I work for
Siemens Corporate Technology and I work in Intelligent Signal Processing. I got
my Ph D in the Electrical and Computer
Engineering at
Current Research Topics:
Visual Tracking
Analysis of Human Actions in Video
Conversion of Semantic Concepts in Medical
Image Analysis to Vision Algorithms
Assistive Low Cost Technologies for Medical
Diagnosis and Treatment
Prior Research Work:
Computer Vision for Camera
Projector based systems
My current research focusses on computer vision problems that arise in
projector camera augmented reality systems. Recent research in projector-camera
systems has overcome many of the obstacles to deploying and using intelligent
displays for a wide range of applications. In parallel with these developments,
projector costs continue to decline with corresponding increase in resolution,
brightness and contrast ratio. In light of this trend, we are exploring the
unique capabilities that camera-projector systems can offer to intelligent
environments and ubiqutous computing. In a recent work, we addressed the
problem of real time contact detection in projected interface using an epipolar constraint between camera and projector. Here is a video of Ken
touching several virtual buttons on the screen and the system responding to it.
Another application we have developed is a smart bookshelf. The system
utilizes a camera pair and a projector to monitor the state of a real world
library shelf. As books are added to the shelf a foreground
detection algorithm which takes into account the projected information
yields new pixels in each view that are then verified using a planar parallax constraint
across both cameras to yield the book spine. An online process updates a
database containing the spine image, a color model of the spine, the location
of the spine in each view, and auxillary information
such as book title. Here is a
video of the system in action.
Prior to coming to UKY, I worked
on the problem of human identification using gait.
Continuous HMM based Approach
We proposed a view
based approach to recognize humans using gait. The width of the outer
contour of the binarized silhouette of a walking
person is chosen as the image feature. A set of exemplars that occur during a
walk cycle is chosen for each individual. Using these exemplars a lower
dimensional Frame to Exemplar Distance (FED) vector is generated. A continuous
HMM is trained using several such FED vector sequences. This methodology serves
to compactly capture structural and dynamic features that are unique to an individual. The statistical nature of the HMM renders
overall robustness to representation and recognition. Human identification
performance of the proposed scheme is illustrated using outdoor video
sequences.
Dynamic time warping based template matching approach :
In this work we proposed an appearance-based method for
gait recognition when the amount of training data is limited. The width of the
outer contour of the binary silhouette was used as the basic feature. Different
features were extracted from the width vector and dynamic time warping was used
to match gait sequences. Eigenanalysis of the width
vector shows that the gait signal evolves on a much lower (two-) dimensional
subspace and that gait possesses discriminative information. The method was
found to be reasonably robust to changes in speed. The contribution of dynamic information for
gait recognition, and
effect of viewing angle was also studied. It was also found that
the leg region by itself gave better recognition performance for one of the
databases. Analysis of component
level evidences to improve gait recognition performance was also performed.
In this work, we have proposed a method
for synthesizing arbitrary views of planar objects, and applying the
synthesized views for gait recognition when people are walking at any arbitrary
angle to the camera. Our method used a perspective projection model and an
optical flow based structure from motion model for estimating the azimuth angle
of the original view from monocular video data. Thereafter, a video sequence at
the new view was synthesized. The entire process was done in 2D, though 3D
structure of the scene played an implicit role. A simple, yet accurate, camera
calibration procedure was also proposed. Gait recognition on two
databases of people was reported using these synthesized views. Though the
method has been explained from the motivation of the gait recognition problem,
it has important applications in other areas too, like multimedia and video
processing. Video based rendering of planar dynamic scenes is one multimedia application we have worked on.
Feel free to write to me if you would like to discuss this
further
