Omnidirectional and Pan-Tilt-Zoom surveillance
MSc dissertation proposal 2008/2009
Surveillance
of Public Spaces using
Omnidirectional and Pan-Tilt-Zoom Cameras
Introduction:
Standard surveillance systems installed in public spaces, often require
a multitude of conventional cameras to cover most of the space under
observation. Despite the low-cost of nowadays cameras, there are huge amounts
of cabling and maintenance costs associated with these systems. Furthermore, in
the surveillance operation center, most of the
cameras are unattended due to the very high costs of having humans watching the
resulting video-imaging. Hence, in this work we propose using novel
non-standard types of cameras do deal with this issue: omnidirectional and
pan-tilt-zoom cameras. Omnidirectional cameras are able to cover very wide
spaces at the cost of reduced resolution. On the contrary, Pan-Tilt-Zoom
cameras are able to move to interesting regions in the visual space, thus being
able to inspect them with high resolution but a narrow view field. The
combination of these complementary types of cameras is able to bring enhanced
fields of view and resolutions, and allow constructing effectively the required
automatic surveillance needs, with a reduced number of cameras, cabling and
computational resources.
Objectives:
The goals of this work proposal are two-fold: (i)
modelling pan-tilt-zoom and omnidirectional cameras, (ii) designing and
implementing computer vision algorithms for surveillance applications such as
detection and tracking of intrusions.
Detailed description:
In this work we propose using novel omnidirectional and pan-tilt-zoom
cameras. Omnidirectional cameras see in all the 360deg azimuthal
field-of-view and are thus interesting for surveillance because of the true omni-awareness: conventional limited
field-of-view cameras observe just part of the scene at a time. Pan-tilt-zoom
cameras allow selecting the field-of-view and therefore give the highest
quality image resolution. Cooperative combining both types of cameras
allows obtaining the best properties of both systems.
In this work-proposal we consider two principal aspects: (i) modelling omnidirectional and pan-tilt-zoom cameras (ii)
using these cameras combined in a network for surveillance applications.
Modelling omnidirectional and pan-tilt-zoom cameras is in general a challenging
task due to the wide variety of the existing designs (many of them involving
non-linear optics). In this work we consider just cameras having simple
geometries as the ones with constant resolution and/or a single effective
view-point.
Networking cameras encompasses creating the communications
infrastructure and calibrating the complete system. For example, in order to
perform target tracking in a camera network, it is necessary to measure, or
estimate, the relationships among the fields-of-view of the overlapping
cameras. We consider a network of cameras having largely redundant
fields-of-view and doing the calibration based on the large superpositions
of the background or by cooperatively observing moving objects.
Work-proposal detailed steps:
(i) Modelling omnidirectional and
pan-tilt-zoom cameras - this involves estimating or measuring some camera
parameters and building environmental background representations [Sinha04]. Given
the geometric model of an omnidirectional camera one can build virtual
cameras, i.e. cameras resembling standard pan-tilt-zoom cameras, by
re-locating pixels. The modelling of the pan-titl-zoom
cameras allows collaborative observation of the same scene point, as the
coordinates of the point in one camera can be translated to another one given
the camera models and some scene priors. Note that translating observations
from an omnidirectional to a pan-tilt-zoom camera is an important service as
conventional pan-tilt images, with the correct zooming, are much simpler to
interpret by humans.
(ii) Surveillance applications - given the background
representations, one main purpose is to detect scene intrusions, moving objects
or generic events in the scene [Boult99, Hall05, Nascimento06].
Other surveillance services include doing statistics as bidirectional
counting of pedestrian passing through a "gate", reports upon request
how many people are gathered in its field of view, counts the number of times
people have dwelt for longer than a pre-set period, the most frequent path in a
shopping mall (average paths, dwell times), etc.
Hardware and software details:
The surveillance cameras will encompass one or more omnidirectional and
pan-tilt-zoom cameras. The software builds upon networking software
[YARP], which provides for example ports for image acquisition from cameras. At
the end of the work, the produced applications will have ports reporting events
and situation assessment detected at the camera network.
References:
[Sinha04]
[Boult99] "Frame-Rate Omnidirectional Surveillance & Tracking
of Camouflaged and Occluded Targets", T.E. Boult
R. Micheals, X. Gao, P.
Lewis, C.Power, W. Yin and A. Erkan,
in Proc. of the IEEE Workshop on Visual Surveillance, June 1999.
[Hall05] "Comparison of target detection algorithms using adaptive
background models", D. Hall, J. Nascimento, P. Ribeiro, E. Andrade, P. Moreno, S. Pesnel,
T. List, R. Emonet, R.B. Fisher, J. Santos Victor, J.
Crowley, Int. workshop on Performance evaluation of Tracking and Surveillance -
Oct 2005
[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.
[YARP] "Yet Another Robot Platform",
http://eris.liralab.it/yarp/
Requirements (grades, required courses, etc):
-
Expected results:
At the end of the work, the students will have enriched their experience
in computer vision applied to video surveillance setups. In particular are
expected to develop and assess:
- geometric modelling methods for
omnidirectional, pan-tilt-zoom, networked cameras;
- algorithms for intrusion detection and
tracking.
Place for conducting the work-proposal:
ISR / IST
More MSc dissertation
proposals on Computer and Robot Vision in:
http://omni.isr.ist.utl.pt/~jag