1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan...

44
1 Hierarchical Part- Hierarchical Part- Based Human Body Pose Based Human Body Pose Estimation Estimation * Ram anan Navaratnam * Arasanathan Thayananthan † Prof. Phil Torr * Prof. Roberto Cipolla * University Of Cambridge † Oxford Brookes University

Transcript of 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan...

Page 1: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

1

Hierarchical Part-Based Hierarchical Part-Based Human Body Pose EstimationHuman Body Pose Estimation

* Ramanan Navaratnam

* Arasanathan Thayananthan

† Prof. Phil Torr

* Prof. Roberto Cipolla

* University Of Cambridge † Oxford Brookes University

Page 2: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

2

IntroductionIntroduction

Input

Page 3: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

3

IntroductionIntroduction

Input Output

Page 4: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

4

OverviewOverview

1. Motivation

2. Hierarchical parts

3. Template search

4. Pose estimation in a single frame

5. Temporal smoothing

6. Summary & Future work

Page 5: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

5

OverviewOverview

1. Problem motivation ???

2. Hierarchical parts

3. Template search

4. Pose estimation in a single frame

5. Temporal smoothing

6. Summary & Future work

Page 6: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

6

OverviewOverview

1. Problem motivation ???

2. Hierarchical parts

3. Template search

4. Pose estimation in a single frame

5. Temporal smoothing

6. Summary & Future work

Page 7: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

7

OverviewOverview

1. Problem motivation ???

2. Hierarchical parts

3. Template search

4. Pose estimation in a single frame

5. Temporal smoothing

6. Summary & Future work

Page 8: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

8

MotivationMotivation ‘Real-time Object Detection for Smart Vehicles’

– D. M. Gavrila & V. Philomin (ICCV 1999)

‘Filtering using a tree-based estimator’ – Stenger et.al. (ICCV 2003)

Page 9: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

9

MotivationMotivation

Exponential increase of templates with dimensions

‘Real-time Object Detection for Smart Vehicles’ – D. M. Gavrila & V. Philomin (ICCV 1999)

‘Filtering using a tree-based estimator’ – Stenger et.al. (ICCV 2003)

Page 10: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

10

MotivationMotivation ‘Pictorial Structures for Object Recognition’

– P. Felzenszwalb & D. Huttenlocher (IJCV 2005)

‘Human upper body pose estimation in static images’ – M.W. Lee & I. Cohen (ECCV 2004)

Page 11: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

11

MotivationMotivation

Part based approach Assembling parts together is complex

‘Pictorial Structures for Object Recognition’ – P. Felzenszwalb & D. Huttenlocher (IJCV 2005)

‘Human upper body pose estimation in static images’ – M.W. Lee & I. Cohen (ECCV 2004)

Page 12: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

12

MotivationMotivation ‘Automatic Annotation of Everyday Movements’

– D. Ramanan & D. A. Forsyth (NIPS 2003)

‘3-D model-based tracking of humans in action:a multi-view approach’ – D. M. Gavrila & L. S. Davis (CVPR 1996)

Page 13: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

13

MotivationMotivation ‘Automatic Annotation of Everyday Movements’

– D. Ramanan & D. A. Forsyth (NIPS 2003)

‘3-D model-based tracking of humans in action:a multi-view approach’ – D. M. Gavrila & L. S. Davis (CVPR 1996)

‘State space decomposition’

Page 14: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

14

Hierarchical PartsHierarchical Parts

Page 15: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

15

Hierarchical PartsHierarchical Parts

Page 16: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

16

Hierarchical PartsHierarchical Parts

Page 17: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

17

Hierarchical PartsHierarchical Parts

Page 18: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

18

Hierarchical PartsHierarchical PartsConditional prior

p(x i

/xpar ent(i))

Spatial dimensions (translation)Joint Angles

Page 19: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

19

Hierarchical PartsHierarchical Parts

Head and torsoUpper armLower Arm

False Positive

Tru

e Po

siti

ve

Page 20: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

20

Hierarchical PartsHierarchical PartsDetection Threshold = 0.81

Detections

Head and torso

6156

Part

Page 21: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

21

Hierarchical PartsHierarchical PartsDetection Threshold = 0.81

Detections

Head and torso

6156

13 199 44 993

Part

Lower arm

Page 22: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

22

Template SearchTemplate Search

Page 23: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

23

Template SearchTemplate Search

Page 24: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

24

Template SearchTemplate Search

Page 25: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

25

Template SearchTemplate Search

Features Chamfer distance Appearance

Page 26: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

26

Template SearchTemplate Search

Features Chamfer distance Appearance

Page 27: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

27

Template SearchTemplate Search

Features Chamfer distance Appearance

Page 28: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

28

Template SearchTemplate Search

Features Chamfer distance Appearance

Page 29: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

29

Template SearchTemplate Search

Features Chamfer distance Appearance

Page 30: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

30

Template SearchTemplate Search

Features Chamfer distance Appearance

Page 31: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

31

Template SearchTemplate Search

Features Chamfer distance Appearance

Page 32: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

32

Template SearchTemplate Search

Features Chamfer distance Appearance

Page 33: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

33

Template SearchTemplate Search

Features Chamfer distance Appearance

Page 34: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

34

Template SearchTemplate Search

Learning Appearance Match ‘T’ pose based on edge likelihood only in initial

frames

Update 3D histograms in RGB space that approximates P(RGB/part) and P(RGB)

Page 35: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

35

Pose Estimation in a Single FramePose Estimation in a Single Frame

Page 36: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

36

Pose Estimation in a Single FramePose Estimation in a Single Frame

Page 37: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

37

Pose Estimation in a Single FramePose Estimation in a Single Frame

Page 38: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

38

Temporal SmoothingTemporal Smoothing

HMM

Page 39: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

39

Temporal SmoothingTemporal Smoothing

HMMT = t

Page 40: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

40

Temporal SmoothingTemporal Smoothing

HMM

Viterbi back tracking

Page 41: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

41

Temporal SmoothingTemporal Smoothing

Viterbi back tracking

Page 42: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

42

Temporal SmoothingTemporal Smoothing

Page 43: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

43

Summary & Future workSummary & Future work

Summary

Realtime process (unoptimized code at 1Hz, 2.4 Ghz IG RAM)

3D pose

Automatic initialisation and recovery from failure

Page 44: 1 Hierarchical Part-Based Human Body Pose Estimation * Ramanan Navaratnam * Arasanathan Thayananthan Prof. Phil Torr * Prof. Roberto Cipolla * University.

44

Summary & Future workSummary & Future work

Summary

Realtime process (unoptimized code at 1Hz, 2.4 Ghz IG RAM)

3D pose

Automatic initialisation and recovery from failure

Future workExtend robustness to illumination changes

Non-fronto-parallel poses

Poses when arms are inside the body silhouette

Simple gesture recognition by assigning semantics to regions of articulation space