Eye tracking using image data

• US 7,197,165 B2
• Filed: 02/04/2003
• Issued: 03/27/2007
• Est. Priority Date: 02/04/2002
• Status: Expired due to Fees

• Current Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Original Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Inventors: Ryan, Mathew David
• Primary Examiner(s): Wu; Jingge
• Assistant Examiner(s): Lu; Tom Y.
• Application Number:
  US10/358,067
• Publication Number:
  US 20030146901A1
• Time in Patent Office: 1,512 Days
• Field of Search: 382/103
• US Class Current: 382/103
• CPC Class Codes: G06F 3/013 Eye tracking input arrangements G06F3/015 takes precedence
• CPC Class Codes: G06F 3/04815 Interaction with a metaphor-based environment or interaction object displayed as three-dimensional, e.g. changing the user viewpoint with respect to the environment or object
• CPC Class Codes: G06V 10/25 Determination of region of interest [ROI] or a volume of interest [VOI]
• CPC Class Codes: G06V 10/62 relating to a temporal dimension, e.g. time-based feature extraction; Pattern tracking
• CPC Class Codes: G06V 40/171 Local features and components; Facial parts eye characteristics G06V40/18; Occluding parts, e.g. glasses; Geometrical relationships
• CPC Class Codes: G06V 40/19 Sensors therefor

Abstract

In a computer processing apparatus 8, frames of image data received from a camera 4 are processed to track the eyes of a user 2 in each image. A 3D computer model of a head is stored, and search regions are defined in the 3D space corresponding to the eyes and eyebrows. For each image, pixels within the projection of the search regions from the 3D space to the 2D image space are sampled to determine a representative intensity value for each of the search regions. Positions for the eyes in the 3D space are then calculated based on the determined values. The 3D computer model and search bands are moved within the 3D space to align the eyes with the calculated eye positions. In this way, when the next image is processed, the search bands project into the image from a head configuration determined from the previous image. This facilitates reliable and accurate eye tracking.

Claims

1. A method of processing data in a processing apparatus to perform eye tracking across a sequence of images of a moving head, the method comprising processing each of at least some of the images in the sequence by:

identifying pixels in the image corresponding to search regions defined relative to a three-dimensional computer model representative of at least part of a head in a three-dimensional space, the search regions being defined with different positions in a portion of the three-dimensional space corresponding to the eyes and eyebrows of the model, and the identification of pixels being performed in dependence upon data defining a transformation between the three-dimensional space and the two-dimensional image;

determining image values of the identified pixels;

generating at least one respective representative image value for each search region using the determined image values; and

determining eye positions in the three-dimensional space in dependence upon the relationship between the generated representative image values for different search regions.

2. A method according to claim 1, further comprising:

rotating the three-dimensional computer head model in dependence upon the determined positions for the eyes, so as to move the search regions within the three-dimensional space for subsequent processing of the next image in the sequence to be processed.

3. A method according to claim 1, wherein the identification of pixels is carried out by projecting sample points defined in the three-dimensional space within the search regions into the image, and identifying the pixels onto which the sample points project.

4. A method according to claim 1, wherein the identification of pixels is carried out by projecting points from the three-dimensional space into the image to generate projected points, defining projected search regions in the image in dependence upon the projected points, and identifying pixels in the projected search regions.

5. A method according to claim 1, wherein each search region is a two-dimensional area of the three-dimensional space.

6. A method according to claim 1, wherein each search region is a line in the three-dimensional space.

33. A method according to any one of claims 1, 7, 8, 20, 21 and 32, wherein the processing of each image further comprises:

processing the image data to determine a position of the head in the image; and

moving the three-dimensional computer model of the head and the associated search regions within the three-dimensional space in dependence upon the determined position.

34. A method according to any one of claims 1, 7, 8, 20, 21 and 32, wherein the processing of each image further comprises:

processing the image data to determine a size of the head in the image; and

moving the three-dimensional computer model of the head and the associated search regions within the three-dimensional space in dependence upon the determined size.

35. A method according to any one of claims 1, 7, 8, 20, 21 and 32, further comprising defining the search regions in the three-dimensional space.

36. A method according to any one of claims 1, 7, 8, 20, 21 and 32, further comprising calculating a gaze direction for each of at least some of the images.

37. A method according to claim 36, further comprising transmitting a signal conveying the determined gaze direction.

38. A method according to claim 37, further comprising making a recording of the signal either directly or indirectly.

48. A storage medium storing computer program instructions for programming a programmable processing apparatus to become operable to perform a method as set out in at any one of claims 1, 7, 8, 20, 21 and 32.

