|
Captured Data for Vaulted Ceiling
Summary:
The
data was captured by the Computer Graphics Group at the Department
of Computer Science, Yale University. You are free to use this
dataset for research purposes, only. If experimental results are
obtained that use this dataset, all publications of these results
should acknowledge its use by referencing the following paper:
@inproceedings{XuGeRuDo06, author = "Chen Xu and
Athinodoros Georghiades and Holly Rushmeier and Julie Dorsey", title
= "A System for Reconstructing Integrated Texture Maps for Large
Structures", booktitle = "3rd Int. Symp. on 3-D Data Processing,
Visualization and Transmission", year = "2006"}
Without permission from Yale University, this data
cannot be incorporated into a larger database, which is then
publicly distributed.
Data
Description:
The data was captured in
the lobby area of the Mechanical Engineering Department building
(known as Mason Lab) at Yale University. The tiled, vaulted ceiling
was designed by Guastavino. The main challenge in this dataset is
the spatially, temporally, and spectrally varying illumination. The
spatial variations can be quite significant for the areas
illuminated by the spot lights, while the spectral variations arise
from the multiple types of light sources that range from fluorescent
to incandescent to daylight.
The range data was
captured with a Cyrax 2500 (now Leica HDS 2500) time-of-flight
scanner. A total of 23 range scans were captured at five different
locations. At each location, a number of scans was captured with
different, overlapping viewpoints, as the scanner was rotated from
one side to the other. For the five locations, the number of scans
was as follows: 3, 4, 6, 5, and 5. Note that the accuracy of the
range data is about +-5 millimeters.
For each scan, a
corresponding set of color images was captured with different
exposures using an Olympus C8080WZ digital camera. For all scans,
except one, five images were captured with a shutter speed of: 1/50,
1/15, 1/4, 1, and 4 seconds. For the fourth scan of position four,
the image with the 4 sec exposure was corrupted. Therefore, a total
of 114 color images were captured. For each set of multiple exposure
images, HDRShop (http://www.hdrshop.com/) can be used to combine
them into a single, high-dynamic range image. (The camera response
function stored in the file NeutralFluorescent_505iterations_curve.m
should be used. For the work in the above-mentioned paper, we used
HDRShop, version 1.)
Each high-dynamic-range image was registered
with respect to its corresponding scan. A UI was used to pick a set
of approximately 20-30 corresponding pairs of points in the 3D
geometry (range scan) and the 2D image. These correspondences are
store in the mason_pos0*_v01*.dat files. Tsai's algorithm was used
to perform the camera to scanner calibration for each HDR image and
its corresponding range scan. The results of the Tsai calibrations
are stored in the mason_pos0*_v02*_cp.dat files.
Details:
The
following
describes the different types of files in this folder in more detail:
A. Data files:
mason_pos0*_v0*__2_05_05_pr02_sc01.ptx
The exported file from the Cyrax 2500
scanner. The first two rows give the size of the range scan (number
of rows and columns). The laser dot moves from top to bottom and
from left to right. The captured data (starting from the 11th row)
is organized in 7 columns: three for the xyz coordinates of each
measured point, followed by the return intensity (a number between 0
and 1, usually around 0.5), and finally the scanner onboard camera
colors corresponding to that point. Note that the resolution of the
onboard camera is fairly low and not at all calibrated with the 3D
geometry -- too cumbersome to calibrate -- hence we resorted to
using an external digital camera. (The only real use of the image
data from the onboard camera was to select the region one wanted to
scan. It is not advisable to use this image data in the last three
columns of the PTX files in any other way.) Last, note that the
coordinate frame of the scanner is such that the z-coordinate is in
the negative viewing direction.
mason_pos0*_v0*__*_*_*_pr02_sc01__2054135__WB@NeutralFluorescent__ISO@50__FL@7.4mm__Focus@220__Aperture@8.0__Shutter@50.TIF
The
captured images at five different exposures, corresponding to each
range scan. (Note that for the fourth scan of position four the
image with a 4 sec exposure was corrupted.) For the example filename
above, "Shutter@50" means that the exposure was 1/50 seconds.
HRDShop can be used to blend these images into a high-dynamic range
image. (The camera response function stored in the file
NeutralFluorescent_505iterations_curve.m should be used with
HDRShop.) The rest of the filename gives further information about
other camera settings, the date of image capture, and the position
and view number of the corresponding range scan. Note that these are
the raw captured images, i.e., they were not rectified to remove the
camera lens distortion calculated using Tsai's algorithm.
mason_pos0*_v0*__2_05_05_pr02_sc01__max_vertex_dist@4cm__clip_depth@7m_med_mm_ed.ply
The PLY files of the triangulated
3D mesh for each range scan. Note that each
PLY file was median filtered to remove some but not all
noise, had its units converted to millimeters, and then edited
by hand in 3D to remove the wires that
supported the spot lights. The distance threshold for incorporating two
neighboring vertices into a triangle was 4cm. mason_pos0*_v0*.dat -- The
2D/3D correspondences between each range scan and its corresponding images. The
first three columns are the xyz
coordinates of
the selected points on the geometry (units in millimeters),
while the last two columns are the pixel coordinates
of the corresponding points in the image. mason_pos0*_v0*_cp.dat -- The output
of the Tsai calibration. The 7th and 8th numbers are
the estimated optical center of the camera (in pixels). The 9th is
the scale factor accounting for the uncertainty of the camera's
resampling of the horizontal scanline. The 10th number is the
effective focal length of the pin-hole
camera. The 11th is the 1st order radial lens distortion coefficient.
The next three numbers are the translational components
(in
millimeters)
for the transfor between the scanner and camera coordinate frames. The next three
numbers are the rotation angles (in radians) about the major
axes for the transform between the scanner and camera coordinate
frames.
Check
http://www.cs.cmu.edu/afs/cs.cmu.edu/user/rgw/www/TsaiCode.html (or google "Tsai Calibration") for more information about the output file.
mason_pos0*_v0*_cp.cam
The GL
camera files relating the camera to the laser scanner coordinate
frame for each range scanner and high-dynamic-range image pair. The
distance units are in mm. (Note that these files assume that the
images have been rectified to remove the lens distortion. The
captured images in this directory need to be rectified before
attempting to do texture-mapping using these GL camera files.)
NeutralFluorescent_505iterations_curve.m -- file containing the
camera response function to be used with HDRShop to create the
high-dynamic-range images.
B. Full Model
(non-texture-mapped):
mason_full_1_3_2ed_b_4_5_ed__5mm_L1_q5__nzf_ifs_ed__n30_d2_15__ed2.wrl
This
combines all 23 scans into 3D model. This was the complete model
used in the above-mentioned paper for creating the integrated
"unlit" texture map of the vaulted ceiling.
Request
a Data DVD
|
