in the left and
right part of the image, and on the object in the middle.
- The rest of the image should be rather dark or even totally black. If it is not totally
black, it must be perfectly constant (no flickering from a neon lamp).
GEOMETRY:
- The laser line must reach from the left to right border of the image.
- The background planes must stand in an angle of exactly 90°.
- The calibration pattern printouts must be attached precisely in the corner.
-
During scanning, I get the error message "INTERSECTION ANGLE TOO LOW!"
What does it mean?
You need to hold your laser at a different position (in most cases: higher).
DAVID calculates 3d coordinates by intersecting the laser plane with "viewing rays" through the camera.
That is why the laser plane must not be too close to the camera.
-
My surface is heavily spiked. How can I reduce that?
Spikes are often caused by laser reflections at the surface. Try to
reduce the camera exposure time / shutter / brightness... such that only one
single laser line (without laser reflections) is visible. You can furthermore
reduce noise by pressing the smoothing buttons multiple times and/or by sliding
the laser more often over the surface.
If this does not help, you might coat the object with a matt paint or a powder.
Wenn dies nicht hilft müssen sie eventuell das Objekt mit einer matten Farbe oder einem Puder versehen.
-
How can I scan bigger or smaller objects?
Theoretically, you can scale the whole setup up or down as far as you
want. For very large objects, you will need a very large Calibration Corner
with an accordingly large calibration pattern, and a bright laser. For a very
small setup, the laser line should be as thin as possible.
Hint: Instead of using two walls in the background, you can simply use a single
vertical wall together with the floor. Floor and wall build a perfect
'calibration corner'; if you rotate the camera (and laser) about 90 degree....
Advanced Scanning
-
How can I scan dark objects?
Scanning of rather dark objects is difficult. Maybe the following idea
helps (not for perfectly black objects, of course): The problem is that only
little laser light is reflected by the object. So you increase the camera's
aperture size and/or exposure time --> now the laser is sufficiently
reflected by the object, but the background is too bright! --> the camera is
oversaturated, the laser line gets too wide, cannot be detected and so on
--> scanning doesn't work. So our simple idea is to replace the bright
background by a dark one (about as dark as the object is). Of course it has to
be the same structure as in the calibration, so you should use the bright
printouts for calibration and attach dark paper for scanning. Or you can use
the "Invert" option in the Calibration window and use white calibration markers
on dark planes.
-
Can I scan a person that has open eyes?
We strongly advise against trying to scan an open eye of a person!!
You find important safety-related warnings on every commercial laser device!
The best way is to scan the head with closed eyes and to modify the mesh
afterwards. The mesh differences between closed and open eyes are not high. The
main differences can be seen in the texture, and to get the texture you need no
laser.
Table of Contents
-
Welcome to DAVID (help_welcome.htm)
-
Camera Calibration (help_camera_calibration.htm)
-
3D Laser Scanning (help_3d_scanning.htm)
-
Shape Fusion (help_shape_fusion.htm)
-
Frequently Asked Questions (help_faq.htm)
-
General Hints (help_hints.htm)