2D to 3D conversion of formula 1 footage

67
2D to 3D conversion of Formula 1 footage Serge Hendrickx 1

Transcript of 2D to 3D conversion of formula 1 footage

Page 1: 2D to 3D conversion of formula 1 footage

2D to 3D conversion of Formula 1 footage

Serge Hendrickx

1

Page 2: 2D to 3D conversion of formula 1 footage

Presentation Overview

• Thesis objective

• Conversion process

• Conclusion and result 2

• 3D rotation

• Car detection and tracking

• Inpainting

• Overlay detection and conversion of background

Page 3: 2D to 3D conversion of formula 1 footage

Objective: 2D to 3D conversion of Formula 1 footage

2D recorded and broadcasted images 3D experience

3

Page 4: 2D to 3D conversion of formula 1 footage

Objective: 2D to 3D conversion of Formula 1 footage

4

What is special about Formula 1?

Undeformable objects with near perfect 3D models -> Investigate how these could be used in this conversion process -> Not just generating best-looking 3D

Page 5: 2D to 3D conversion of formula 1 footage

Introduction to Formula1 footage Example footage

5

Page 6: 2D to 3D conversion of formula 1 footage

Presentation Overview

• Thesis objective

• Conversion process

• Conclusion and result 6

• 3D rotation

• Car detection and tracking

• Inpainting

• Overlay detection and conversion of background

Page 7: 2D to 3D conversion of formula 1 footage

3D effect

• Translation + rotation

7

Retinal Disparity

Page 8: 2D to 3D conversion of formula 1 footage

3D effect

• Translation + rotation

8

Retinal Disparity

1) Nearest-neighbor

2) 3D Projection

Page 9: 2D to 3D conversion of formula 1 footage

3D effect

• Translation + rotation

9

Retinal Disparity

1) Nearest-neighbor

2) 3D Projection

Page 10: 2D to 3D conversion of formula 1 footage

Nearest neighbor rotation

10

For each pixel: 1) Find corresponding RGB 2) Detect nearest match in

rotated view 3) Copy pixelvalue

XYZ represented by RGB during rendering

Page 11: 2D to 3D conversion of formula 1 footage

11

Nearest neighbor rotation

Page 12: 2D to 3D conversion of formula 1 footage

3D effect

• Translation + rotation

12

Retinal Disparity

1) Nearest-neighbor

2) 3D Projection

Page 13: 2D to 3D conversion of formula 1 footage

3D projection

13

Traverse rendering pipeline

(X,Y) –coordinate in new view

(X,Y,Z) coordinate

(X,Y) –coordinate in known view

Using ModelView and Projection matrices of new view

Using ModelView and Projection matrices of known view

Page 14: 2D to 3D conversion of formula 1 footage

3D projection

14

Traverse rendering pipeline

Source:

Generated:

Page 15: 2D to 3D conversion of formula 1 footage

3D rotation

• 3D projection : + Fast + Produces clear image - Holes

15

• Neighborhood: - Very slow - Artifacts + Fills in holes

Page 16: 2D to 3D conversion of formula 1 footage

3D rotation

• 3D projection : + Fast + Produces clear image - Holes

16

• Neighborhood: - Very slow - Artifacts + Fills in holes

-> Combine both methods

Page 17: 2D to 3D conversion of formula 1 footage

17

Source:

3D rotation: final result

Rotated 10 degrees:

Page 18: 2D to 3D conversion of formula 1 footage

18

• Translation + rotation

Retinal Disparity

3D effect

Page 19: 2D to 3D conversion of formula 1 footage

19

• Translation + rotation

Retinal Disparity

3D effect

-> Exact position and pose of car needed

Page 20: 2D to 3D conversion of formula 1 footage

Presentation Overview

• Thesis objective

• Conversion process

• Conclusion and result 20

• 3D rotation

• Car detection and tracking

• Inpainting

• Overlay detection and conversion of background

Page 21: 2D to 3D conversion of formula 1 footage

Car detection and tracking

• Car detection

• Tracking throughout fragment

21

Page 22: 2D to 3D conversion of formula 1 footage

Car detection and tracking

• Car detection

• Tracking throughout fragment

22

Page 23: 2D to 3D conversion of formula 1 footage

Car detection

23

Sliding window -> binary classifier: contains car yes or no?

