STEREO - A Digital System for Surface Reconstruction
This is the project that I did at SUNY-ESF for my Ph.D. degree under the direction of
Dr. R. Brock and Dr. P. Hopkins.
Residing on a SUN SPARC station with an ANDROX image board and two CCD cameras, a
digital photogrammetric system, STEREO, is implemented with approximately 13,000
lines of C code to determine three dimensional coordinates of points of interest
(POI) defined with a grid on a texture-less smooth specimen. Two CCD cameras were set up
with unknown orientation and took digital images of a model and
the specimen. Points on the model were selected as control or
check points for calibrating or assessing the system. With a new algorithm for
edge-detection called local maximum convolution (LMC), the POI's were extracted from
the stereo image pairs and matched automatically. Based on the result of image
matching, the image coordinates of the POI's, together with the image (and object)
coordinates of the check (and control) points, are directly imported into a least
squares "bundle" adjustment procedure. This procedure solves for the camera
orientation parameters, the object
coordinates of the POIs and the ground coordinates of the
check and the control points. By comparing the known and the computed object
coordinates of the check points, the standard deviation of
the residuals at the check points was approximately 24%, 49% and 56% of the pixel
size in the X, Y and Z directions, respectively. The average of the absolute values
of the residuals at the check points was approximately 19%, 36% and 49% of the pixel
size in the X, Y and Z directions, respectively. After interpolating the calculated
coordinates of the POI's, the reconstructed surface can be
displayed for visual examination. With the graphical user interface, STEREO
demonstrated a high degree of automation and its operation does not require special
knowledge of photogrammetry, computers or image processing.
A paper describing this work was published in a
special issue
of Photogrammetric Engineering & Remote Sensing (PE&RS) on
softcopy photogrammetry. It can be downloaded in PDF format
(330Kbyte), which can be read with Adobe
Acrobat Reader.
An abstract of my thesis is also available in Chinese.
Main Page
|| Model
|| Stereo Images
Points of Interest
|| Reconstructed Surface
|| Quality Check
