Form RG5
NATIONAL UNIVERSITY OF SINGAPORE
ACADEMIC RESEARCH FUND
RESEARCH GRANT FINAL REPORT
THE FINAL REPORT MUST BE SUBMITTED WITHIN 30 DAYS OF THE PROJECT COMPLETION
DATE.
IN ADDITION, A FINAL RESEARCH REPORT WHICH CONTAINS A DETAILED
DESCRIPTION OF THE PROJECT (PLEASE SEE PARAGRAPH 12 OF THE TERMS AND CONDITIONS
OF RESEARCH GRANT) SHOULD BE SUBMITTED AND MUST BE IN A FORM SUITABLE FOR
DISPLAY IN THE UNIVERSITY’S LIBRARY.
FAILURE TO DO SO WILL RESULT IN THE REJECTION OF SUBSEQUENT RESEARCH GRANT
APPLICATIONS FROM THE PRINCIPAL INVESTIGATOR UNTIL THE FINAL REPORT IS RECEIVED.
IF THERE ARE EXCEPTIONAL REASONS WHICH PREVENT THE FINAL REPORT FROM BEING
SUBMITTED ON TIME, A WRITTEN REQUEST MUST BE MADE THROUGH HEAD, FACULTY
RESEARCH COMMITTEE TO BURSAR’S OFFICE BEFORE THE DEADLINE, FOR THE SUBMISSION
PERIOD TO BE EXTENDED.
TITLE OF RESEARCH PROJECT: Reconstruction of 3-D Objects from Video Sequences
REFERENCE
NUMBER: R-252-000-
TOTAL PROJECT
COST: $28,500
APPROVAL DATE (dd/mm/yy): 25.11.1998
051-112
APPROVING AUTHORITY:
*FRC/URC/ARC/MLC
Note:
START DATE (dd/mm/yy): 25.11.1998
COMPLETION DATE (dd/mm/yy):30.6.2002
FRC - Faculty Research Committee
URC - University Research Committee
ARC - Academic Research Committee
MLC - Ministerial Level Committee
RESEARCH TEAM PARTICULARS
Principal investigator: Huang Zhiyong
Assistant Professor, Dept of CS, SoC
Other key team members: Leow Wee-Kheng
Associate Professor, Dept of CS, SoC
Other collaborating parties/agencies:
To provide names, posts, departments.
1(6)
PURPOSE AND OBJECTIVES (IN ORDER OF PRIORITY)
Provide details as stated in
the original proposal.
1.
To research the novel polygonization algorithm.
2. To develop a software and hardware platform for reconstructing 3D models of objects from video sequences.
ACHIEVEMENTS
Major achievements are two novel methods have been developed:
(1) Frontier Advancing Polygonization Method. It is devised specially
for 3D points recovered from image sequences of objects and scenes.
The results are polygon meshes representing the objects and scenes.
A mesh model represents an object, terrain or other surface as a series
of interconnected planar shapes.
The 3D points recovered are constrained by the features present in the
images. They are not densely or uniformly sampled that the existing
methods are not suitable. The method addresses the problem from the
following aspects:
a.
b.
c.
d.
Describe the extent to which
the objectives of the research
have been achieved and relate
the significance of the
research to recent work in the
field.
Provide details on the
conferences/seminars/
workshops held in
conjunction with the research
project. Please compare
against targets.
It constructs a mesh from advancing the frontier of smooth
surfaces, formed by more reliable 3D points, towards surface
discontinuities. Thus, the surface discontinuities, i.e., corner
points and edges of the shape, can be computed.
It only adds triangles to the mesh and each added triangle is in the
final mesh.
It constructs Delaunay triangulation/Voronoi diagram of sample
points and the resulting mesh is therefore optimal.
It can incorporate many different mesh construction requirements
easily.
(2) Error-Predictive Correspondenceless 3D Shape Reconstruction. This
method is devised for recovering 3D points from image sequences of
objects and scenes. The results are 3D points representing the captured
2D feature points of objects and scenes. The method has the following
features:
a.
b.
c.
d.
It is a correspondenceless approach. It is not necessary to
compute the correspondence of 2D feature points in image
sequences. It would bypass the difficulties and potential errors of
deriving the correspondence.
It is based on the use of optimizing spheres in 3D space that is
different from the existing volume based approach. The
advantage over the volume approach is that the result is
independent of the accuracy of the volume decomposition.
Moreover, it can predict the error of reconstruction.
The theoretic analysis and experimental have shown the method
has better accuracy over the existing methods. Its accuracy is
about 1% of the object size if actual model is required and about
0.5% if structurally identical model is required.
It is used as a core technique in a desktop 3D reconstruction
system consisting of a PC, a step-motor driving turntable
controlled by PC, and a CCD camera.
STUDENTS TRAINED
Number of students trained
4 MSc: Zhang Yong, Indriyati Atmosukarto, Zhou Luping, Liu Sujin
2 Honours student: Merzad Gotla, Neo Chee-Seng
during the research project,
type of training received.
