This UOW Online Action3D Dataset consists of action sequences of skeleton videos, the 20 actions are from the original MSR Action3D Dataset. The action videos are recorded by Microsoft Kinect V.2 with average 20fms/s frame rate.
There are 20 participants to perform these actions, every participant performs each action according to his/her personal habits. For each participant, he/she first repeats each action 3--5 times, then performs 20 actions continuously in a random order. These continuous action sequences can be used for online action recognition testing. The repeated action sequences will be used for training.
In order to make the dataset can be used for cross dataset test, the 20 participants perform the actions in 4 different environments.
Please cite the following papers if you use the dataset.
Chang Tang, Wanqing Li, Pichao Wang, Lizhe Wang, Online Human Action Recognition Based on Incremental Learning of Weighted Covariance Descriptors, Information Sciences, 2018 (to appear).