My research mainly focus on computer vision, image processing, data acquisition, computer interface, field bus transfer protocol, optoelectronic inspection, control system simulation, distributed system and human computer interaction.
re-identification for security surveillacen
This project aims to recognize
suspicious target from multiple cameras without overlapping
field of view. This is a very challenging problem for
security surveillance. The most challenging part of the
project is finding discriminative features from the target
since the videos are at low resolution with periodic jitter
and illumination condition and angle of view are different
for every camera. It is also very crucial yet difficult to
design a human detector and tracker that are efficient and
errorless. Currently we have finished the algorithm design
of target tagging and precisely locating, and we are working
on finding a good descriptor as well as an effective
online-discriminative feature selection algorithm.
calibration and dynamic correction
for pan-tilt-zoom cameras
Most modern wide-area camera
surveillance networks make extensive use of PTZ cameras.
However, since such cameras are in constant motion,
accumulated errors from imprecise mechanisms, random noise,
and power cycling render any calibration in absolute world
coordinates useless after many hours of continuous
operation. We propose a method based on the automatic
detection and matching of scene features to maintain the
calibration of a PTZ camer after its initial calibration, so
that when a user directs the camera to given (pan, tilt,
zoom) coordinates, the same field of view is always
attained. Consequently, the absolute PTZ coordinates for a
given camera can be trusted to be accurate, leading to
improved performance on important tasks like the 3D
triangulation of a tracked target.
counter-flow detection for airport
Counterflow detection is an important
problem in security surveillance applications, such as
one-way passages from secure areas to non-secure areas in
airports. We describe a solution to detecting counterflow in
a fixed camera that leverages the detection of scene
features to improve performance. We also introduced a
classifier that simultaneously identifies scene feature
trajectories and detects counterflow motion. Explicitly
including scene features in the flow classifier improves
robustness, reduces the mixing of foreground and background
in point tracks, and allows the detection of jitter frames.
real-Time airport security
checkpoint surveillance system
This project aims to design an airport security
checkpoint surveillance system using a camera network.
The system tracks the movement of each passenger and
carry-on bag, continuously maintains the association
between bags and passengers, and verifies that
passengers leave the checkpoint with the correct bags.
We investigat methods for calibrating the camera
network and tracking the many moving objects in the
environment. We define a state machine for bag
tracking and association, dividing the imaged area
into several semantically meaningful regions.
visual tracking system for EMPAC
This project aims to design a visual tracking system for a 2500 ft2 black box studio at RPI's new Experimental Media and Performing Arts Center, enabling real-time multi-person tracking for a wide range of research applications. The first goal of the project is to provide the user with real-time estimates of the (x,y) floor positions of all the people in the studio. Subsequent research goals include highly efficient intrinsic and extrinsic calibration of the camera array, automatic multi-person tracking algorithms under complex conditions such as dimly lit or crowded scenes, and interactions with other sensors and actuators such as wall-mounted pan-tilt-zoom cameras, projectors, lights and audio recorders. We expect the system to became a valuable research testbed and infrastructure addition to EMPAC that will improve the utility of the studio space and promote interaction between scientists, engineers, and artists.