Please provide targets and
actual numbers.
Type of Student
i) PhD
ii) Masters
iii) Undergraduate
If possible, indicate whether
these students intend to
continue research.
Neo C. S. and Huang Z. Texture Mapping on Doo-Sabin Subdivision
Surfaces Using Multiple Images, accepted by Eurographics 2002 for short
presentation.
List titles of papers and
internal reports etc arising
from the research (indicate if
they were refereed or nonrefereed).
Zhou L, Atmosukarto I, Leow W K, and Huang Z. Reconstructing surface
discontinuities by intersecting tangent planes of advancing mesh frontiers.
In Proc. Computer Graphics International, 2002.
Describe any significant
publicity obtained for this
work.
PUBLICATION AND DISSEMINATION OF RESULTS
Atmosukarto, I., Zhou, L., Leow, W. K. and Huang Z. Polygonizing Nonuniformly Distributed 3D points by Advancing Mesh Frontiers. In Proc.
Computer Graphics International, 2001. 175-182.
Neo C. S. and Huang Z. Texture mapping on subdivision surfaces using
multiple images. CADGraphics 2001. 170-179.
Atmosukarto I., Leow W. K., Huang Z., Zhang Y. and Sung K. K., Mesh
Construction from Non-Uniformly Distributed and Noisy 3D Points
Recovered from Image Sequence. PG 2000, 423-424.
Leow W. K., Huang Z., Zhang Y., and Setiono R. Rapid 3D Model
Acquisition from Images of Small Objects. GMP 2000, 33-41.
Liu S. and Huang Z. Interactive 3D Modeling Using Only One Image.
ACM Symposium of Virtual Reality Software Technology (VRST) 2000,
49-54.
Leow W. K., Huang Z., Zhou L., Atmosukarto I., Zhang Y. Acquiring 3D
Models from Images for Multimedia Systems. MMM 2000. 153-166.
Gotla M. and Huang Z., A Minimalist Approach to Facial Reconstruction.
Eurographics 1999 Short Papers and Demos. 12-14.
Leow W. K., Huang Z, Accurate and True Reconstruction of 3D Models
with Closed-Form Solutions, Technique Report, School of Computing,
NUS, 1998. No. TR12/98.
EXPLOITATION OF RESULTS
One patent Frontier Advancing Polygonization. has been filed in USA,
October, 2001.
Describe any
patentable/commercially
exploitable results, and what
follow-up arrangements have
been made or are planned.
2(6)
PROGRESS OF THE RESEARCH
(1)
Was there any significant change in the research compared with
the original proposal?
If YES, give reasons for
changes.
No.
If NO, give details.
(2)
Did the research proceed as expected and on time?
Yes.
(3)
Were there any circumstances which aided or impeded the
progress of the research?
No.
If YES, give details of
problems and explain the steps
you took to overcome them.
Examples could include
difficulties in staff recruitment,
late delivery of equipment,
equipment malfunction etc.
FURTHER RESEARCH
(1)
Has the research led to further grant applications from the
Academic Research Fund?
If YES, give details of the
outcome. List grant
applications, giving dates and
grant reference number.
Indicate value of any grants
awarded.
Has the research stimulated further work in other ways?
If YES, provide details on
research parties and agencies.
No.
(2)
Yes.
“3D object retrieval”, A/P Leow Wee-Kheng is the PI.
EXTERNAL REVIEW (for ARC and MLC projects only)
Please attach external referee’s post-project review.
3(6)
FINAL STATEMENT OF ACCOUNTS
Items
Manpower
E/F
M/C
O/T
Total approved grant
5,400
8,700
2,900
(2)
Total actual expenditure
1,900
5,966.79
1,082.75 4,076.23 5,877.92 18,903.69
(3)
Total outstanding commitments
0
0
0
(4)
= (1-2-3)
Total grant balance
Interest Earned*
3,500
2,733.21
1,817.25 223.77
1,322.08 9,596.31
0
0
0
0
= (4+5)*
Total Uncommitted Balance
(To Be Returned to Academic
Research Fund)
3,500
2,733.21
1,817.25 223.77
(6)
Note:
0
0
7,200
TOTAL
(1)
(5)
4,300
T/M
0
Part time undergraduate student assistantship:
Wu Yinghui, $500
Xu Jie, $500
Top-up PhD student:
Shou Lidan, $900
0
0
1,322.08 9,596.31
The project account will be closed within 1 month from the date of this Form BUR/RG5.
Principal Investigator should therefore submit all claims/invoices for any outstanding purchase
orders or commitments as soon as possible but no later than the aforementioned deadline.
*
Not Applicable for projects approved by the URC
E/F denotes new equipment/facilities
M/C denotes materials/consumables
O/T denotes overseas travel
T/M denotes training costs/other miscellaneous costs
MANPOWER TRAINED
28,500
List the staff employed directly on
the grant and any other research
fellows/research students
associated with the research
project, highlighting the
manpower grades and numbers.