Development of a Vision-Based Measurement System for ... · 9/9/2014  · Development of a...

Preview:

Citation preview

Development of a Vision-Based Measurement System for Relative Motion Compensation

Johan Lindal Haug, Morten Ottestad & Geir Hovland

2

Introduction

• Motivation, about the system

• Image processing

• Computations

• Results from experiments

• Conclusion

• Q&A

3

Introduction

• Why vision-based?

4

5

Introduction

• Wave frequency 0.05 Hz to 0.125 Hz

• Accuracy: +25mm heave, +0.5° roll and pitch

• Compute all 6 DOF

6

Camera model

• Frontal pinhole camera model

7

Radial distortion

8

Pattern

• Pattern with rounded outer corners

9

Pattern

10

Equipment: Initial Experiments

11

Equipment: www.motion-lab.no

12

Image Processing

13

Image Processing - Thresholding

14

Image Processing - Thresholding

15

Image Processing – Region of Interest

• Assume pattern is largest dark area

16

Image Processing – Feature Detection

•Harris corner detector

u-gradient v-gradient

17

Image Processing – Feature Detection

•Harris corner response

18

Image Processing - Feature Localization

•Binary Large OBject

19

Image Processing - Feature Correspondences

Features inside pattern Features along edge

20

Image Processing - Feature Correspondences

21

Image Processing - Feature Correspondences

22

Image Processing - Feature Correspondences

23

Experiments

• Static experiments

• Dynamic experiments

24

Experiments – Static results, heave

25

Experiments – Static results, heave

• Raw measurements

• Calibrated measurements

26

Experiments – Static results, roll

27

Experiments – Static results, roll

• Raw measurements

• Calibrated measurements

28

Experiments – Static results, pitch

29

Experiments – Static results, pitch

• Raw measurements

• Calibrated measurements

30

Experiments – Dynamic results, heave

31

Experiments – Dynamic results, heave

• 0.05 Hz measurements

• 0.125 Hz measurements

32

Experiments – Dynamic results, roll

33

Experiments – Dynamic results, roll

• 0.05 Hz measurements

• 0.125 Hz measurements

34

Experiments – Dynamic results, pitch

35

Experiments – Dynamic results, pitch

• 0.05 Hz measurements

• 0.125 Hz measurements

36

Experiments – Delay

• Zoomed in dynamic heave experiments

• Delay ≈ 0.1 seconds

37

Conclusion

• Solution has been found

• Reduce cycletime

• Improve quality of input-data -> global shutter

• Find and eliminate pixel-noise giving variance

Recommended