|
|
Motion and Dynamics
|
|
|
|
Dynamic Scene Analysis and Understanding
|
|
|
|
o
Robust
Image Recognition and Matching
Developing
high-performance image matching and recognition techniques for multimedia
applications. The key problems to be
addressed are invariant feature extraction, description, and fast matching
from live video or archived image sources.
Targeted applications are mobile computing (e.g. handheld devices
for entertainment, advertising, and dynamic media sharing) and multimedia
data searching and retrieval [more
info].
o
Detection
and Tracking of Dynamic Events and Objects
Developing
algorithms for automatic analysis video imagery for detecting and tracking
of dynamic events and objects through the scene. Methods to estimate poses of targets from
single or multiple images [more info].
o 2D Image Motion Estimation
Tracking
natural scene features from video streams acquired by stationary or moving
cameras. The tracked scene features
can be used for recovering camera pose, image stabilization, image
registration, and creation of image mosaic from video sequences. We developed a robust tracking approach
and the software has been licensed to Rhythm & Hues, a major
special-effects production firm in Santa Monica CA. This software, named “Fastrack” by the firm, has been successfully used
for creation of special effects in such films as “X-Men 2”,
“Daredevil”, and “Dr. Seuss’ The Cat in the
Hat” - is capable of tracking hundreds of features from one frame to
another with sub-pixel accuracy in only a few seconds on a standard
personal computer, processes roughly 40 percent of movie shots without
having to provide extensive input to the computer, which is highly
appreciated by the effects artists of the firm [more info].
o Automatic Image Mosaicing
Automatic
creation of high quality image mosaic from image/video sequences based on
robust image motion estimation, where no assumption is made about the 3D
camera motion or the scene structure [more info].
o
Natural
Scenes Analysis Using Wavelet and Fractal Models
This
work focuses on the study of new methodology of natural scene analysis
using the wavelet and fractal theories.
The goal is to create new models and approaches which are efficient
for dealing with natural scene images, hence helping with image
representing of natural scenes, extraction semantic information, and
pattern analysis [more
info].
An
introduction talk
|
|
|
|
|
|
|
|
Motion Estimation and Navigation
|
|
|
|
o
Landmark-based
Camera Pose Tracking and Estimate
Automatic
estimation of camera 6DOF pose using artificial landmarks. Methods for detecting, recognizing, and
tracking landmarks in real-time [more info].
o Markerless Pose Tracking for Unprepared
Environments (Structure from Motion)
This
work addresses the case where neither camera motion, nor structure
information is available. The
approach uses naturally-occurring features tracking to recover relative
camera motion and scene structures (Structure From Motion) [more info].
o
Wide-area
Tracking Using Panoramic (omni-directional) Imaging Sensor
Camera pose tracking from
multiple video streams acquired by omni-directional imaging system that
provides a full 360-degree horizontal viewing [more info].
o
Point and Line Feature
Auto-calibration
Approach to simultaneously
estimate 6D pose of a camera and 3D parameters of tracked features (points
or lines) in the scene. An initial
camera pose estimate is computed from a set of known calibrated features. Other features (intentional fiducials (IF) or natural features (NF)), at initially
unknown positions, are tracked in the images produced as the camera
moves. The IF or NF 3D positions are
estimated (automatically calibrated) and their position estimates are used,
in turn, to estimate the pose of the camera. This computation iterates and converges
to produce both 6D camera pose and 3D IF or NF positions over a sequence of
images [more
info].
o Model-based Tracking and Visual
Navigation
Automatic
estimation of sensor pose motion using scene knowledge including: scene 3D
models (buildings, varied utility signs, facility tags or labels, etc.);
natural occurring features; and auto-calibrated scene knowledge
(runtime-calibrated features constrained to the model database) [more info].
o Hybrid Vision/INS/GPS
Hybrid
tracking technology attempt to compensate for the shortcomings of each
single technology by using multiple measurements to produce robust
results. We focus the research on
robust pose estimate by integrating
the computer vision, inertial, and GPS sensors for unprepared outdoor
environments. Methods to fuse complementarily
the diverse data resources [more info].
o Optimal Estimation and Filtering
Motion estimation is a
typical nonlinear estimate or probabilistic inference problem. The optimal solution to the nonlinear
estimate problem is given by the recursive Bayesian estimation technology. However, for the real-world problems, the
optimal Bayesian recursion is intractable and approximate solutions must be
used. Extended Kalman
Filter (EKF) is the most widely used estimation approach due to its
simplicity and tractability. The
EKF, however is based on a sub-optimal implementation of the recursive Bayesian
estimation framework applied to Gaussian random variables, this can
seriously affect the pose estimate accuracy or even lead to a divergent
solution. That is where we propose
to focus our attention to make the most significant impact [more info].
