Intruder
detection, face zooming.
MSc dissertation proposal 2010/2011
Target
Tracking with Pan-Tilt-Zoom Cameras
Introduction:
While most nowadays video surveillance
systems aim to document events happening in the field of view, too few of them
actually try to optimize the imaging quality
of the events. In many cases one is able to see that a number of robbers
broke into a store, but there is not enough zoom towards any of them allowing
the recognition of the face.
Detecting and tracking events, such as
intrusions, is therefore a desirable property for the modern video surveillance
systems. In this work we want precisely to research and develop some detection
and tracking methodologies.
Objectives:
The objectives of this work are twofold: (i)
construction of a user interface
allowing the manual selection of regions to track; (ii) programming the algorithm for tracking and zooming the selected area, encompassing a robust
controller of the pan and tilt axis of the camera. At the end is expected a
successful demonstration of the tracking of a person in a room, having
eventually obstacles as tables between the person and the camera.
Detailed description:
Using pan and tilt cameras to track people and mobile objects, e.g.
cars, is still a very active research-and-development topic mainly because of
the extremely wide variability one can observe in the shape of a person walking
or in the almost infinite options of clothing textures and colours. Even in the
case of cars, the situation is similar, as not only one finds a very wide variety
of shapes, as in addition one has to account the perspective effects associated
to the various poses of the cars.
Nevertheless, one finds that some camera manufacturers are already
evolving to provide some conveniences such as intrusion detection or face
tracking applications. For example Panasonic is proving a camera with a two
step zoom that places a detected person occupying a significant area allowing
than to be more identifiable (see Fig.2).
Fig.2 Commercial pan, tilt and zoom camera.
Notice the zooming after detecting the person.
http://www.youtube.com/watch?v=vQKGhIPiqs8
Even with these modern cameras, some aspects continue to be challenging,
such as visually locking on one person and avoiding influences from occluding
obstacles (e.g. people or vehicles crossing in front of the camera), or
handing-over the events found in one camera to a neighbouring one. Recognizing
that the automation of mobile cameras has still a long walk to be done, with
this work we look to find contributions to the state of the art.
The objective of this work is stated therefore briefly as the
implementation of an automatic tracker of people, or mobile objects, using
mobile (pan and tilt) cameras. More specifically, the work-proposal consists of
implementing image segmentation [Nascimento06], using representations adequate
for pan-tilt cameras [Sinha04, Vicente09], and selecting proper controlling
modalities.
The work is organized in the following main parts:
1) construction of an user interface allowing the
manual selection of the region to track: the system operator selects a salient
area in the image (e.g. part of the clothing of a person, or a pattern drawn in
a car) and the mobile camera tracks the selected region
2) programming of an algorithm for segmenting
and tracking the selected area
3) implementation of a robust controller, computing the pan, tilt and zoom control signals
of the camera
References:
[Nascimento06] "Performance Evaluation of Object Detection
Algorithms for Video Surveillance", Nascimento,
J.C.; Marques, J.S.; IEEE T-Multimedia, V8.4, 2006 pp:761
- 774.
[Sinha04] "Towards Calibrating a Pan-Tilt-Zoom Camera
Network",
[Vicente09] "Assessing Control Modalities Designed for Pan-Tilt
Surveillance Cameras", D. Vicente, J. Nascimento,
J. Gaspar, RecPad'09
[OpenCV] Open Computer Vision Library, http://sourceforge.net/projects/opencv/
Expected results:
At the end of the work the students will have enriched their knowledge
in:
* Computer vision
* Video Surveillance
Examples of expected demonstrations in simulated and/or real
environments:
* Tracking of a person walking in a room
* Tracking of the person having some obstacles in between (e.g. tables
or chairs)
Observations:
--
More MSc dissertation
proposals on Computer and Robot Vision in:
http://omni.isr.ist.utl.pt/~jag