49. A signal carrying computer program instructions for programming a programmable processing apparatus to become operable to perform a method as set out in any one of claims 1, 7, 8, 20, 21 and 32.

7. A method of processing data in a processing apparatus defining a sequence of images of a person as the person'"'"'s head moves, to track the eyes of the person in the images, the method comprising processing each of at least some of the images in the sequence by:

determining image values of pixels located in the image at positions defined by the projection of search bands or points therein from a three-dimensional space into the two-dimensional image, wherein the search bands have different positions in the three-dimensional space, and wherein the projection is dependent upon the position and orientation of the search bands in the three-dimensional space resulting from the processing of the previous image in the sequence to be processed;

generating at least one respective representative image value for each search band using the determined image values;

determining eye positions in the three-dimensional space in dependence upon the relationship between the generated representative image values for different search bands; and

moving the search bands within the three-dimensional space in dependence upon the calculated positions for the eyes, so as to move the search bands for subsequent processing of the next image in the sequence to be processed.

8. A method of processing data in a processing apparatus to perform eye tracking across a sequence of images of a moving head, the method comprising processing each of at least some of the images in the sequence by:

identifying pixels in the image corresponding to search regions defined relative to a three-dimensional computer model representative of at least part of a head in a three-dimensional space, the search regions being defined with different positions in a portion of the three-dimensional space corresponding to the eyes and eyebrows of the model, and the identification of pixels being performed in dependence upon data defining a transformation between the three-dimensional space and the two-dimensional image;

determining image values of the identified pixels;

generating at least one respective representative image value for each search region using the determined image values;

generating an image value profile from the generated representative image values representative of the variation in the representative image values with variation in search region position; and

comparing the generated image value profile with a reference profile to determine eye positions in the three-dimensional space.

9. A method according to claim 8, further comprising:

rotating the three-dimensional computer head model in dependence upon the determined positions for the eyes, so as to move the search regions within the three-dimensional space for subsequent processing of the next image in the sequence to be processed.

10. A method according to claim 8, wherein the identification of pixels is carried out by projecting sample points defined in the three-dimensional space within the search regions into the image, and identifying the pixels onto which the sample points project.

11. A method according to claim 8, wherein the identification of pixels is carried out by projecting points from the three-dimensional space into the image to generate projected points, defining projected search regions in the image in dependence upon the projected points, and identifying pixels in the projected search regions.

12. A method according to claim 8, wherein each search region comprises a line in the three-dimensional space of the computer model.

13. A method according to claim 12, wherein the spacing between adjacent search regions in the three-dimensional space is such that, when the search regions are transformed into a two-dimensional image in accordance with the transformation data, the spacing between adjacent transformed search regions is less than or equal to one pixel in the image.

14. A method according to claim 8, wherein the search regions extend in a transverse direction to a line through an eye and an eyebrow of the three-dimensional computer model.

15. A method according to claim 14, wherein the search regions extend in a perpendicular direction to the line through the eye and the eyebrow.

16. A method according to claim 14, wherein the search regions intersect the line through the eye and the eyebrow at different respective positions therealong.

17. A method according to claim 8, wherein a respective average value of the plurality of image values is calculated as the representative value for each search region.

18. A method according to claim 8, wherein the processing to compare the generated image value profile with the reference profile comprises processing to compare the generated profile and reference profile for different relative positions thereof.

19. A method according to claim 18, wherein the processing to compare the generated image value profile with the reference profile comprises processing to compare the generated profile and reference profile for different relative positions thereof to determine the match therebetween for each relative position and to determine the relative position for which the match between the generated profile and reference profile is greatest.

20. A method of processing data in a processing apparatus defining a sequence of images of a person as the person'"'"'s head moves, to track the eyes of the person in the images, the method comprising processing each of at least some of the images in the sequence by:

determining image values of pixels located in the image at positions defined by the projection of search bands or positions therein from a three-dimensional space into the two-dimensional image, wherein the search bands have different positions in the three-dimensional space, and wherein the projection is dependent upon the position and orientation of the search bands in the three-dimensional space resulting from the processing of the previous image in the sequence to be processed;

generating at least one representative image value for each search band using the determined pixel image values;

comparing the generated representative image values for different search bands with reference image values representative of image values for an eye and an eyebrow to calculate positions for the eyes in the three-dimensional space; and

moving the search bands within the three-dimensional space in dependence upon the calculated positions for the eyes, so as to move the search bands for subsequent processing of the next image in the sequence to be processed.