A summary
talk
|
|
|
|
|
|
|
|
Recognition and Human Motion
|
|
|
|
o
Face
Detection and Tacking
Real-time
detection of multiple faces and estimation of head pose for collaborative
workspace [more
info].
o Automatic Recognition of Human Face
and Expression
The
important of developing face recognition system by computer is not only in
the cognitive aspects, but also in its practical applications. Most existing systems only operate on
frontal view of face or facial profile, the aims of the research are to
develop automatic face recognition techniques that can work under less constrained
imaging conditions (varying pose and expressions). We approach methods including the
multiple view-based, synergetic neural network, and principal component
analysis, to deal with the complex problems of pose and expression varying
[more info].
o Real-time Landmark Detection and
Recognition
Accurate
landmark detection and recognition are crucial for the real-time pose
tracking system. We develop a
principal component analysis (PCA) based method extending from the above
face recognition work that can robustly detect and recognize the designed
B/W square landmarks (an
alphanumeric or symbol region embedded in their design facilitates unique fiducial recognition from sets of 50-100 different
symbols), achieving 28 fps and
allowing the viewpoint varying up to 70 degree in depth [more info].
o Computer Operation via Human Face
Orientation
We
approach a passive human head tracking and locating system through determining
the gaze of face in images. When a
head is captured by video camera, the face area and some facial features
are extracted automatically. To
determine the pose of input head, a vision approach is employed for
estimating and tracking the gaze direction of the face. The direction of face gaze gives a good
estimation of the normal direction of face plane; therefore it could be
used for determining the head pose and location with respect to the camera
[more info].
|
|
|
|
|
|
|
|
Geometry
and Appearance
|
|
|
|
Scene Reconstruction
and Modeling
|
|
|
|
o
Large-scale
Scene Modeling
Approaches
to rapidly create large-scale urban site models from LiDAR
and imagery, extract significant urban features (buildings, streetscapes),
and refine the reconstructed models [more info].
o Rapid Modeling of Dynamic Objects
Rapid
modeling of dynamic events and objects (people, vehicle) from video images,
allowing the objects to be visualized in 3D world [more info].
o 3D Scene Reconstruction From Stereo
Modeling
3D scene from stereo imagery. We
developed stereo matching approaches for different scenarios (man-made and
natural scenes), and produced a task-oriented stereo vision system for
automatic measure and reconstruction of SEM (Scanning Electronic
Microscope) imagery, and digital photogrammerty
(SPOT imagery). Approaches using
the wavelet to stereo matching: a wavelet zero-crossing matching, and a
wavelet phase-based stereo matcher [more info].
A
technical talk
|
|
|
|
|
|
|
|
Dynamic Texture
|
|
|
|
o
Real-time
Texturing from Video
Producing
model textures from real-time video streams captured by stationary or
moving cameras. By employing the
live video as texture resource, we are not only able to create an accurate
and photo-realistic appearance of the rendering scene, but also support
dynamic spatio-temporal update in the structure
of texture model, database, and rendering system [more info].
|
|
|
|
|
|
|
|
Data Fusion and Comprehension
|
|
|
|
o Dynamic Fusion and Visualization of Imagery and 3D Models
A technique combines all
manner of images, video, 3D models, and data in a coherent visualization
that supports varied media types and layers of abstraction. Our
work focuses on a novel approach, called augmented virtual environment
(AVE), fusing dynamic
imagery with 3D models. The AVE
provides a unique approach to visualize and comprehend multiple streams of
temporal data or images. Models are
used as a 3D substrate for the visualization of temporal imagery, providing
improved comprehension of scene activities.
