articulated human detection

53
Student: Yao-Sheng Wang Advisor: Prof. Sheng- Jyh Wang ARTICULATED HUMAN DETECTION 1 Department of Electronics Engineering National Chiao Tung University Hsinchu, Taiwan 1 Vision Lab 2012

Upload: althea

Post on 15-Feb-2016

35 views

Category:

Documents


0 download

DESCRIPTION

Articulated Human Detection . Student: Yao-Sheng Wang Advisor: Prof. Sheng- Jyh Wang. Department of Electronics Engineering National Chiao Tung University. Hsinchu, Taiwan. Vision Lab 2012. 1. Outline. Introduction Related Works Idea Proposed Method Experimental Results - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Articulated Human  Detection

Student: Yao-Sheng WangAdvisor: Prof. Sheng- Jyh Wang

ARTICULATED HUMAN DETECTION

1

Department of Electronics Engineering

National Chiao Tung UniversityHsinchu, Taiwan

1Vision Lab 2012

Page 2: Articulated Human  Detection

Introduction Related Works Idea Proposed Method Experimental Results Conclusion Reference

OUTLINE

2

Page 3: Articulated Human  Detection

Introduction Motivation Challenge Representative Works Potential Problems Target

Related Works Idea Proposed Method Experimental Results Conclusion Reference

OUTLINE

3

Page 4: Articulated Human  Detection

4

Why we care about human detection? We are human beings!

Wide range of applications: Automotive safety Surveillance system

Indoor care Crime alert

Human-Computer Interface … etc.

MOTIVATION

Page 5: Articulated Human  Detection

Introduction Motivation Challenge Representative Works Potential Problems Target

Related Works Idea Proposed Method Experimental Results Conclusion Reference

OUTLINE

5

Page 6: Articulated Human  Detection

6

What makes human detection so difficult? Illumination condition Cluttered background Change of viewpoints Occlusion Wearing difference Diversity of human Pose variation

CHALLENGE

Page 7: Articulated Human  Detection

7

What makes human detection so difficult? Illumination condition Cluttered background Change of viewpoints Occlusion Wearing difference Diversity of human Pose variation

CHALLENGE

Page 8: Articulated Human  Detection

8

What makes human detection so difficult? Illumination condition Cluttered background Change of viewpoints Occlusion Wearing difference Diversity of human Pose variation

CHALLENGE

Page 9: Articulated Human  Detection

9

What makes human detection so difficult? Illumination condition Cluttered background Change of viewpoints Occlusion Wearing difference Diversity of human Pose variation

CHALLENGE

Page 10: Articulated Human  Detection

10

Progress on “Machine Learning” technologyHandle more general and complicate cases.

Definition: “Articulated Human Detection”.

CHALLENGE

Page 11: Articulated Human  Detection

Introduction Motivation Challenge Representative Works Potential Problems Target

Related Works Idea Proposed Method Experimental Results Conclusion Reference

OUTLINE

11

Page 12: Articulated Human  Detection

12

Deformable Part Model Root filter (mask). Part filter (mask). Penalty function.

REPRESENTATIVE WORKS (I)

[P. Felzenszwalb, D. McAllester, and D. Ramanan. A discriminatively trained, multi-scale, deformable part model. In CVPR, 2008.]

Page 13: Articulated Human  Detection

13

Pose-let:

REPRESENTATIVE WORKS (II)

[Lubomir Bourdev, Jitendra Malik. Poselets: Body Part Detectors Trained Using 3D Human Pose Annotations. In ICCV, 2009.].

Page 14: Articulated Human  Detection

Introduction Motivation Challenge Representative Works Potential Problems Target

Related Works Idea Proposed Method Experimental Results Conclusion Reference

OUTLINE

14

Page 15: Articulated Human  Detection

15

Problems: System complexity increased with the complexity of human poses. More detectors needed.

Exhaustive search. Sliding window method + Image pyramid.

Both problems leads to unacceptable speed for applications in real life.

POTENTIAL PROBLEMS

Page 16: Articulated Human  Detection

Introduction Motivation Challenge Representative Works Potential Problems Target

Related Works Idea Proposed Method Experimental Results Conclusion Reference

OUTLINE

16

Page 17: Articulated Human  Detection

17

Target in the thesis: Propose a detection scheme with acceptable detection speed in dealing with highly intra- class variation from the change of pose and viewpoint.

TARGET

Page 18: Articulated Human  Detection

Introduction Related Works Idea Proposed Method Experimental Results Conclusion Reference

OUTLINE

18

Page 19: Articulated Human  Detection

19

Better features: Cheap to compute and capture crucial information at the same time. Ex: HOG.

Better classifiers: Linear classifiers.

Ex: Adaboost, Linear-SVM and Random-forests.

Better prior knowledge: Ex: Information about ground plane.

RELATED WORKS

Page 20: Articulated Human  Detection

20

Cascades: Cascade the part filters to reduce the searching regions.

RELATED WORKS

[P. Felzenszwalb, R. Girshick, D. McAllester. Cascade Object Detection with Deformable Part Models. In CVPR, 2010.]

Page 21: Articulated Human  Detection

21

Discard non-promising hypotheses. Class-dependent:

Branch and bound. (CVPR, 2008) Class-independent:

What is an object? (CVPR, 2010) Closure boundary, different appearance or salience.

Segmentation as selective search. (ICCV, 2011)

RELATED WORKS

Start

Page 22: Articulated Human  Detection

22

Feature response approximation: Feature approximation in testing step. Feature approximation in training step.

RELATED WORKS

[R. Benenson, M. Mathias, R. Timofte, and L. Van Gool. Pedestrian detection at 100 frames per second. In CVPR, 2012.]

[P. Dollár, S. Belongie, P. Perona. The fastest pedestrian detector in the west. In BMVC, 2010.]