58. A method according to any one of claims 7, 20 and 32, wherein each search band has a rectangular shape.

21. A method of processing data in a processing apparatus to perform eye tracking across a sequence of images of a moving head, the method comprising processing each of at least some of the images in the sequence by:

identifying pixels in the image in dependence upon search regions defined relative to a three-dimensional computer model representative of at least part of a head in a three-dimensional space, the search regions dividing into parts a portion of the three-dimensional space corresponding to the eyes and eyebrows of the model, and in dependence upon data defining a transformation between the three-dimensional space and the two-dimensional image;

determining image values of the identified pixels;

generating at least one respective representative image value for each search region using the determined image values;

determining which of the search regions project onto dark regions in the image by comparing the generated representative image values of different search regions; and

determining eye positions in the three-dimensional space in dependence upon the position of at least one search region determined to project onto a dark region in the image.

22. A method according to claim 21, further comprising:

rotating the three-dimensional computer head model in dependence upon the determined positions for the eyes, so as to move the search regions within the three-dimensional space for subsequent processing of the next image in the sequence to be processed.

23. A method according to claim 21, wherein the identification of pixels is carried out by projecting sample points defined in the three-dimensional space within the search regions into the image, and identifying the pixels onto which the sample points project.

24. A method according to claim 21, wherein:

the search regions include first and second groups of search regions, the first group dividing into parts a portion of the three-dimensional space extending over a first two of the left eyebrow, left eye, right eyebrow and right eye of the three-dimensional computer model, and the second group dividing into parts a portion of the three-dimensional space extending over a second two of the left eyebrow, left eye, right eyebrow and right eye of the three-dimensional computer model;

first and second search regions which correspond to dark regions in the image are identified within each group; and

positions for the eyes are determined in dependence upon the relative positions of the identified search regions within each group.

25. A method according to claim 24, wherein:

the search regions include first and second groups of search regions, the first group dividing into parts a portion of the three-dimensional space extending from above the left eyebrow to below the left eye of the three-dimensional computer model, and the second group dividing into parts a portion of the three-dimensional space extending from above the right eyebrow to below the right eye of the three-dimensional computer model;

first and second search regions corresponding to dark regions in the image are identified within each group; and

positions for the eyes are determined in dependence upon the position of the identified search region within each group which corresponds to a lower position on the three-dimensional computer head model.

26. A method according to claim 24, wherein:

the search regions include first and second sets of search regions;

the first set of search regions contains first and second groups of search regions, the first group dividing into parts a portion of the three-dimensional space extending over a first two of the left eyebrow, left eye, right eyebrow and right eye of the three-dimensional computer model and the second group dividing into parts a portion of the three-dimensional space extending over a second two of the left eyebrow, left eye, right eyebrow and right eye of the three-dimensional computer model;

within the second set of search regions, each search region comprises a plurality of spatially separated parts having relative positions dependent upon the spacing of the eyes and eyebrows in the three-dimensional computer model;

values of a first coordinate of the eye positions are determined by;

identifying first and second search regions within each group of the first set corresponding to dark regions in the image; and

determining values of the first coordinate for the eye positions in dependence upon the relative positions of the identified search regions within each group in the first set; and

values for a second coordinate of the eye positions are determined by;

identifying a search region in the second set corresponding to dark regions in the image; and

determining values of the second coordinate for the eye positions in dependence upon the positions of the spatially separated parts within the identified search region of the second set.

27. A method according to claim 26, wherein:

the plurality of search regions in the second set divide into parts a portion of the three-dimensional space extending from outside the left eye to outside the right eye of the three-dimensional computer model; and

the processing to determine the values of the second coordinate includes;

selecting from among the search regions in the second set the search region which corresponds to the darkest region in the image; and

determining values of the second coordinate for the eye positions in dependence upon the positions of the spatially separated areas within the identified search region of the second set.

28. A method according to claim 21, wherein the search regions are defined with overlaps in the three-dimensional space.

29. A method according to claim 21, wherein each search region is a two-dimensional area of the three-dimensional space.

30. A method according to claim 21, wherein the processing to determine which search regions project onto dark regions in the image includes:

spatially smoothing the representative image values; and

identifying search regions which project onto dark regions in the image using the smoothed representative image values.

31. A method according to claim 21, wherein an average value of the pixel image values is calculated as the representative value for each search region.

