工研院之三次元擷取與顯示技術發展

lynxherringΤεχνίτη Νοημοσύνη και Ρομποτική

18 Οκτ 2013 (πριν από 3 χρόνια και 10 μήνες)

79 εμφανίσεις

影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



25

Computer Vision and Virtual Reality Techniques and Applications
---

A Progressive Report on Recent Research Results in Computer
Vision Laboratory at National Chiao Tung University


Wen
-
Hsiang Tsai, Ph. D.

Professor & Dean of Academic Affairs


Department of

Computer & Information Science

National Chiao Tung University

whtsai@cis.nctu.edu.tw



Abstract

Computer vision and virtual reality are the key technologies to the success of
man
-
machine interfacing. In the pas
t five years the research group in the Computer
Vision Laboratory in the Department of Computer and Information Science at
National Chiao Tung University has developed many computer vision and virtual
reality techniques and applications. Emphasis of the de
velopments was placed on
vision
-
based virtual reality studies in the following four research directions: (1)
autonomous land vehicle guidance techniques; (2) environment learning by land
vehicle navigation; (3) human body image analysis for virtual reality

applications;
and (4) computer vision techniques for miscellaneous applications. In this report, the
research ideas for each of the four directions are introduced, followed by briefly
descriptions of the main research results. A conclusion is found at the

end.



I.

Introduction

With the fast advances of computer technologies, more and more applications of
computer vision (CV) and virtual reality (VR) become feasible due to the availability
of fast
-
speed computers and related equipment. On the other hand, the
3D nature of
CV and VR technologies contributes much to the friendliness of their applications in
various fields of man
-
machine interactions because human vision is stereo. It is
expected that CV and VR applications will penetrate every corner of human liv
es in
the next century.

With such a view
-
scope, the research group in the Computer Vision Laboratory
of the Department of Computer and Information Science at National Chiao Tung
University has, in the past five years, devoted themselves to research and
影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



26

dev
elopments of many fields related to CV and VR technologies and applications.
The emphasis is placed on the concept of vision
-
based VR, and the results may be
categorized into the following four directions:


1.

Autonomous land vehicle guidance techniques;

2.

Envi
ronment learning by land vehicle navigation;

3.

Human body image analysis for VR applications;

4.

CV techniques for miscellaneous applications.


In this report, we will present the achievements of our laboratory in the past five
years in each of the above four d
irections, in the way of describing the basic research
ideas of each direction first, followed by giving a brief introduction to each study in
the direction. A conclusion can be found at the end of this report.


II.

Autonomous Land Vehicle Guidance Techniques

Vehicles on the land need guidance to achieve safe navigation. And it is always a
dream of human beings to accomplish automatic vehicle guidance. Guidance
automation is beneficial to many purposes, like ease of driving, help to handicapped
people, promotio
n of driving safety, realization of unmanned navigation in dangerous
areas, etc. Emulation of human vision in vehicle guidance to achieve desired
automation is the most intuitive way everyone into this research field will adopt. And
this leads to the use o
f CV. Video cameras are the most frequently used sensors for
vision
-
based autonomous land vehicle guidance researches.

Issues involved in such researches include vehicle structure and sensor
configuration design, environment image analysis and feature extr
action, vehicle
location computation, wheel movement control, etc. It is also desired to equip the
vehicle with the capability of collision avoidance. More interesting is the design of an
autonomous land vehicle, which can follow a human being in front. Ma
ny
applications follow from the availability of such an intelligent vehicle. The inherent
principle we adopted in these studies is basically model
-
based. That is, vehicle
guidance is realized by comparing sensed environment features with environment
models

that were learned before navigation sessions.

Furthermore, there are two types of vehicle navigation environments
---

indoor
and outdoor. Images of indoor environments are simpler to process while those of
outdoor ones more complicated. Adopted image fea
tures in indoor environments
include wall base lines, wall corners, vertical lines on walls, lines of wall intersections,
ceiling lamp shapes, special marks artificially attached on walls, etc. And useful
features in outdoor environments include road lane
lines, sidewalk curb edges and
影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



27

corners, road surface areas, traffic lines at crossings, colors of lines and areas, etc.

Research results related to this research direction which we have obtained in the
past five years are introduced briefly in the followin
g. Instead of giving technical
details, we describe major research ideas and application significance.


A1.

