of Human Identification Systems for Visual Surveillance
in Security-Sensitive Environments
- Develop human identification technologies for surveillance
and security systems, which combine advantages of biometrics
(face recognition, gait recognition) and human motion analysis
(motion analysis of human body parts, human motion tracking)
to recognize a person in a crowd or singularly.
- Design surveillance systems to identify individuals at
a distance in security-sensitive environments such as banks
- Develop and evaluate algorithms for face recognition over
image sequences, gait recognition, human movement tracking,
and human pose and motion analysis.
- Synchronize the 4 different stages of human identification
procedure, i.e., face recognition, gait recognition, human
movement tracking, and human pose and motion analysis.
- Design a hierarchical decision-making scheme for identity
recognition based on the results of the 4 different stages.
- Face recognition over image sequences: Develop face recognition
algorithms using the advantage of additional information
obtained in image sequences, which can be used for either
tracking persons over image sequences or identify a person.
The developed algorithms can also estimate head pose, remove
sunglasses and scarf, and recognize facial expressions.
- Gait recognition: Develop gait recognition algorithms
based on silhouette analysis from videos, which aims to
identify individuals by the way they walk. These algorithms
should produce stable and accurate results in the presence
of variations such as shoe-wear type, weight carried, camera
viewpoint, and data collecting time.
- Human movement tracking: Develop model-based tracking
algorithms for moving humans by focusing on the motion of
the whole body, which can be used as a guide for predicting
and estimating the movement of body parts (such as head,
neck, shoulders, elbows, wrists, torso, hips, knees, ankles).
The human movement tracking algorithms should segment rapidly
changing scenes in natural environments involving non-rigid
motion and occlusion, and track humans either from a single
view or from multiple perspectives.
- Human body pose and motion analysis: Qualitatively and
quantitatively analyze human body pose and motion by using
high level reasoning and kinematic modeling, respectively.
High level reasoning can discriminate the relative position
and motion of body parts without requiring the precise location
of joints or edges. Kinematic modeling can detect, track
and identify articulated objects (humans) by modeling the
underlying dynamics of human locomotion and biomechanics.
2D and 3D methods will be designed for interpretation of
human body structure by focusing on motion estimation of
the joints of body parts.
- Synchronization of the 4 different stages: Design a real-time
dynamic framework to evaluate which stages will work well
in the current frame streaming, to conduct the selected
stages and to produce the estimation results for the current
- Hierarchical decision-making scheme for identity recognition:
Design a dynamic multilayered identity estimation method
based on the estimation results from the different stages
and from the different frame streaming.
- Database collection: Build up a number of databases to
aid in the evaluation of algorithms for face recognition,
gait recognition, human movement tracking, and human pose
and motion analysis.
The computer system will be able to convert the pixels
from the camera into data sequences that would then
be assigned to the parts of the body that would have
specific fields assigned , such as head, neck, chest,
arms etc. The pattern for each area would be further
broken down on an in-depth ( micro ) mapping system,
then converted again to pick out detail ( micro ) movements
of the joint, muscle, skin, the system would have to
be able to virtually remove clothing and accessories
and (peer) beneath. This could be accomplished with
a form of ( x ray, laser ) scanning. For the face ,
should it be covered by e.g.: sunglasses, facial hair,
or plastic surgery ( which would not necessarily change
the shape or size of the skeletal form) .
To match a body profile in 3D using software that will
allow downloading a moving picture of the subject and
divide the entire body using the body's joints as markers
focusing on the movement of that particular part in
depth using a grid method. With the transfer of all
the points and movements from the grid to an overall
movement structure, this will create a format for this
individual which would be unique.