Presentation:
Different from existing dynamic hand gesture datasets, LMDHG contains unsegmented sequences of hand gestures performed with either one hand or both hands. This dataset can therefore be used for recognition of both pre-segmented and unsegmented dynamic hand gestures using skeleton data.
Download the database
Download database information
Content:
There were 21 participants, each participant performed at least one sequence, resulting in 50 sequences. Each sequence contains 13 ± 1 class gestures leading to a total of 608 gesture instances. At the end of each gesture, the participant was asked to keep his hands above the Leap Motion before performing an other gesture. We labelled this no-gesture as an idle class. Furthermore, the order of classes in each sequence is aleatory and is different from one sequence to another.
Each frame contains the 3D coordinates of 46 joints (23 joints for each hand). If one of the hands is not tracked then the position of its joints are set to zero. The Leap Motion sensor is used to collect our dataset. We provide also the labels and ground truth start and end of each gesture class in each sequence. The gestures composing our dataset are listed in Table bellow.
| Gesture | #Hands | tag name |
| Point to | 1 | HG1 |
| Catch | 1 | HG2 |
| Shake with two hands | 2 | HG3 |
| Catch with two hands | 2 | HG4 |
| Shake down | 1 | HG5 |
| Shake | 1 | HG6 |
| Draw C | 1 | HG7 |
| Point to with two hands | 2 | HG8 |
| Zoom | 2 | HG9 |
| Scroll | 1 | HG10 |
| Draw Line | 1 | HG11 |
| Slice | 1 | HG12 |
| Rotate | 1 | HG13 |
Reference:
This dataset can be used freely for research purpose. However, any published work using it should refer to:
Boulahia, S. Y., Anquetil, E., Multon, F., & Kulpa, R. Dynamic hand gesture recognition based on 3D pattern assembled trajectories. In 7th IEEE International Conference on Image Processing Theory, Tools and Applications (IPTA 2017), (pp. 1-6).
Acknowledgment:
We would like to thank all those who have participated in collecting the LMDHG dataset.
Related publications:
[1] Boulahia, S. Y., Anquetil, E., Kulpa, R., & Multon, F. (2016, December). HIF3D: Handwriting-Inspired Features for 3D skeleton-based action recognition. In Pattern Recognition (ICPR), 2016 23rd International Conference on (pp. 985-990). IEEE.
[2] Boulahia, S. Y., Anquetil, E., Multon, F., & Kulpa, R. Dynamic hand gesture recognition based on 3D pattern assembled trajectories. In 7th IEEE International Conference on Image Processing Theory, Tools and Applications (IPTA 2017), (pp. 1-6).
Contact:
If you have any questions or suggestions, you can contact Eric Anquetil (eric.anquetil@irisa.fr)