Vision
-
based autonomous land vehicle guidance in outdoor environments
using combined line and road following techniques [1]
---

Line following and road
following are two techniques often used in
outdoor vehicle guidance. However due to the complexity of outdoor scenes
and environments, either of them alone is insufficient, and in this study we
achieved good vehicle guidance results by adaptive use of the
two techniques
in the navigation process. The features we used are road lines and surfaces,
and the detection of them is based on the use of a clustering algorithm.


A2.

Vision
-
based guidance for autonomous land vehicle navigation in outdoor
road environments
with static and moving cars [2]
---


A further step to the success of outdoor vehicle guidance is achieved in
this study, in which we have designed effective algorithms for avoidance of
collisions with the static and moving cars faced by the autonomous veh
icle
during the navigation session. Detection of the existence of a car is completed
by matching of the road model with sensed road image input.


A3.

Vision
-
based obstacle detection and avoidance for autonomous land
vehicle navigation in outdoor roads [3]
---

In this study we developed further methods for detecting obstacles
found in front of the autonomous vehicle (not limited to cars), as well as
methods for deriving navigation paths to avoid the obstacles. Obstacle
detection is based on a hypothesis test to
decide if a shape area in the road path
is the shadow of a 3D rigid body or not. The results of the three consecutive
studies A1 through A3 above together offer an effective mechanism for
automatic outdoor vehicle guidance in outdoor campus roads.


A4.

Obstacl
e avoidance for autonomous land vehicle navigation in indoor
environments by quadratic classifier [4]
---

In this study, we emphasized autonomous vehicle guidance in indoor
environments. Before this study, we have completed quite a number of indoor
guidanc
e studies. Here we developed mainly a unified technique that can be
used both for detecting regular corridor walls and irregular obstacles found in
影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



28

front of the vehicle. The key technique is the use of the quadratic classification
scheme in pattern recogni
tion theory. Taking corridor walls and obstacles are
taken as input to the classifier, the decision boundary of the classifier offers
exactly a smooth navigation path for the vehicle navigation,


A5.

Smooth Autonomous Land Vehicle Navigation in Indoor Environm
ents by
Person Following Using Sequential Pattern Recognition [5]
---

Person following is an interesting topic in autonomous vehicle guidance.
We found sequential pattern recognition techniques are useful here, being
effective to keep the vehicle stable in

a straightforward driving and avoid
inappropriate following of a person when he/she walks in a curve but does not
mean to do so. The tracked target here is a rectangular shape attached on the
back of the person. The features extracted for use in sequentia
l pattern
recognition are the length of one side of the shape in the image and the
position of the person with respect to the vehicle navigation path.


A6.

Obstacle avoidance in person following for vision
-
based autonomous land
vehicle guidance using vehicle l
ocation estimation and quadratic pattern
classifier [6]
---


This study is a further development of the last research topic, merging
the technique of collision avoidance using the quadratic classifier into the
result of the person following technique menti
oned previously. A vehicle with
complete person following and collision avoidance capability was thus
completed. It can be used in many applications, like a cart or vehicle for use in
a supermarket, a hospital, or a golf field. It can also be used as a toy

dog for
children.



III.

Environment Learning by Land Vehicle Navigation

For many CV and VR applications, it is required to set up environment models,
such as in the cases of autonomous vehicle navigation, VR environment creation,
building model establishment,

hazard environment exploration, etc. Driving a land
vehicle is a good way to collect indoor environment images, and it is even more
exciting to let the vehicle drive by itself, i.e., to accomplish automatic vehicle
navigation without the existence of an e
nvironment model. This defines a problem of
environment learning, which involves the study of learning mechanisms.

There are two possible ways of learning, supervised and unsupervised. For the
case of vehicle navigation here, supervised learning means to
have a man driving the
影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



29

vehicle through the environment, while unsupervised learning means to let the vehicle
navigate automatically by itself without human involvement. The latter approach is a
much more challenging work. We have tried both approaches with

good results.
Furthermore, we have proposed schemes for learning vehicle guidance strategies in
addition to environment model structures.

After environment images are analyzed and relevant features collected, some
ways to merge feature data and create ima
ge
-
based or graphics
-
based environment
models are desired. Visualization of the models is also required. Some results of our
studies in this research direction are briefly described in the following.


B1.