56. A method according to claim 21, wherein:

each search region comprises a plurality of spatially separated parts having relative positions dependent upon the spacing of the eyes and eyebrows in the three-dimensional computer model; and

eye positions are determined in the three-dimensional space by;

identifying a search region having parts corresponding to dark regions in the image; and

determining eye coordinate values in dependence upon the identified parts.

57. A method according to claim 21, wherein the generated representative image values are compared to determine which of the search regions corresponds to the darkest region of the image relative to the other search regions, and the eye positions are determined in dependence upon the determined search region.

32. A method of processing data in a processing apparatus defining a sequence of images of a person as the person'"'"'s head moves, to track the eyes of the person in the images, the method comprising processing each of least some of the images in the sequence by:

determining the image values of pixels located in the image at positions defined by the projection of search bands or points therein from a three-dimensional space into the two-dimensional image, wherein the search bands have different positions in the three-dimensional space so as to divide into parts a portion of the three-dimensional space, and wherein the projection is dependent upon the position and orientation of the search bands in the three-dimensional space resulting from the processing of the previous image in the sequence to be processed;

calculating a respective representative image value for each search band using the determined pixel image values;

calculating positions for the eyes in the three-dimensional space by comparing the representative image values for different search bands to identify at least one search band corresponding to a dark region in the image; and

moving the search bands within the three-dimensional space in dependence upon the calculated positions for the eyes, so as to move the search bands for subsequent processing of the next image in the sequence to be processed.

39. Image processing apparatus for processing image data of at least some images in a sequence of images of a moving head to perform eye tracking across the sequence, the apparatus comprising:

a data store configured to store data defining a three-dimensional computer model representative of at least part of a head in a three-dimensional space, and data defining search regions having different positions in a portion of the three-dimensional space corresponding to the eyes and eyebrows of the model;

a pixel identifier operable to identify pixels in an image corresponding to the search regions, the pixel identifier being operable to identify pixels in dependence upon a transformation between the three-dimensional space and the two-dimensional image;

an image value determinator operable to determine image values of the identified pixels;

a representative value calculator operable to generate at least one respective representative image value for each search region using the determined image values; and

an eye position determinator operable to determine eye positions in the three-dimensional space in dependence upon the relationship between the generated representative image values for different search regions.

40. An apparatus according to claim 39, further comprising:

a computer model rotator operable to rotate the three-dimensional computer head model in dependence upon the determined positions for the eyes, so as to move the search regions within the three-dimensional space for subsequent processing of the next image in the sequence to be processed.

41. Apparatus for processing image data of each of at least some images in a sequence of images of a person as the person'"'"'s head moves, to track the eyes of the person in the images, the apparatus comprising:

a pixel identifier operable to identify pixels located in the image at positions defined by the projection of search bands or points therein from a three-dimensional space into the two-dimensional image, wherein the search bands have different positions in the three-dimensional space, and wherein the projection is dependent upon the position and orientation of the search bands in the three-dimensional space resulting from the processing of the previous image in the sequence to be processed;

an image value determinator operable to determine image values of the identified pixels;

a representative value calculator operable to generate at least one respective representative image value for each search band using the determined image values;

an eye position calculator operable to calculate positions for the eyes in the three-dimensional space in dependence upon the relationship between the generated representative image values for different search bands; and

a search band mover operable to move the search bands within the three-dimensional space in dependence upon the calculated positions for the eyes, so as to move the search bands for subsequent processing of the next image in the sequence to be processed.

61. Apparatus according to any one of claims 41, 44 and 47, wherein each search band has a rectangular shape.

42. Apparatus for processing image data of each of at least some images in a sequence of images of a moving head to perform eye tracking across the sequence, the apparatus comprising:

a data store configured to store data defining a three-dimensional computer model representative of at least part of a head in a three-dimensional space, and data defining search regions having different positions in a portion of the three-dimensional space corresponding to the eyes and eyebrows of the model;

a pixel identifier operable to identify pixels in the image corresponding to the search regions, the pixel identifier being operable to identify pixels in dependence upon a transformation between the three-dimensional space and the two-dimensional image;

an image value determinator operable to determine image values of the identified pixels;

a representative value calculator operable to generate at least one respective representative image value for each search region using the determined image values;

a profile generator operable to generate an image value profile from the generated representative image values representative of the variation in the representative image values with variation in search region position; and

a profile comparer operable to compare the generated image value profile with a reference profile to determine eye positions in the three-dimensional space.

43. Apparatus according to claim 42, further comprising:

a computer model rotator operable to rotate the three-dimensional computer head model in dependence upon the determined positions for the eyes, so as to move the search regions within the three-dimensional space for subsequent processing of the next image in the sequence to be processed.

