Transcript multicamera and distributed surveillance

http://imagelab.ing.unimo.it
Tutorial: multicamera and distributed
video surveillance
Prof. Rita Cucchiara
Università di Modena e Reggio Emilia, Italy
Third
ACM/IEEE
International
Conference on
Distributed
Smart Cameras
ICDSC 2009
30/08/2009
Como (Italy)
Distributed surveillance
 Problem of tracking and distributed
consistent labeling : Problem of
matching or recognizing objects
previously viewed by other cameras.
 Some constraints:
 Constraints on the motion models and
transition times
 Scene planarity for both overlapping
and not overlapping FOVs
 Constraints of recurrent paths
[70] V. Kettenker, R. Zabih Bayesian multi
camera surveillance CVPR 1999
[54] C.Stauffer, K.Tieu automated multicamera planar tracking
correspondence modeling cvpr
2003
Distributed surveillance (cont.)




Network of (smart) cameras; Not overlapped
FoVs; loosely coupled.
Problems of node communication
If moving cameras: problems of calibration and
tracking. The simultaneous localization and
tracking (SLAT) problem, to estimate both the
trajectory of the object and the poses of the
cameras.
Problem of color calibration
original
Look-up table
[71]Zoltan Safar, John Aa. Sørensen, Jianjun
Chen, and K°are J. Kristoffersen
MULTIMODAL WIRELESS NETWORKS:
DISTRIBUTED SURVEILLANCE WITH
MULTIPLE NODES Proc of ICASSP 2005
[72]Funiak, S.; Guestrin, C.; Paskin, M.;
Sukthankar, R.; Distributed localization of
networked cameras Int conf on
Information Processing in Sensor Networks,
2006.
Independent channels
Full matrix
Color calibration
 Methods:
 Linear transformation
 Independent channels
 AR   1 BR  1 
  

A


B


2
 G  2 G
 AB   3 BB  3 
  

 Full matrix ( M conmputed with LSQ)
 AR 
 
 AG   M
 AB 
 
 BR 
 
 BG 
 BB 
 
 Look-up table
 for non linear
 transformation
[73]Roullot, E., "A unifying framework for
color image calibration," 15th
International Conference on Systems,
Signals and Image Processing, 2008.
IWSSIP 2008, pp.97-100, 25-28 June
2008
[74]K. Yamamoto and J. U “Color
Calibration for Multi-Camera System
by using Color Pattern Board”
Technical Report MECSE-3-2006
Feature to match






Color (single / multiple)
Shape (geometrical ratios / spline / elliptical models)
Motion (speed, direction)
Gait (Fourier transform)
SIFT + , grey level co-occurrence matrix, Zernike
moments and some simple colour features
Polar color histogram + Shape
[75]Nicholas J. Redding, Julius Fabian
Ohmer1, Judd Kelly1 & Tristrom
Cooke Cross-Matching via Feature
Matching for Camera Handover with
Non-Overlapping Fields of View Proc.
Of DICTA2008
[76]Kang, Jinman; Cohen, Isaac; Medioni,
Gerard, "Persistent Objects Tracking
Across Multiple Non Overlapping
Cameras," IEEE Workshop on Motion
and Video Computing, 2005.
WACV/MOTIONS '05, vol.2, no.,
pp.112-119, Jan. 2005
Distributed Surveillance at ImageLab
 The problem: a people disappeared in the scene exiting from a
camera FoV, where can be detected in the future?
 1) tracking within a camera FoV multi hypothesis generation
 2) tracking in exit zones

3) Prediction into new cameras’ FoVs
 4) matching in the entering zones
[77]R. Vezzani, D. Baltieri, R.
Cucchiara, "Pathnodes
integration of standalone Particle
Filters for people tracking on
distributed surveillance systems"
in Proceedings of 25°
ICIAP2009, 2009
 Using Particle Filtering + Pathnodes


In computer graphic all the possible avatar positions are represented by nodes and the
connecting arcs refers to allowed paths. The sequence of visited nodes is called pathnodes.
A weight can be associated to each arc in order to give some measures on it, such as the
duration, the likelihood to be chosen with respect to other paths, and so on.
Weights can be defined or learned in a testing phase
Exploit the knowledge about the scene
 To avoid all-to-all matches, the tracking system can exploit the
knowledge about the scene




Preferential paths -> Pathnodes
Border line / exit zones
Physical constraints & Forbidden zones NVR
Temporal constraints
Tracking with pathnode
A possible path
between
Camera1 and Camera 4
Pathnodes lead particle diffusion
Results with PF and pathnodes
Single camera tracking:
Recall=90.27%
Precision=88.64%
Multicamera tracking
Recall=84.16%
Precision=80.00%
Example
Frame 431. a man #21 exits and
his particles are propagated
Frame 452 a person # 22 exits
too and also his particles are
propagated
Frame 471 a people is detected in
Camera #2 and the particles of
both # 21 and #22 are used but
the ones of #22 match and
person 22 is recognized