An incremental
-
learning
-
by
-
navigation approach to visio
n
-
based
autonomous land vehicle guidance in indoor environments using
vertical line information and multi
-
weighted generalized Hough
transform [7]
---


It is attractive to design algorithms with learning capability during
regular navigation sessions. And t
his goal has been reached in this study for
the autonomous vehicle. An incremental learning process is performed when
the vehicle is navigating in the indoor environments. More navigation sessions
will result in a more accurate model in the system. It is a
ccomplished by
merging the current learned model with the newly
-
sensed environment
features using a new line
-
pattern matching scheme designed in this study,
called multi
-
weighted generalized Hough transform.


B2.

Unsupervised learning of unexplored environment

by pushdown
transducer for autonomous land vehicle navigation [8]
---


The main achievement of this study is to find out the effectiveness of
the pushdown transducer in the automata theory for use in the design of a
systematic scheme for exploring a given

closed space. Our algorithm based on
the transducer guarantees no repetitive visit of identical locations and no
missing of any spot in the given space, as proved by several theorems. This
research result is exciting because it essentially has achieved th
e goal of
unsupervised learning.


B3.

A new approach to vision
-
based unsupervised learning of unexplored
indoor environment for autonomous land vehicle navigation
---


This study basically is a follow
-
up of the previous one with the aim of
applying the theory
of pushdown transducer to vehicle learning in real
environments. The research was successful in conducting experiments of
影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



30

vehicle navigation in indoor corridors without human involvement. The
corridor sections in a building were explored completely by a re
al
experimental vehicle and a model of the corridors was set up.


B4.

An intelligent system for learning environment models and guidance
strategies for vision
-
based indoor autonomous land vehicle navigation
[10]
---


In this study, we designed an even more int
elligent learning system,
which not only can learn environment models but also guidance strategies
simultaneously. Guidance strategies for each indoor corridor sections are
different, including line following, model matching and dead reckoning. A
mechanism

for guidance strategy switching at corridor intersections was also
proposed. This might be the first result of learning these two types of
information simultaneously in the world. Experimental results showed the
feasibility of the proposed approach.


B5.

3
-
D
i
ndoor
v
irtual environment modeling
via

vehicle navigation and
multi
-
camera image data fusion

[11]
---


In this study, an offline automatic 3
-
D virtual indoor environment
modeling method by the autonomous vehicle navigation is developed. A

land
vehicle

is f
irst driven manually along
an

indoor corridor

path

decided by a
driver and the
environment

data are collected
with

three cameras mounted on
the
vehicl
.
Two of the cameras collect baseline information from the two sides
of the corridor in a building and the

third takes images of the corridor ceiling.
T
he environment model
is then generated
by a data fusion approach using the
multiple
-
view data, and a
corresponding
3
-
D
VR
model
is created accordingly
.
The result can be used for various applications in which u
sers can have a
better feeling of involvement in the VR
environment
.



IV.

Human Body Image

Analysis for VR Applications

There are two major approaches to VR technologies, namely, graphics
-
based
and vision
-
based. The latter approach means the use of CV techniq
ues to help VR
studies, in which a major issue is to track the human body activity for correct
corresponding VR interaction. The tracking approaches by CV may be categorized
into two types: outside in and inside out. The former approach adopts the way of
m
onitoring the operator’s body movement from sensors (like cameras) set up around
the operator, while the latter utilizes sensors mounted on some tools worn by the
影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



31

operator. We prefer the former approach to the latter because in the latter the operator
does

not have to wear any sensor
-
mounted tool and so feels more comfortable.

The parts of the body that are tracked for VR interaction in our studies include
the operator’s face, head, hands, and legs, or a certain object (like a hat or glasses)
which are spe
cially designed in advance with abundant shape information useful for
the tracking purpose. Relevant techniques used here include calibration for sensor
setup, analysis of sensed images, computation of 3D information of tracked objects or
shapes, transform
ation of body movement into VR interaction, etc. Some results of
our researches in this aspect are described briefly as follows.


C1.

Position and orientation tracking using
perspective
-
transformation
-
invariant generalized Hough transform for
virtual reality a
pplications [12]
---


In this study, a user wears a hat on whose flat top five white points with
unequal distances are marked, forming a shape of “+” with the five points
being located at the four ends and the center. And a camera mounted on the
ceiling fr
om the top of the user is used to take images of the hat for point
detection and pose determination. The computation result is a set of the
position and orientation parameters of the hat with respective to the camera,
which may be used for computing the re
lative location and direction of the
user’s head with respective to the VR environment in the monitor. This is
useful for human interaction with the VR environment.


C2.

