A White-Light Based Range Scanner
Investigators: A. Goshtasby
(WSU) and R. Jacobsen (MLPC)
Students: J. Britton, D. Smith, B.
Birkmire, and Ben Scott
This
project is set out to develop a range scanner with 10-micron accuracy
based on subtractive light principle. Rather than projecting a laser
light onto a surface and analyzing the shape of the light stripe on the
surface to determine the geometry of the surfaces, light is obstructed
by a thin fiber and the shape of the shadow profile is analyzed to
recover the surface geometry. The hardware organization of the scanner
is shown below.
Fig.
1.
Hardware
organization of the scanner.
White
light is guided through a thin fiber to the focal point of a lens,
creating cylindrical light toward the surface being scanned. A thin
opaque fiber in front of the surface creates a shadow on the surface.
The shadow profile is detected by computer vision techniques and from
its shape the 3-D geometry of the surface at the shadow is estimated.
By sweeping the fiber over the surface and processing images of the
shadow, the 3-D geometry of the surface is recovered.
The
process of detecting the shadow profile is shown below. A particular
frame is first convolved with two different smoothing operators and
subtracted. This is similar to determining the Laplacian
of Gaussian of the image. The resultant image will have positive and
negative values. The zero-crossings in the image identify the shadow
boundaries.
Fig.
2.
The
process involved in detecting the shadow boundaries.
The
zero-crossings not only detect the shadow boundaries, they detect all
sharp intensity changes on the surface. There is a need to distinguish
zero-crossings that are due to the shadow of the fiber and those that
are due to sharp changes in surface intensity. Information from
consecutive frames is used to determine that. Since the surface and the
camera are fixed and only the fiber moves over the surface, by
subtracting consecutive frames, all edges except shadow edges can be
filtered out. The process is depicted below. Image differencing keeps
information only from the motion of the fiber and filters out other
image details and, thus, removes edges that do not belong to the shadow
boundaries.
Fig.
3.
Subtracting
consecutive frames erases all image details except in the neighborhood
of the shadow profile.
An
example shadow boundary determined by this algorithm is shown below.
From the shape of such a boundary, the 3-D geometry of the surface
along the boundary is calculated.
Fig.
4.
A
detected shadow boundary.
A scan of
the back of the Penny is shown below. Due to occlusion, noise, and
other factors, surface height at all points cannot be determined.
Heights at missing points are estimated by an inverse distance method.
Fig. 6 shows the image after estimating the missing heights in this
manner.
Fig.
5.
An
example scan
of the back of a Penny by the scanner. Black spots show points where
surface height could not be calculated by processing the shadow
profiles.
Fig.
6.
Estimating
missing heights by inverse distance interpolation.
While the above scanner can scan an area of size 1 by 1 square centimeter, a new version of the scanner, using an XY translation stage, can scan an area as large as 10 by 10 square centimeters. The hardware organization of the new XY scanner is shown in Fig. 7.
Fig. 7. Hardware organization of the XY scanner.
Example scans by the new scanner are shown in Figs. 8 and 9. Resolution of this new scanner is 15 microns in X, Y, and Z directions.
Fig. 8. (left) A 1-yen Japanese coin. (middle) The height map of the scanned coin shown in psuedo color. (right) 3-D surface of the coin reconstructed by fitting a surface to the height map.
Fig. 9. This is another scan of the penny shown in Fig. 6 by the new scanner.
For more
information about this scanner, please contact A. Goshtasby
(agoshtas@wright.edu).