Dynamic multi-texture projections enable real time
update and “painting” of scenes to reflect the most recent
visual scene data. The dynamic
controls, including viewpoint as well as image inclusion, blending, and
projection parameters, make for interactive real-time visualization of
events occurring over wide areas such as a campus, airport, security
infrastructure, military base, or battlefield [more info].
|
|
|
|
|
|
|
|
Graphics
Immersive Reality, HCI
|
|
|
|
Augmented/Virtual Reality and HCI
|
|
|
|
o
Mobile AR System and Collaboration
Developing
mobile AR that enables information annotation overlaying on real images
aiding on-site people to perform a verity of complex tasks. The effort targets the development of a
practical mobile AR system that provides accurate position aware computing
and information assistance to users throughout a facility or field [more info].
o Multi-sensors Fusion for Outdoor
Augmented Reality
This
project is for the development of multi-sensor fusion technology that specifically
targets outdoor augmented reality.
The aim is to produce technology and system that support and
integrate into systems such as NRL’s BARS
(Battlefield Augmented Reality System) system. The effort targets the development of a
complete system integrating current or near term technologies with the
needs of applications [more info].
o Augmented Reality for Space Flight
This
project deals with the production of annotated video (augmented reality)
and its use in NASA’s training and operations applications. With the assistance of Dr. Anthony Majoros at the Boeing Company, we propose to develop
and construct a prototype AR authoring system and evaluate its utility and
human performance benefits in terms of learning, recall, problem-solving,
and time to complete tasks [more info].
o Geospatial Registration of
Information for Dismounted Soldiers (GRIDS)
GRIDS
is an Augmented Reality system that can offer an intuitive, natural way for
dismounted soldiers to understand electronic information. The fundamental goal of GRIDS is to
implement geospatial registered information in an untethered
and unstructured environment [more info].
o Virtual Guidance for Aerospace
Targets Recognition and Visual Navigation
This
is a part of a telerobotics vision system which
is an integrated operation platform for telerobotics
vision, simulation, and manipulation research. The main work we carried out are on the
following problems: environment modeling, dynamic sensor calibration, and
virtual view generation aiding for visual navigation.
AR
technical talk
|
|
|
|
|
|
|
|
Rendering
and Visualization
|
|
|
|
o
4-D
Distributed Modeling and Visualization
Developing
methodologies and testbed required for dynamic
time-critical modeling and visualization systems, using augmented reality,
visualization, and 4D dynamic models of the environment, which incorporate
fast, robust, automatic, and accurate modeling of visual data, leading to
enhanced and improved tools and techniques, as well as compact
representations for time-space visualization of real world scenery [more info].
o Volume Rendering for 3D Virtual
Colonoscopy
The
motivation of this research is to employ advanced visualization techniques
for imaging and exploring the mucosal surface of human colon. A prototype system has been developed
including interactive navigation, fast rendering, and segmentation of the
colon, which allows the user to achieve both planned and guided navigations
inside the colon using the 3D image as a virtual environment –
“a virtual explorer” [more info].
|
|
|
|
|
|
|
|
Systems
and Applications
|
|
|
|
Surveillance and Situational Awareness
|
|
|
|
o
V-Sentinel:
A Novel System for Wide-area Situational Awareness
Developing
robust and intelligent systems for wide-area situational awareness is vital
to many applications including national security, transportation
management, environment monitoring, catastrophe response, and tactical
decision-making and military rations in battlefield environments. We employ our AVE framework to fuse and
present the aggregate sensing information so that it is easily browsed,
interpreted, and comprehended to support situational awareness and
decision-making processes [more info].
Overview
talk
|
|
|
|
|
|
|
|
Scene Reconstruction and Modeling
|
|
|
|
o
V-Urban3D:
A System for Rapid Creation of Large-scale 3D Urban Model
Developing
a state-of-the-art software solution that facilitates the rapid and
reliable creation of large-scale 3D urban models from range sensing
data. Based on the latest computer
vision, graphics, and modeling technologies, the V-Urban3D offers many
superexcellent features over existing systems, including true 3D urban site
reconstruction, rapid extraction and modeling buildings with a variety of
irregular shapes and rooftops, and accurate geometry model refinement, etc.
- all performed in an integrated environment [more
info].
|
|
|