Reliable determination of object pose from line features by hypothesis
testing [13]
---


I
n this study, to ensure the quality of the
object
pose estimated from line
features
, which is necessary for a reliable computer vision system
, two simple
test functions based on statistical hypothesis testing are
proposed
.
The first is
an error function ba
sed on the relation between the line features and some
quality thresholds, from which

poor input can be detected before estimating
the
object
pose.
T
he second test function
is

used
,

after

pose estimation
,

to
decide if the estimated result is sufficiently a
ccurate. Experimental results
show that the first test function can detect input with low qualities or
erroneous line correspondences, and that the overall proposed method yields
reliable estimated results.


C3.

Vision
-
based tracking and interpretation of huma
n leg movement for
virtual reality applications [14]
---


影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



32

A system for tracking and interpreting leg motion in image sequences
using a single camera is developed
, which is useful

for a user to control his
movement in the virtual world by his legs. Twelve c
ontrol commands are
defined. The trajectories of the color marks placed on the shoes of the user are
used to determine the types of leg movement by a first
-
order Markov process.
And

the types of leg movement are encoded symbolically as input to Mealy
machi
nes to recognize the control command associated with a sequence of leg
movements.


C4.

Model
-
based analysis of free hand gestures from single images
by
projections of dot
-
matrix laser beams [15]
---


In this
study
, a new model
-
based system for analyzing free
hand gesture
from a single image by
CV

techniques is proposed.

T
he orientation and
position of the hand, and the joint angles of the fingers and the thumb
,

are
estimated separately
.

The estimated parameters are shown suitable for 3
-
D
hand gesture animation

by experiments. In addition, the applicability of the
proposed system is also demonstrated by a simple hand
-
gesture recognition
system.


C5.

D
etermination of head pose and facial expression from a single
perspective view by successive scaled orthographic appr
oximations

[16]
---


Human faces are the main organs for expressing human emotion. In this
study, a new iterative approach to analyzing the head pose and the facial
expression of a human face from a single image is proposed. The approach
is
based on an ext
ension of
the concept of successive scaled orthographic
approximations
.
The implementation of the proposed method is simple;
furthermore, no initial guess is required. The convergency property of the
proposed method has also been analyzed theoretically and

experimentally.


C6.

Knowledge
-
based
tracking and modeling of facial expressions

by stereo
vision techniques [17]
---


In this study, a
three
-
camera
stereo vision system for
real
-
time
knowledge
-
based tracking and modeling of facial expressions is
designed
.
Fe
ature points on a human face
are extracted, from which
related facial
parameters, including the head pose caused by rigid motion and the
displacements of feature points caused by local motion
, are computed
. And
computer graphics techniques
using a
muscle
-
b
ased face model are used to
影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



33

re
-
animate the facial expressions.
F
eature
-
point correspondences in
consecutive

image frame
s

are
computed

to calculate the 3
-
D coordinates of the
feature points.

And k
nowledge
-
based prediction is
employed

to speed up
the
search
of the corresponding feature points



V.

CV Techniques for Miscellaneous Applications

Because of the mature of the CV techniques and advance of hardware
technologies, more and more real
-
time CV
-
related applications can be developed. In
our laboratory, a main

emphasis is the study of using camera calibration techniques
for various 3D applications. Besides its original meaning, camera calibration means
localization of a camera, or in other words, determination of the position and
orientation of the camera. A so
lution of this problem leads to the decision of the
location or pose of an object or a vehicle on which the camera is mounted, creating
the possibility of various applications. We have developed in this aspect several
applications, including localization o
f a vehicle in indoor environments, a car moving
in outdoor roads, and a helicopter in the air. We have also used the camera calibration
technique to design a so
-
called 3D mouse that can be used for browsing VR
environments on the monitor screen. Finally,
we have developed a system for
scanning large
-
scale pictures.


D1.

Viewing Corridors as Parallelepipeds for Vehicle Localization [18]
---


When a vehicle navigates in a corridor inside a building, it is required to
locate the vehicle consecutively in the indoo
r environment. In this study, it
found that by viewing the corridor as a parallelepiped, the camera calibration
principle could be utilized to locate the camera on the vehicle, and so
determine the position and direction of the vehicle in the corridor. The

principle of vanishing line is employed in solving the camera calibration
problem. The result of this research can be used for indoor autonomous
vehicle guidance.


D2.

