Comparison of Background Subtraction Methods for a Multimedia Applications

1
Comparison of Background Subtraction Methods for
a Multimedia Applications

• Fida EL BAF - PhD Student
• Thierry BOUWMANS - Doctor
• Bertrand VACHON Professor
• Laboratory of Applied Mathematics and Image
  (M.I.A.)
• University of La Rochelle - France
  
• IWSSIP 2007

2
Summary

• Description of Aqu_at_thèque application
• Identification Three steps
• Educational information
• Virtual Representation
• Difficulties in Identification step
• Background Subtraction Methods
• Qualitative comparison
• Quantitative comparision
• Results
• Perspectives

3
Description of Aqu_at_thèque application

• Projection of the images acquired by a camera
  movie directed to a tank
• of an aquarium
• Appoint a fish by a visitor
• System of recognition of fish species
• System of behavioral modeling

4
Identification Three steps

• Segmentation Extraction of regions corresponding
  to fishes in video sequences using a background
  subtraction method (Ex Sliding Median)
• Features extraction Color, texture, motion, etc
  (24 features)
• Classification of fishes using the extracted
  features (Recognition ratio 92)

Background Image
Foreground Detection Mask
Current Image
5
Educational information

• Indexed pedagogic cards Pictures and real video
  on the way of life, origin and the environment of
  the selected animal (e.g preys, predators).
• Example Pork fish

6
Virtual Representation

• Fishes can be represented in their real
  environment with their prey and predators in a
  virtual tank created by the user.
• Technical innovation Behavior engine.
• Pedagogical interest Nature Stability

7
Difficulties in Identification step

• Bootstrapping
• Global or local illumination changes
• Crossing
• Camouflage
• Foreground aperture
• Occultation

8
Background Subtraction Methods

• Single Gaussian (SG) Wren, Medioni, Zhao
• Mixture of Gaussians (MOG) Stauffer
• Parametric, 3 5 Gaussians
• Kernel Density Estimation (KDE) Elgammal
• Non Parametric Complexity O(NN)

Pavlidis 2001 - I. Pavlidis, V. Morellas, P.
Tsiamyrtzs, S. Harp, Urban surveillance
systems from the laboratory to the commercial
world, Proceedings of the IEEE, volume 89, no.
10, pages 1478 -1497, 2001
9
Qualitative comparison of Background Subtraction
Methods

• Qualitative Performance Evaluation
• Time consuming and memory consuming
• Detection KDE, MOG, SG, Median, Mean
• Neglected algorithm KDE

10
Quantitative comparison of Background Subtraction
Methods

• Quantitative Performance Evaluation
• Let A be a detected region and B be the
  corresponding ground truth, the similarity
  between A and B is defined by Li as
• It integrates the false positive and negative
  errors (FPR, FNR) in one measure
• Selected algorithm MOG offers the best
  compromise Time consuming/Detection

SG MOG (5 gaussians) MOG (3 gaussians)
FPR 0.0005 0.0068 0.007
FNR 0.6419 0.4001 0.3683
Detection Rate 0.3580 0.5998 0.6317
S(A,B) 0.3564 0.5710 0.6004
11
Results of Detection

• Aquatheque sequence

a) b)


c) d) e)
a) Image 201, b) Groundthruth, c) SG Foreground
Mask, d) MOG Foreground Mask and e) KDE
Foreground Mask
12
Perspectives

• Future research
• Unified framework for background subtraction
• Evaluation of Background subtraction in the
  context of Markerless Motion Capture system and
  Video Surveillance.
• Fuzzy model for background subtraction
  (Imprecision, uncertainty)

13
Bibliography

• Aquathèque
• Semani al 2002 D.Semani, C. Saint-Jean, C.
  Frélicot, T. Bouwmans, P. Courtellemont, Alive
  Fish species characterization for on line
  video-based recognition, Proceeding of the SPR
  2002, Windsor, Canada, pages 689--698, January
  2002 .
• Semani al 2002 D. Semani, T. Bouwmans, C.
  Frélicot, P. Courtellemont, Automatic Fish
  Recognition in Interactive Live Videos,
  Proceeding of the IVRCIA 2002, volume XIV,
  Orlando, Florida, pages 94-99, July 2002.
• Desfieux al 2002 J. Desfieux, L. Mascarilla,
  P. Courtellemont, "Interactivity and Educational
  Information in Virtual Real-Time 3D Videos",
  Proceeding of the IVRCIA 2002, 2002.
• Background Subtraction
• Wren 1997 C. Wren, A. Azarbayejani, T.
  Darrell, A. Pentland, Pfinder Real-Time
  Tracking of the Human Body, IEEE Transactions on
  Pattern Analysis and Machine Intelligence, Volume
  19, No. 7, pages 780 785 , July 1997.
• Stauffer 1999 C. Stauffer, "Adaptive
  background mixture models for real-time
  tracking", Proceedings IEEE Conference on
  Computer Vision and Pattern Recognition, pages
  246-252, 1999.
• Bowden al 2001 P. KaewTraKulPong, R. Bowden,
  "An Improved Adaptive Background Mixture Model
  for Real-time Tracking with Shadow Detection",
  Proceedings 2nd European Workshop on Advanced
  Video Based Surveillance Systems, AVBS 2001,
  Kingston, UK, September 2001.
• Elgammal 2000 A. Elgammal, Non-parametric
  model for Background Subtraction, in
  Proceedings of IEEE ICCV'99 FRAME RATE Workshop,
  1999
• Bo Han 2004 B.Han, D. Comaniciu, L.Davis,
  "Sequential kernel density approximation through
  mode propagation applications to background
  modeling, Proceeding Asian Conference on
  Computer Vision (ACCV 2004), 2004.
• Vasile Gui 2005 C. Ianasi, V. Gui, C. Toma, D.
  Pescaru, A Fast Algorithm for Background
  Tracking in Video Surveillance, Using Non
  parametric Kernel Density Estimation, Facta
  Universitatis, Series Electronics and
  Energetics, volume 18, No. 1, pages 127-144,
  April 2005.
• Li al 2004 L. Li, W. Huang, I.Y.H. Gu,
  Q. Tian, Statistical modeling of complex
  backgrounds for foreground object detection,
  IEEE Trans. Image Process, volume 13, pages
  14591472, 2004.
• Kim al 2006 K. Kim, T. Chalidabhongse, D.
  Harwood, L. Davis, "PDR Performance Evaluation
  Method for Foreground-Background Segmentation
  Algorithms", EURASIP Journal on Applied Signal
  Processing, 2006.