By Joaquim Rocha. With the release of the Kinect device, there was finally an affordable camera capable of giving depth information. This, together with the Kinect's open USB connection, led to a lot of innovative projects. Still, the Kinect just gives raw signals and the only way to obtain more complex information, such as skeleton tracking was to use either the Microsoft SDK or the OpenNI framework. Both of these solutions are closed, proprietary and, in the case of Microsoft's, only for non-commercial work. To solve the issue above, Igalia developed Skeltrack, a Free and Open Source library published under LGPL that performs human skeleton tracking and identifies a number of skeleton joints. It is a more atomic solution than the other commercial counterparts because it does not connect directly to the Kinect nor to any other depth camera, instead it expects to be given the buffer corresponding to the depth buffer. In this talk I will present how Skeltrack was developed and show a demo of it working.

1. Skeltrack - Open Source Skeleton Tracking
Joaquim Rocha, Igalia
LinuxTag 2012 - Wunderbare Berlin

2. Guten Tag!
✩ I am a developer at Igalia
✩ I like doing innovative stuff like OCRFeeder and SeriesFinale
✩ and today I am presenting my latest project: Skeltrack

3. The Kinect

4. Microsoft's Kinect was the first camera
with a price affordable to the public

5. The USB connection is open and thus hackable

6. This originated Open Source projects like the libfreenect,
a library to control the Kinect device and get its information

7. We created a GLib wrapper for libfreenect called GFreenect

8. GFreenect offers asynchronous functions (and some synchronous as
well) and makes it easy to use with other GNOME technologies

9. GObject Introspection = free bindings (Python, Javascript, Vala)

10. Kinect has a structured light camera which gives depth information

11. But that's raw information... values from 0-2048

12. libfreenect/GFreenect can give those values in mm

14. Still...

15. It does NOT tell you there is a person in the picture

16. Or a cow

17. Or an ampelmann

18. Let alone a skeleton and where its joints are

19. For this you need a skeleton tracking solution

20. Three proprietary/closed solutions exist:

21. Microsoft Kinect SDK: non-commercial only

22. OpenNI: commercial compatible

23. Kinect for Windows: commercial use allowed
but incompatible with the XBox's Kinect

25. Conclusion: There were no Free solutions to
perform skeleton tracking... :(

26. So Igalia built one!

27. Enter Skeltrack

28. What we wanted:
✩ A shared library, no fancy SDK
✩ Device independent
✩ No pattern matching, no databases
✩ Easy to use (everybody wants that!)

29. Not as easy as it sounds!

30. After some investigation we found Andreas Baak's
paper "A Data-Driven Approach for Real-Time Full
Body Pose Reconstruction from a Depth Camera"

31. However this paper uses a database of
poses to get what the user is doing

32. So we based only part of our work on it

33. How does it work?

34. First we need to find the extremas

35. Make a graph whose nodes are the depth pixels

36. Connect two nodes if the distance is less than a
certain value

37. Connect the different graph's components by using
connected-component labeling

38. Choose a starting point and calculate Dijkstra to
each point of the graph; choose the furthest point.
There you got your extrema!

39. Then create an edge between the starting point
and the current extrema point with 0 cost and
repeat the same process now using the current
extrema as a starting point.

40. This comes from Baak's paper and the difference
starts here: choosing the starting point

41. Baak chooses a centroid as the starting point
We choose the bottom-most point starting from the
centroid (this showed better results for the upper
body extremas)

42. So we got ourselves some extremas!
What to do with them?

43. What extrema is a hand, a head, a shoulder?

44. For that we use educated guesses...

45. We calculate 3 extremas

46. Then we check each of them hoping they are the head

47. How?

48. For each extrema we look for the points in places
where the shoulders should be, checking their distances
between the extrema and between each other.

49. If they obey those rules then we assume they are
the head'n'shoulders (tm)

50. With the remaining 2 extremas, we will try to see if
they are elbows or hands

51. How to do it?

52. Calculate Dijkstra from the shoulders to each extrema

53. The closest extrema to any of the shoulders is either a
hand of an elbow of that shoulder

54. How to check if it's a hand or an elbow?

55. If the distance between the extrema and the shoulder is
less than a predefined value, then it is an elbow. Otherwise
it is a hand.

56. If it is a hand, we find the elbow by choosing the first point
(in the path we created with Dijkstra before) whose distance
exceeds the elbow distance mentioned before

58. There is still some things missing...

59. Future work

60. Hands from elbows: If one of the extremas is an elbow, we
need to infer where the hand is

61. Smoothing: Smooth the jittering of the joints

62. Robustness: Use restrictions to ignore objects that are not
the user

63. Multi-user: Track more than one person at a time

64. And of course, get the rest of the joints: hips, knees, etc.

65. How to use it?

66. Asynchronous API

67. SkeltrackSkeleton *skeleton = SKELTRACK_SKELETON (skeltrack_skeleton_new ());
skeltrack_skeleton_track_joints (skeleton,
depth_buffer,
buffer_width,
buffer_height,
NULL,
on_track_joints,
NULL);

69. Synchronous API

70. SkeltrackJointList list;
list = skeltrack_skeleton_track_joints_sync (skeleton,
depth_buffer,
buffer_width,
buffer_height,
NULL,
NULL);

71. Skeleton Joint:
ID: HEAD, LEFT_ELBOW, RIGHT_HAND, ...
x: X coordinate in real world (in mm)
y: Y coordinate in real world (in mm)
screen_x: X coordinate in the screen (in pixels)
screen_y: Y coordinate in the screen (in pixels)

72. Code/Bugs:
https://github.com/joaquimrocha/Skeltrack

73. Nifty Tools for Development:
GFreenect:
https://github.com/elima/GFreenect
GFreenect Utils:
https://github.com/joaquimrocha/gfreenect-utils

74. GFreenect Python Example

75. Tool: record-depth-file

76. Tool: depth-file-viewer

77. Questions?

78. Creative Commons pictures from flickr:
Kinect: Auxo.co.kr
Ampelmann: echiner1
Kid Playing: Rob Welsh
Skeleton: Dark Botxy