Using parallel line information for vision
-
based landmark location
estimation and an applic
ation to automatic helicopter landing [19]
---

The landmark of “H” on the roofs of hospitals or high buildings is used
as the landing site of helicopters. There exists abundant shape information,
mostly parallel lines, in the mark for use in camera calibra
tion, which in turn
leads to yield parameters representing the helicopter pose (position and
orientation). In this study, we have designed methods for computing analytic
影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



34

solutions to the helicopter pose determination problem. The result can be used
in auto
matic helicopter landing.


D3.

An inside
-
out vision
-
based 3
-
D mouse [20]
---

This is an interesting research in which we have designed a hand
-
held
mouse with a small camera inside, viewing out at a rectangular shape attached
on a side of the monitor screen of
a computer. The mouse can be used as a
browser in a 3D environment in the monitor. By analyzing the
perspective
-
transformed shape boundaries using camera calibration techniques,
the pose of the camera with respective to the monitor can be computed, which
o
ffers a basis for computing the position and orientation of the cursor
controlled by the mouse.


D4.

Estimation of moving vehicle locations using wheel shape information in
single 2
-
D lateral vehicle images by 3
-
D computer vision techniques [21]
---

The circul
ar wheel shape is useful for locating the position of a car
running on a street, as investigated in this research. A fast algorithm was
designed to detect the wheel shape in given car images. And the camera
calibration technique is used for wheel pose dete
rmination. The perspective
distortion of the circular wheel shape offers 3D information for computing the
parameters of the wheel pose. The result can be used in car driving automation
or assistance.


D5.

A
l
arge
-
s
cale
p
icture
s
canning
s
ystem based on automati
c image detection
and content corrections

[22]
---

We have designed a system in this study for scanning large
-
scale
pictures, which is constructed by combining an upright X
-
Y table with a
camera mounted on the moving head of the table. Because of the persp
ective
orientation of the picture with respective to the camera optical axis, 3D CV
techniques were applied here to correct the perspective distortion in the
scanning result. Other types of image distortion were also corrected. The
system is a useful tool
for scanning pictures with sizes difficult for regular
scanners to handle.



VI.

Conclusions

With the fast developments of CV and VR hardware and processing techniques,
影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



35

more and more intensive studies of CV and VR applications are just to begin although
lots o
f them have been developed. In the next century we will see the penetration of
various interesting applications of CV and VR into the human life. CV and VR
products will be part of consumer electronics, and will become common home
appliances. Studies of CV

and VR have no limit because friendly visual man
-
machine
interaction relies heavily on successful CV and VR developments. It is a view of our
research group that effective combination of CV and VR techniques is the key to the
success of human interfacing
in the future. Therefore, studies on vision
-
based VR will
still be emphasized in our researches in the Computer Vision Laboratory at National
Chiao Tung University.




References

[1]

K H. Chen and
W. H. Tsai

(1998). “Vision
-
based autonomous land vehicle
guidan
ce in outdoor environments using combined line and road following
techniques,”
Journal of Robotic Systems
, Vol. 14, No. 10, pp. 711
-
728, 1997.

[2]

K H. Chen and
W. H. Tsai

(1998). “Vision
-
based guidance for autonomous land
vehicle navigation in outdoor road en
vironments with static and moving cars,”
Proceedings of National Science Council (Part A), Republic of China
, Vol. 22,
No. 5, pp. 691
-
702.

[3]

K H. Chen and
W. H. Tsai
. “Vision
-
based obstacle detection and avoidance for
autonomous land vehicle navigation in ou
tdoor roads,” accepted and to appear
in
Automation in Construction
.

[4]

C. H. Ku and
W. H. Tsai

(1999). “Obstacle avoidance for autonomous land
vehicle navigation in indoor environments by quadratic classifier,”
IEEE
Transactions on Systems, Man, and Cyberneti
cs
---

Part B: Cybernetics
, Vol. 29,
No. 3, pp. 416
-
426.

[5]

C. H. Ku and
W. H. Tsai

(1999). “Smooth Autonomous Land Vehicle Navigation
in Indoor Environments by Person Following Using Sequential Pattern
Recognition,”
Journal of Robotic Systems
, Vol. 16, No. 5
, pp. 249
-
262.

[6]

C. H. Ku and
W. H. Tsai
, “Obstacle avoidance in person following for
vision
-
based autonomous land vehicle guidance using vehicle location
estimation and quadratic pattern classifier”, accepted and to appear in
IEEE
Transactions on Industrial

Electronics
.

