Original subtitles

1

Write in the previews Statoil toil we detected the faces using the detect Montell scale method from

2

our face cascade object and we gather the coordinates and the width and height of the detector rectangles

3

in faces we iterated through the faces to draw the rectangles around them.

4

And now now that we have those rectangles we can detect the eyes because we are actually going to take

5

the eyes in the referential of the faces that is in the referentially of these rectangles.

6

So the first thing we're going to do to detect the eyes is to get to region of interest inside the rectangles

7

and y2 region of interest because we need one region of interest for the black and white image in which

8

the cascade is going to be applied to detect the eyes.

9

And of course one other region of interest for the original colored image because we will draw the rectangles

10

in the original color image.

11

So let's get these two regions of interest.

12

So the first one for the black and white image.

13

I'm going to call it our I underscore gray and I'm going to take my gray image that is the black and

14

white image and I'm going to take the zone of the rectangles which corresponds to the first range y

15

Collin y plus H H is the height and then Same for x which is the x coordinate of the upper left corner

16

of the rectangle.

17

And we're taking the range from x to x.

18

Plus the width which is W..

19

Right now we have our zone of interest which corresponds exactly to the zone inside the detector rectangle.

20

But remember we need to do that.

21

We need to get a region of interest for both the gray images and the original color image.

22

So I'm introducing a new variable here which is going to be our own eye color design for the original

23

color image and therefore I need to here not gray but frame which is my original color image and the

24

same zones of coordinates.

25

All right so now we have two region of interest the zone inside the rectangle detecting the faces both

26

in the black and white image and the colored image.

27

And now now that we have this we can detect the eyes.

28

And we're going to detect them in the region of interest.

29

Of course the gray image because the Cascades are applied in black and white images.

30

And so how are we going to detect the eyes.

31

Well we're going to do exactly the same as what we did for the faces we're going to introduce a new

32

variable which we're going to call ice then we're going to take our eye cascade object from the cascade

33

classified class and we're going to apply to detect multihulls scale method to detect the ice that is

34

to get the coordinates of the upper left corner of the rectangle detecting the eyes and the width and

35

the height of these rectangles.

36

So basically what we simply need to do is take this and here paste it replies here faces by eyes.

37

That's a new variable to get these doubles of coordinates with and height for the eyes the rectangle

38

detecting the eyes.

39

And here we of course need to take not our face Cascada object but our eye Cascada objects and then

40

same to detect multifocal method to detect Montell scale method as a method of the Cascade pass first

41

class so we can use it for as many objects of the Cascade class for class as we want and then do we

42

need to replace here Grey by something.

43

Yes of course because we're detecting the eyes in the referential of the face there is a zone inside

44

the rectangle detecting the face.

45

That's just in order to save some computation time.

46

So of course we need to replace Gray here by the region of interest in black and white and that is all

47

I underscore.

48

Gray.

49

Perfect.

50

And then are we going to change these numbers here.

51

I remind that this first one here corresponds to descale factor by which we reduce the size of the images

52

or equivalently by which we increase the size of the filters.

53

That is the kernels.

54

And yes we are going to replace it by 1 point 1 again.

55

This is by experimenting.

56

We get some good results with that some fast result.

57

So 1.1 and then are we going to keep the same number of minimum neighbors.

58

I remind this is the minimum number of zones around the accepted zone to be accepted and yes we are

59

going to replace it by 3.

60

That will give us some good results.

61

All right.

62

So we now have our eyes.

63

That is a table of the coordinates of the upper left corner.

64

The width and the height of the rectangle surrounding the eyes.

65

And so now we are ready to start a new for loop in which.

66

Well again we will draw the rectangles around the ice so we'll do that in the next to tutorial and until

67

then enjoy computer vision.