References

If you use the IAS-Lab Action Dataset, please cite the following works:

M. Munaro, G. Ballin, S. Michieletto, and E. Menegatti.
"3D Flow Estimation for Human Action Recognition from Colored Point Clouds."
In Biologically Inspired Cognitive Architectures (BICA), vol. 5, pp. 42-51, ISSN: 2212-683X, 2013

 M. Munaro, S. Michieletto, and E. Menegatti.
"An evaluation of 3D motion flow and 3D pose estimation for human action recognition." 
In RSS 2013 Workshop on RGB-D: Advanced Reasoning with Depth Cameras, Berlin (Germany), June 2013.

 

This dataset was used to test the action recognition algorithms in:

 

 

 

Downloads

All the samples in the IAS-Lab Action Dataset are provided as ROS bags.

 

Any sample contains:

  • point cloud (registered RGB and depth)
  • rgb image (RGB images at 640x480 resolution)
  • skeleton tracker joints (estimated by the NITE middleware)

 

Download the dataset in a compressed rar file: here.

Overview

The IAS-Lab Action Dataset contains 540 video samples:

  • 15 different actions;
  • 12 different people;
  • 3 different attempts.

 

We asked the subjects to perform well defined actions:

01 0001

1. check watch

02 0001

2. cross arms

03 0001

3. get up

04 0001

4. kick

05 0001

5. pick up

06 0001

6. point

07 0001

7. punch

08 0001

8. scratch head

09 0001

9. sit down

10 0001

10. standing

11 0001

11. throw from bottom up

12 0001

12. throw over head

13 0001

13. turn around 

14 0001

14. walk

15 0001

15. wave

 

For questions and remarks directly related to the IASLab Action Dataset please contact Questo indirizzo email è protetto dagli spambots. È necessario abilitare JavaScript per vederlo. or Questo indirizzo email è protetto dagli spambots. È necessario abilitare JavaScript per vederlo.