[7]

G. Y. Chen and
W. H. Tsai

(1998). “An incremental
-
learning
-
by
-
navigation
影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



36

approach to vision
-
based autonomous land vehicle guidance in indoor
environments using vertical line information and multi
-
weighted generalized
Hough transform,”
IEEE Tr
ansactions on Systems, Man, and Cybernetics
-

Part
B: Cybernetics
, Vol. 28, No. 5, pp. 740
-
748, Oct. 1998.

[8]

G. Y. Chen and
W. H. Tsai

(1997/8). “Unsupervised learning of unexplored
environment by pushdown transducer for autonomous land vehicle navigation,”
Proceedings of 1997 Conference on Computer Vision, Graphics and Image
Processing
, Taichung, Taiwan, Republic of China, pp. 335
-
342.

[9]

G. Y. Chen and
W. H. Tsai

(1999) “A new approach to vision
-
based unsupervised
learning of unexplored indoor environment for
autonomous land vehicle
navigation,”
Robotics and Computer
-
Integrated Manufacturing
, Vol. 15, pp.
353
-
364.

[10]

G. Y. Chen and
W. H. Tsai

(1999/12). “An intelligent system for learning
environment models and guidance strategies for vision
-
based indoor
autonomou
s land vehicle navigation,”
Proceedings of National Computer
Symposium
, Taipei, Taiwan, Republic of China, pp. C159
-
C165.

[11]

H. P. Fan and
W. H. Tsai

(2000/12). “
3
-
D
i
ndoor
v
irtual environment modeling
via

vehicle navigation and multi
-
camera image data fusion
,” submitted.

[12]

R. C. Lo and
W. H. Tsai

(1995/9). "Position and orientation tracking using
perspective
-
transformation
-
invariant generalized Hough transform for virtual
reality applications,"
Proceedings of 1995 Workshop on Artificial Intelligence
,
Taipei, Ta
iwan, Republic of China, pp. 165
-
179.

[13]

C. C. Chang and
W. H. Tsai

(1999). “Reliable determination of object pose from
line features by hypothesis testing,”
IEEE Transactions on Pattern Analysis and
Machine Intelligence
, Vol. 21, No. 11, pp. 1235
-
1241.

[14]

C. C.

Chang and
W. H. Tsai

(2000) “Vision
-
based tracking and interpretation of
human leg movement for virtual reality applications,” accepted and to appear in
IEEE Transactions Circuits and Systems for Video Technology
.

[15]

C. C. Chang and
W. H. Tsai

(2000). “
Model
-
based analysis of free hand gestures
from single images
by projections of dot
-
matrix laser beams,” accepted and to
appear in
IEEE Transactions Circuits and Systems for Video Technology
.

[16]

C. C. Chang and
W. H. Tsai

(2000). “
Determination of Head Pose and Fa
cial
Expression from a Single Perspective View by Successive Scaled Orthographic
Approximations
,” submitted.

影像與識別

2000, Vol. 6, No.2 CV and VR Techniques



37

[17]

C. Y. Huang and
W. H. Tsai

(2000). “
Knowledge
-
Based
Tracking and Modeling
of Facial Expressions

by Stereo Vision Techniques,”
Proceedings of 2000
C
onference on Computer Vision, Graphics, and Image Processing
, Taipei,
Taiwan, Republic of China, Aug. 20
-
22, 2000.

[18]

Z. F. Yang and
W. H. Tsai

(1999). “Viewing Corridors as Parallelepipeds for
Vehicle Localization,”
IEEE Transactions on Industrial Electronic
s
, Vol. 46, No.
3, pp. 653
-
661.

[19]

Z. F. Yang and
W. H. Tsai

(1998). “Using parallel line information for
vision
-
based landmark location estimation and an application to automatic
helicopter landing,”
Robotics and Computer
-
Integrated Manufacturing
, Vol. 14,
p
p. 297
-
306.

[20]

Z. F. Yang and
W. H. Tsai

(1998/8). "An inside
-
out vision
-
based 3
-
D mouse,"
Proceedings of 1998 Conference on Computer Vision, Graphics, and Image
Processing
, Taipei, Taiwan, Republic of China.

[21]

C. C. Lai and
W. H. Tsai

(1999). “Estimation of mo
ving vehicle locations using
wheel shape information in single 2
-
D lateral vehicle images by 3
-
D computer
vision techniques,”
Robotics and Computer Integrated Manufacturing
, Vol. 15,
pp. 111
-
120.

[22]

K. W. Chou and
W. H. Tsai

(2000). “
A
l
arge
-
s
cale
p
icture
s
ca
nning
s
ystem based
on automatic image detection and content corrections
,” submitted.