Data contain 3D hand trajectories collected with Leap Motion device. There are 13 subjects, each performs 26 interface-command gestures. Each gesture is encoded as a sequence of 3D points, representing the position of the dominant-hand forefinger. There are 26 classes corresponding to unique gestures. See Fig. 1 of [2] for the list of gestures and their visualisation.

1. Class 1: arc3Dleft
2. Class 2: arc3Dright
3. Class 3: caret
4. Class 4: check
5. Class 5: circle
6. Class 6: curly-bracket-left
7. Class 7: curly-bracket-right
8. Class 8: delete
9. Class 9: left-swipe
10. Class 10: pigtail
11. Class 11: poly3Dxyz
12. Class 12: poly3Dxzy
13. Class 13: poly3Dyxz
14. Class 14: poly3Dyzx
15. Class 15: poly3Dzxy
16. Class 16: poly3Dzyx
17. Class 17: rectangle
18. Class 18: right-swipe
19. Class 19: spiral
20. Class 20: square-bracket-left
21. Class 21: square-bracket-right
22. Class 22: star
23. Class 23: triangle
24. Class 24: v
25. Class 25: x
26. Class 26: zig-zag

We make three datasets out of these data, one for each dimension. We follow the original paper (see [2]) and use data of 8 subjects for training and 5 subjects for testing. Data of a same subjects does not appear in both train and test set, so resampling may introduce bias.

Data created by Fabio M. Caputo et al. (see [1], [2]). Data edited by Hoang Anh Dau.