1 / 17

Unified Crowd Segmentation

Unified Crowd Segmentation. P . Tu , T. Sebastian, G. Doreto , N. Krahnstoever , J. Rittscher , T. Yu ECCV 2008. Presenter: Ramin Mehran. Goal: Segmentation. Crowd Analysis Conventional Approach Segmentation Object Detection Tracking Analysis. Ideas: a Robust Algorithm. Approaches.

melia
Télécharger la présentation

Unified Crowd Segmentation

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. Unified Crowd Segmentation P. Tu, T. Sebastian, G. Doreto, N. Krahnstoever, J. Rittscher, T. Yu ECCV 2008 Presenter: RaminMehran

  2. Goal: Segmentation • Crowd Analysis • Conventional Approach • Segmentation • Object Detection • Tracking • Analysis CV Lab @ UCF - Presenter: RaminMehran

  3. Ideas: a Robust Algorithm CV Lab @ UCF - Presenter: RaminMehran

  4. Approaches • Bottom up • Low level feature grouping • Whole body classifiers • Top down • Background segmentation • BOTH! CV Lab @ UCF - Presenter: Ramin Mehran

  5. Algorithm Overview CV Lab @ UCF - Presenter: Ramin Mehran

  6. Affinity of Patches to Hypotheses • Head-Shoulder Detection • Grouping Patches • Assigning Patches to Heads Ck gk(zi) zi zj gk(zi, zj) • Top Down  Head/Shoulder Detection • Bottom up  Patch Affinities CV Lab @ UCF - Presenter: Ramin Mehran

  7. Maximum Likelihood How likely is the assignment a patch to a head E-M Steps Is it likely near the head? Is it like neighboring patches? Assignment indicator Consistency Affinity CV Lab @ UCF - Presenter: Ramin Mehran

  8. E-Step k Increase the expectations i CV Lab @ UCF - Presenter: Ramin Mehran

  9. E-Step affect CV Lab @ UCF - Presenter: Ramin Mehran

  10. M-Step • Enforce Consistency • Where is the head? • Contiguous patches • Path from Patches to the Head • Dynamic Programming • Shortest low cost path Ck Cm • Inconsistent patches • No Contribution to V (assignment) CV Lab @ UCF - Presenter: Ramin Mehran

  11. Head/Shoulder Body Detection Strong Classifier out of Weak Classifiers Strong Classifier (SC(s,z)) Orientation Orientation Orientation Orientation Aggregation of weak Classifiers CV Lab @ UCF - Presenter: Ramin Mehran

  12. Weak Classifier R Ck Needs Camera Calibration zi • Weak Classifier • Ratio of the intersection of patch and the region • Sobel Orientation CV Lab @ UCF - Presenter: Ramin Mehran

  13. What’s the Affinity? patch Head and body bounding box Patch to Head/Shoulder Affinity • Patch to Patch Affinity • Histogram matching of colors • Motion matching of flow CV Lab @ UCF - Presenter: Ramin Mehran

  14. Patch Size • Camera Calibration • Width of the average pedestrian • Large Patches • Occlusion Handling • Small Patches • Discrimination of correct and incorrect assigments CV Lab @ UCF - Presenter: Ramin Mehran

  15. Sample Results CV Lab @ UCF - Presenter: Ramin Mehran

  16. Results Compared to Histogram of Oriented Gradient method Superior mostly because of handling occlusions CV Lab @ UCF - Presenter: Ramin Mehran

  17. Thank you! CV Lab @ UCF - Presenter: Ramin Mehran

More Related