Page 24: 2D to 3D conversion of formula 1 footage

Car detection

24

Sliding window -> binary classifier: contains car yes or no?

Page 25: 2D to 3D conversion of formula 1 footage

Car detection

25

Sliding window -> binary classifier: contains car yes or no?

Page 26: 2D to 3D conversion of formula 1 footage

Car detection

26

Sliding window -> binary classifier: contains car yes or no?

Page 27: 2D to 3D conversion of formula 1 footage

Render car from all different angles

27

Car detection

Page 28: 2D to 3D conversion of formula 1 footage

Car detection

28

Sliding window -> binary classifier: contains car yes or no?

-> Does it match one of the renders?

Page 29: 2D to 3D conversion of formula 1 footage

Car detection

29

How to compare? Pixel per pixel -> Not resistant to illumination changes and other differences Solution: Histogram of Oriented Gradients (HoG)

Page 30: 2D to 3D conversion of formula 1 footage

HoG based object detection

30

Histogram of Oriented Gradients (HoG)

Gradient computation

Gradient binning

Page 31: 2D to 3D conversion of formula 1 footage

HoG based object detection

31

Parts-based detection

Star-based representation:

Real example:

Page 32: 2D to 3D conversion of formula 1 footage

HoG based object detection

32

Page 33: 2D to 3D conversion of formula 1 footage

Analysis of detection method on F1 footage

33

Analysis of distibution of certainty-scores of 1 particular render on 1 frame

Example: best certainty-score is 10 -> only 4% of matches score > 9 -> ony 14% of matches score > 5

Page 34: 2D to 3D conversion of formula 1 footage

Analysis of detection method on F1 footage

34

Score deduction in neighboring renders

Page 35: 2D to 3D conversion of formula 1 footage

Car detection and tracking

• Car detection

• Tracking throughout fragment

35

Page 36: 2D to 3D conversion of formula 1 footage

Car tracking

36

36,000 renders (360*100 angles) * 500 frames = 18,000,000 detections -> coarse selection of renders and frames for first pass.

Tier 1

Page 37: 2D to 3D conversion of formula 1 footage

Car tracking

37

36,000 renders (360*100 angles) * 500 frames = 18,000,000 detections -> coarse selection of renders and frames for first pass.

Tier 1

Page 38: 2D to 3D conversion of formula 1 footage

Car tracking

38

Tier 2 Starting from best detection: 1) Rerun detection on best frame 2) Detect surrounding frames

Surrounding frames: - Car is at nearly same position - Car viewed under nearly same angle

Page 39: 2D to 3D conversion of formula 1 footage

Car tracking

39

Tier 3: smoothing Smooth angles Smooth boudingbox location and size

Horizontal angle

Vertical angle

Page 40: 2D to 3D conversion of formula 1 footage

Car tracking

40

Tier 3: smoothing Smoothed angles can still be used Exact size and position needed

Horizontal angle

Vertical angle

Page 41: 2D to 3D conversion of formula 1 footage

Car tracking

41

Tier 3: smoothing Keypoint tracking

Track points using feature-tracking algorithm Calculate new positions using rendering pipeline