44. Apparatus for processing image data of each of at least some images in a sequence of images of a person as the person'"'"'s head moves, to track the eyes of the person in the images, the apparatus comprising:

a pixel identifier operable to identify pixels located in the image at positions defined by the projection of search bands or positions therein from a three-dimensional space into the two-dimensional image, wherein the search bands have different positions in the three-dimensional space, and wherein the projection is dependent upon the position and orientation of the search bands in the three-dimensional space resulting from the processing of the previous image in the sequence to be processed;

an image value determinator operable to determine image values of the identified pixels;

a value generator operable to generate at least one representative image value for each search band using the determined pixel image values;

a comparer operable to compare the generated representative image values for different search bands with reference image values representative of image values for an eye and an eyebrow to calculate positions for the eyes in the three-dimensional space; and

a search band mover operable to move the search bands within the three-dimensional space in dependence upon the calculated positions for the eyes, so as to move the search bands for subsequent processing of the next image in the sequence to be processed.

45. Apparatus for processing image data of at least some images in a sequence of images of a moving head to perform eye tracking across the sequence, the apparatus comprising:

a data store configured to store data defining a three-dimensional computer model representative of at least part of a head in a three-dimensional space, and data defining search regions dividing into parts a portion of the three-dimensional space corresponding to the eyes and eyebrows of the model;

a pixel identifier operable to identify pixels in the image in dependence upon the search regions and in dependence upon data defining a transformation between the three-dimensional space and the two-dimensional image;

an image value determinator operable to determine image values of the identified pixels;

a representative value calculator operable to generate at least one respective representative image value for each search region using the determined image values;

a search region selector operable to determine which of the search regions project onto dark regions in the image by comparing the generated representative image values of different search regions; and

an eye position determinator operable to determine eye positions in the three-dimensional space in dependence upon the position of at least one search region determined to project onto a dark region in the image.

46. An apparatus according to claim 45, further comprising:

a computer model rotator operable to rotate the three-dimensional computer head model in dependence upon the determined positions for the eyes, so as to move the search regions within the three-dimensional space for subsequent processing of the next image in the sequence to be processed.

59. Apparatus according to claim 45, wherein:

each search region comprises a plurality of spatially separated parts having relative positions dependent upon the spacing of the eyes and eyebrows in the three-dimensional computer model;

the search region selection means is operable to identify a search region having parts corresponding to dark regions in the image; and

the eye position determining means is operable to determine eye coordinate values in dependence upon the identified parts.

60. Apparatus according to claim 45, wherein the search region selection means is operable to compare the generated representative image values to determine which of the search regions corresponds to the darkest region of the image relative to the other search regions, and the eye position calculating means is operable to determine the eye positions in dependence upon the determined search region.

47. Apparatus for processing image data of each of at least some images in a sequence of images of a person as the person'"'"'s head moves, to track the eyes of the person in the images, the apparatus comprising:

a pixel identifier operable to identify pixels located in the image at positions defined by the projection of search bands or points therein from a three-dimensional space into the two-dimensional image, wherein the search bands have different positions in the three-dimensional space so as to divide into parts a portion of the three-dimensional space, and wherein the projection is dependent upon the position and orientation of the search bands in the three-dimensional space resulting from the processing of the previous image in the sequence to be processed;

an image value determinator operable to determine image values of the identified pixels;

a value calculator operable to calculate a respective representative image value for each search band using the determined pixel image values;

an eye position calculator operable to calculate positions for the eyes in the three-dimensional space by comparing the representative image values for different search bands to identify at least one search band corresponding to a dark region in the image; and

a search band mover operable to move the search bands within the three-dimensional space in dependence upon the calculated positions for the eyes, so as to move the search bands for subsequent processing of the next image in the sequence to be processed.

50. Image processing apparatus for processing image data of at least some images in a sequence of images of a moving head to perform eye tracking across the sequence, the apparatus comprising:

means for storing data defining a three-dimensional computer model representative of at least part of a head in a three-dimensional space, and data defining search regions having different positions in a portion of the three-dimensional space corresponding to the eyes and eyebrows of the model;

pixel identifying means for identifying pixels in an image corresponding to the search regions, the pixel identifying means being operable to identify pixels in dependence upon a transformation between the three-dimensional space and the two-dimensional image;

means for determining image values of the identified pixels;

representative value calculating means for generating at least one respective representative image value for each search region using the determined image values; and

means for determining eye positions in the three-dimensional space in dependence upon the relationship between the generated representative image values for different search regions.

