�ݺ�ߣ

1. Robust Online Face Detection and Tracking 4th October 2016 Francesco Comaschi

2. Detection

3. Detection

4. Tracking

5. Applications Robot vision Video-surveillance Mobile applications Gaming

6. Challenge: accuracy Detect all visible faces Do not detect background regions

7. Challenge: robustness

8. Challenge: computational efficiency

9. Objective Accurate Robust Efficient

10. Detection Multi-face detection and tracking framework Overview Tracking Post processing Performance evaluation 1 2 3 4

11. Efficient face detection

12. Efficient face detection

13. Efficient face detection 1920 1080 ≈ 13.6 trillions of windows ≈ 13,589,544,960,000

14. Run-time adaptive Sliding window 2x speed-up No loss in accuracy 1920 1080 Efficient face detection

15. Robust face tracking Adaptive online tracker

16. Robust face tracking

23. Robust face tracking Adaptive online trackerOnline adaptation Drifting

24. Robust face tracking Stability Expensive Template-based face detector No adaptation False positives

25. Robust face tracking Tested on 15 videos Success rate over 70% Improvement over 9 state-of-the-art trackers

26. Online multi-face detection and tracking

32. Online multi-face detection and tracking Target removal

36. Online multi-face detection and tracking Target removal Target creation

37. Online multi-face detection and tracking Target removal Target creation

38. Online multi-face detection and tracking Target creation Target removal

39. Online multi-face detection and tracking Target creation Target removal Accuracy/efficiency balance

40. Online multi-face detection and tracking Target creation Target removal Accuracy/efficiency balance Fully automated system Multiple faces HD-resolution videos

41. Online multi-face detection and tracking Target creation Target removal Accuracy/efficiency balance Fully automated system Multiple faces HD-resolution videos 30% more accuracy 2x speed-up

42. Conclusions Detection Multi-face detection and tracking framework Tracking Post processing Performance evaluation 1 2 3 4

43. Thank you for your attention

�ݺ�ߣ

PhD_final_presentation_FC

More Related Content

PhD_final_presentation_FC

Editor's Notes