Page 42: 2D to 3D conversion of formula 1 footage

Car tracking

42

Tier 3: smoothing Keypoint tracking

Page 43: 2D to 3D conversion of formula 1 footage

Car tracking

43

Tier 3: smoothing Car rotation and template matching

1) Best match

2) Rotate to desired angle 3) Find rotated car in cropped image

Page 44: 2D to 3D conversion of formula 1 footage

Car tracking

44

Tier 3: smoothing Car rotation and template matching

Page 45: 2D to 3D conversion of formula 1 footage

Car tracking

45

Tier 3: smoothing Car rotation and template matching

1) Detect car 2) Rotate car to desired angle 3) Average multiple rotated cars

Page 46: 2D to 3D conversion of formula 1 footage

Car tracking

46

Tier 3: smoothing

Page 47: 2D to 3D conversion of formula 1 footage

Presentation Overview

• Thesis objective

• Conversion process

• Conclusion and result 47

• 3D rotation

• Car detection and tracking

• Inpainting

• Overlay detection and conversion of background

Page 48: 2D to 3D conversion of formula 1 footage

48

• Translation + rotation

Retinal Disparity

3D effect

Page 49: 2D to 3D conversion of formula 1 footage

Inpainting

49

Single frame methods:

Geometrical inpainting: Patch-based inpainting:

Page 50: 2D to 3D conversion of formula 1 footage

Inpainting

50

Single frame methods:

Geometrical inpainting: Patch-based inpainting:

Page 51: 2D to 3D conversion of formula 1 footage

Inpainting

51

Single frame methods:

Objective comparison Only outermost mixels important

Page 52: 2D to 3D conversion of formula 1 footage

Inpainting

52

Video inpainting

Page 53: 2D to 3D conversion of formula 1 footage

Inpainting

53

Video inpainting

Page 54: 2D to 3D conversion of formula 1 footage

Inpainting

54

Video inpainting

Page 55: 2D to 3D conversion of formula 1 footage

Inpainting

55

Video inpainting

Page 56: 2D to 3D conversion of formula 1 footage

Inpainting

56

Video inpainting

Page 57: 2D to 3D conversion of formula 1 footage

Presentation Overview

• Thesis objective

• Conversion process

• Conclusion and result 57

• 3D rotation

• Car detection and tracking

• Inpainting

• Non-object based 3D conversion

Page 58: 2D to 3D conversion of formula 1 footage

Non-object based 3D conversion

• Overlay detection

• Conversion of background to 3D

58

Page 59: 2D to 3D conversion of formula 1 footage

Non-object based 3D conversion

• Overlay detection

• Conversion of background to 3D

59

Page 60: 2D to 3D conversion of formula 1 footage

Overlay detection

60

Graphical overlay: - Not always present - If present, always at fixed location

Template: averaged over 100 frames

Page 61: 2D to 3D conversion of formula 1 footage

Non-object based 3D conversion

• Overlay detection

• Conversion of background to 3D

61

Page 62: 2D to 3D conversion of formula 1 footage

Conversion of background to 3D

62

Make3D algorithm - Texture gradients - Texture variations - Reduced contrast

Page 63: 2D to 3D conversion of formula 1 footage

Presentation Overview

• Thesis objective

• Conversion process

• Conclusion and result 63

• 3D rotation

• Car detection and tracking

• Inpainting

• Non-object based 3D conversion

Page 64: 2D to 3D conversion of formula 1 footage

Overview of 3D conversion process

64

1) Inpaint car and overlay 2) Generate depthmap 3) Shift background 4) Inpaint background 5) Add overlay 6) Add car 7) Merge to 3D

Page 65: 2D to 3D conversion of formula 1 footage

Presentation Overview

• Thesis objective

• Conversion process

• Conclusion and result 65

• 3D rotation

• Car detection and tracking

• Inpainting

• Non-object based 3D conversion

Page 66: 2D to 3D conversion of formula 1 footage

Result and conclusion

66

Is it a feasible method of 2D-3D conversion?

Yes, but 1) very time-consuming -> also the reason why most of this thesis featured 1 single fragment 2) differences between chosen render-viewpoint and car can become noticeable -> can be solved/prevented by using realistically rendered car instead of the actual car from the frame Car angle, size and location known at each frame -> could be used for other purposes, for example advanced graphical overlays

Result: video on 3D tv

Page 67: 2D to 3D conversion of formula 1 footage

Questions?

67