51. Apparatus for processing image data of each of at least some images in a sequence of images of a person as the person'"'"'s head moves, to track the eyes of the person in the images, the apparatus comprising:

means for identifying pixels located in the image at positions defined by the projection of search bands or points therein from a three-dimensional space into the two-dimensional image, wherein the search bands have different positions in the three-dimensional space, and wherein the projection is dependent upon the position and orientation of the search bands in the three-dimensional space resulting from the processing of the previous image in the sequence to be processed;

means for determining image values of the identified pixels;

representative value calculating means for generating at least one respective representative image value for each search band using the determined image values;

means for calculating positions for the eyes in the three-dimensional space in dependence upon the relationship between the generated representative image values for different search bands; and

means for moving the search bands within the three-dimensional space in dependence upon the calculated positions for the eyes, so as to move the search bands for subsequent processing of the next image in the sequence to be processed.

52. Apparatus for processing image data of each of at least some images in a sequence of images of a moving head to perform eye tracking across the sequence, the apparatus comprising:

means for storing data defining a three-dimensional computer model representative of at least part of a head in a three-dimensional space, and data defining search regions having different positions in a portion of the three-dimensional space corresponding to the eyes and eyebrows of the model;

pixel identifying means for identifying pixels in the image corresponding to the search regions, the pixel identifying means being operable to identify pixels in dependence upon a transformation between the three-dimensional space and the two-dimensional image;

means for determining image values of the identified pixels;

representative value calculating means for generating at least one respective representative image value for each search region using the determined image values;

means for generating an image value profile from the generated representative image values representative of the variation in the representative image values with variation in search region position; and

means for comparing the generated image value profile with a reference profile to determine eye positions in the three-dimensional space.

53. Apparatus for processing image data of each of at least some images in a sequence of images of a person as the person'"'"'s head moves, to track the eyes of the person in the images, the apparatus comprising:

means for identifying pixels located in the image at positions defined by the projection of search bands or positions therein from a three-dimensional space into the two-dimensional image, wherein the search bands have different positions in the three-dimensional space, and wherein the projection is dependent upon the position and orientation of the search bands in the three-dimensional space resulting from the processing of the previous image in the sequence to be processed;

means for determining image values of the identified pixels;

means for generating at least one representative image value for each search band using the determined pixel image values;

means for comparing the generated representative image values for different search bands with reference image values representative of image values for an eye and an eyebrow to calculate positions for the eyes in the three-dimensional space; and

means for moving the search bands within the three-dimensional space in dependence upon the calculated positions for the eyes, so as to move the search bands for subsequent processing of the next image in the sequence to be processed.

54. Apparatus for processing image data of at least some images in a sequence of images of a moving head to perform eye tracking across the sequence, the apparatus comprising:

means for storing data defining a three-dimensional computer model representative of at least part of a head in a three-dimensional space, and data defining search regions dividing into parts a portion of the three-dimensional space corresponding to the eyes and eyebrows of the model;

pixel identifying means for identifying pixels in the image in dependence upon the search regions and in dependence upon data defining a transformation between the three-dimensional space and the two-dimensional image;

means for determining image values of the identified pixels;

representative value calculating means for generating at least one respective representative image value for each search region using the determined image values;

search region selection means for determining which of the search regions project onto dark regions in the image by comparing the generated representative image values of different search regions; and

eye position determining means for determining eye positions in the three-dimensional space in dependence upon the positions of at least one search region determined to project onto a dark region in the image.

55. Apparatus for processing image data of each of at least some images in a sequence of images of a person as the person'"'"'s head moves, to track the eyes of the person in the images, the apparatus comprising:

means for identifying pixels located in the image at positions defined by the projection of search bands or points therein from a three-dimensional space into the two-dimensional image, wherein the search bands have different positions in the three-dimensional space so as to divide into parts a portion of the three-dimensional space, and wherein the projection is dependent upon the position and orientation of the search bands in the three-dimensional space resulting from the processing of the previous image in the sequence to be processed;

means for determining image values of the identified pixels;

means for calculating a respective representative image value for each search band using the determined pixel image values;

means for calculating positions for the eyes in the three-dimensional space by comparing the representative image values for different search bands to identify at least one search band corresponding to a dark region in the image; and

means for moving the search bands within the three-dimensional space in dependence upon the calculated positions for the eyes, so as to move the search bands for subsequent processing of the next image in the sequence to be processed.