1 / 39

Two Examples Of Indoor And Outdoor Surveillance Systems: Motivation, Design, And Testing

Two Examples Of Indoor And Outdoor Surveillance Systems: Motivation, Design, And Testing. Ioannis Pavlidis Vassilios Morellas Honeywell Laboratories. Graduate Seminar in CIS Video Processing and Mining CIS 750 – Spring 2003. Presented by: Ken Gorman. Agenda.

chick
Télécharger la présentation

Two Examples Of Indoor And Outdoor Surveillance Systems: Motivation, Design, And Testing

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Two Examples Of Indoor And OutdoorSurveillance Systems:Motivation, Design, And Testing Ioannis Pavlidis Vassilios Morellas Honeywell Laboratories

  2. Graduate Seminar in CISVideo Processing and MiningCIS 750 – Spring 2003 Presented by: Ken Gorman

  3. Agenda • CCN – Cooperative Camera Network • DETER - Detection of Events for Threat Evaluation and Recognition

  4. Cooperative Camera Network (CCN) • Network of cooperating cameras • Controlled by computer vision software • Features: • mechanism for counting the people present in various parts of the building • An automatic or semi-automatic mechanism for tagging people. • Report tagged individuals whereabouts whenever they are within the field of view

  5. Major Components • COTS Hardware & Software Set-Up • Change Detection • Counting People • Tracking People

  6. State of the Art • Active badges • small, electronic devices worn by people • transmit an ID signal to receivers placed around the building • ID signal corresponds to the identity of the badge’s wearer • received signals are used to compute the wearer’s location

  7. Badge Examples • Infrared-transmitting badges at Olivetti Research and Xerox PARC, • Olivetti ultrasonic badges at AT&T Laboratories in Cambridge, UK • Radio frequency tags from PinPoint • Wired and unwired motion trackers from • Ascension Technology • Polhemus

  8. Disadvantages • Consumers unwilling to wear badges • Cumbersome

  9. Alternatives • ????

  10. Cameras • Pro – Leaves users unencumbered • Cons – Not as reliable as badge methods

  11. Camera Arrangement • Overlapping Fields of View

  12. Fundamentals • Imaging Technologies for Surveillance Systems • Image Segmentation • Tracking Mechanism • Multi-Camera Fusion • Threat Assessment

  13. Multi-Normal Pixel Representation

  14. Initialization • Goal - provide statistically valid values for the pixels corresponding to the scene. • Starting point for the dynamic process of foreground and background awareness

  15. Initialization • Methods Used: • K-Means – better for plazas and malls • Expectation-Maximization – better for changing weather conditions [1]

  16. Image Segmentation • Each pixel is considered as a mixture of five time-varying trivariate normal distributions

  17. Image Segmentation • The term represents a trivariate Normal distribution with vector mean and variance-covariance matrix

  18. Image Segmentation • The distributions are trivariate to account for the three component colors (red, green, and blue) of each pixel in the general case of a color camera.

  19. Image Segmentation • For simplification, the variance-covariance matrix is assumed to be diagonal with xR,xG,xB , having identical variance within each Normal component, but not across all components

  20. Update Cycle • Distributions are ordered based upon their weights. • Member of a Distribution • Distribution is in background or foreground • Jeffreys [2] algorithm used for “matching” pixel to distribution • Distributions are Updated

  21. Matching • We use the Jeffreys divergence measure (J) to determine whether the incoming data point belongs to one of the existing distributions • The Jeffreys number measures how unlikely it is that one distribution (g) was drawn from the population represented by the other (f) • K* - prespecified cut-off value

  22. Update – Match Found • Incoming pixel state is labeled either background or foreground • All the parameters of the matched distribution are updated according to the method of moments • Only the weights of the other distributions are updated

  23. Update – No Match • Incoming pixel state is labeled foreground • Last distribution in the ordered list is replaced • All the parameters of the new distribution are updated • Only the weights of the other distributions are updated

  24. Jeffrey’s Algorithm • Jeffreys number measures how unlikely it is that one distribution (g) was drawn from the population represented by the other (f).

  25. Broken Clouds • Preferential, in order No Proference

  26. Segmentation of Moving Objects • The form of some of the distributions could change • Some of the foreground states could revert to back-ground and vice versa. • One of the existing distributions could be dropped and replaced with a new distribution.

  27. Predictive Tracking • On-line segmentation of foreground pixels • Calculation of blob centroids • Multiple-Hypotheses Tracking Algorithm

  28. Multiple-Hypotheses Tracking • Recursive Bayesian probabilistic procedure • Does NOT commit early to trajectory

  29. Multiple Hypothesis Tracking

  30. Multiple Hypothesis Tracking • Kalman filtering prediction based on constant velocity models • K-best hypothesis trajectory tree generation, pruning and merging • Bayesian probability calculations for matching input data to track hypothesis • See references [5] and [6] for exact algorithm

  31. Multi-Camera Fusion • Monitoring of large areas can only be accomplished using multiple cameras • Panoramic View is created by fusing individual camera FOVs • Object motion registered against a global coordinate system

  32. Multi-Camera Fusion • Optimal Coverage Scheme is created • Minimal use of cameras to minimize cost

  33. Multi-Camera Fusion • Compute HOMOGRAPHY matrix H between two cameras based on CoG of moving objects appearing in the overlapping areas of the two fields of view • Requirement: Information exchange between respective computers (e.g., pixel intensity data and CoG of moving objects in pixel coordinates)

  34. Homography Matrices Computation • Least Squares Method • Very popular • Relatively simple • Defined in Reference [6]

  35. Homography Matrix • Used Kanatani Method • Based on a statistical optimization theory for geometric computer vision • Cures the deficiencies exhibited by Least-Squares

  36. Kanatani Method • Epipolar constraint may be violated by various noise sources due to the statistical nature of the imaging problem

  37. Multi-Camera Fusion • O1 andO2 are Optical Centers • P(x,y,z) is a point in the scene that falls in the common area between the two camera • Vector O1p, O2q, and O1O2 are co-planar

  38. References • Paolo Remagnino , et al (Editors). Video-Based Surveillance Systems: Computer Vision and Distributed Processing. Kluwer Academic Publishers, 2002. • http://www.htc.honeywell.com/projects/deter/ • [1] - A. P. Dempster, N. M. Laird, and D. B. Rubin, “Maximum likelihood from incomplete data via the em algorithm (with discussion),” J. Roy. Stat. Soc. B, vol. 39, pp. 1–38, 1977. • [2] - J. Lin, “Divergence measures based on the Shannon entropy,” IEEE Trans. Inform. Theory, vol. 37, pp. 145–151, Jan. 1991. • [3] - C. Stauer and W.E.L. Grimson, “Adaptive background mixture models for real-time tracking," in Proceedings 1999 IEEE Conference on Computer Vision and Pattern Recognition, Fort Collins, CO, June 23-25 1999, vol. 2, pp. 246-252.

  39. References (cont.) • [4] D. B. Reid, “An algorithm for tracking multiple targets”, IEEE Transactions on Automatic Control, vol. 24, pp. 843{854, 1979. • [5] I. J. Cox and S. L. Hingorani, “An efficient implementation of Reid's multiple hypothesis tracking algorithm and its evaluation for the purpose of visual tracking," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 18, no. 2, pp. 138-150, 1996. • [6] L. Lee, R. Romano, and G. Stein, “Monitoring activities from multiple video streams: Establishing a common coordinate frame," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 22, no. 8, pp. 758 { 767, 2000.

More Related