Page 23: Articulated Human  Detection

Introduction Related Works Idea Proposed Method Experimental Results Conclusion Reference

OUTLINE

23

Page 24: Articulated Human  Detection

24

Recall the memory of the first problem: System complexity increased with the complexity of human poses (include variation of viewpoints).

How can we break the relation between the complexity of system and the one of human poses? Choose stable features or body parts for detection.

IDEA

Page 25: Articulated Human  Detection

25

Better prior knowledge:

IDEA

Page 26: Articulated Human  Detection

26

Recall the memory of the second problem: Exhaustive search.

“Sliding Window” + “Image Pyramid”. How can we reduce the searching region?

Detect the common feature among these parts.

Use the cumulative characteristic of the feature to handle the variation of scale.

IDEA

Page 27: Articulated Human  Detection

27

Common feature Body parts consist of combination of two edge segments.

Cumulative characteristic Edge detector with fixed size + Combination.

IDEA

Page 28: Articulated Human  Detection

28

The previous works focus on reducing the searching regions. Specifically against “Exhaustive Search”.

Our method starts from breaking the relation between complexity of system and that of poses. Then, use the common feature and cumulative characteristic to cut down the searching space.

COMPARISON

Page 29: Articulated Human  Detection

Introduction Related Works Idea Proposed Method Experimental Results Conclusion Reference

OUTLINE

29

Page 30: Articulated Human  Detection

30

Fast Part Detection

Part Combinati

on

Combination

Refinement

SYSTEM BLOCK

Bottom-up system:

Page 31: Articulated Human  Detection

31

Fast Part Detection

Part Combinati

on

Combination

Refinement

SYSTEM BLOCK

Bottom-up system:

Page 32: Articulated Human  Detection

32

Steps: Detection of edge candidates. Production of part candidates. Refinement of part candidates.

FAST PART DETECTION

Page 33: Articulated Human  Detection

33

Detection and combination of segments (9 orientations).

DETECTION OF PART CANDIDATES

Page 34: Articulated Human  Detection

34

Constraints on combination of edges. Orientation, length ratio and color symmetry.

PRODUCTION OF PART CANDIDATES

Neighbor orientation consideration

Page 35: Articulated Human  Detection

35

HOG feature + Random forest training

REFINEMENT OF PART CANDIDATES

Feature = [Length Orientation HOG_features]

feature134

feature33

feature2

? ?

feature400

Page 36: Articulated Human  Detection

36

Fast Part Detection

Part Combinati

on

Combination

Refinement

SYSTEM BLOCK

Bottom-up system:

Page 37: Articulated Human  Detection

37

Problem: No information about the classes of the limbs

due to the low resolution of images or variation from hand gestures or appearance of shoes...etc.

Need another step to refine the combinations. What information left?

Head-shoulder or head-torso.

PART COMBINATION

Page 38: Articulated Human  Detection

38

Any possibility for us to estimate the position and orientation of head-torso based     on the architecture of current combinations?

PART COMBINATION

Page 39: Articulated Human  Detection

39

Problem: How to select body parts belong to specific human from lots of part candidates?

Too much possibilities for exhaustive search.

PART COMBINATION

Page 40: Articulated Human  Detection

40

Clues for reducing the number of possible combinations. Center distance, length ration or width ratio between two parts.

Combination with the number of parts more than four.

PART COMBINATION

Page 41: Articulated Human  Detection

41

Conclusion for the clues mentioned in the previous slide. Too complicate to combine the parts for the whole body.

Start from low-level combination of parts to reveal the benefits of physical constraints.

Break the problems into two levels. Low-level combination. High-level combination.

PART COMBINATION

Page 42: Articulated Human  Detection

42

How far can we reach for low-level combination? 4-parts combination = lower body.

LOW-LEVEL COMBINATION

Page 43: Articulated Human  Detection

43

False alarm exists.

Joints relative position + Random Forest

LOW-LEVEL COMBINATION

feature134

feature33

feature2

? ?

feature400

Page 44: Articulated Human  Detection

44

Combination between the arms, legs, lower bodies and uncombined single parts from the low-level combination step. Upper bound of the number of combination:

HIGH-LEVEL COMBINATION

Page 45: Articulated Human  Detection

45

Fast Part Detection

Part Combinati

on

Combination

Refinement

SYSTEM BLOCK

Bottom-up system:

Page 46: Articulated Human  Detection

46

Pose prediction. Detection with DPM detector.

COMBINATION REFINEMENT

Page 47: Articulated Human  Detection

47

Feature: Relative size ratio and positions between low-level combinations and architecture of each low-level combination.

Random Forest.

POSE PREDICTION

Page 48: Articulated Human  Detection

48

Use DPM detector to cover the intra-class variation.

Model:

DETECTION WITH DPM DETECTOR

Page 49: Articulated Human  Detection

49

Much stronger than information of limbs. Head-shoulder to head-torso. Start from head-torso to combine limbs back.

USAGE OF HEAD-SHOULDER INFORMATION

Page 50: Articulated Human  Detection

50

SYSTEM ILLUSTRATION

Edge Candidates

Part Candidates

Low Level Part Combine

Pose PredictionHead-Torso Detection

High Level Part Combine

Part Detector

Parts

Low Level Combination

Result ofDetection

High Level Combination

Page 51: Articulated Human  Detection

Introduction Related Works Idea Proposed Method Experimental Results Conclusion Reference

OUTLINE

51

Page 52: Articulated Human  Detection

Introduction Related Works Idea Proposed Method Experimental Results Conclusion Reference

OUTLINE

52

Page 53: Articulated Human  Detection

Introduction Related Works Idea Proposed Method Experimental Results Conclusion Reference

OUTLINE

53