Human action recognition has focused more on three-dimensional data as the cost of the wearable devices and depth cameras reduce. The main challenge of the human action recognition is to find a suitable representation of joint actions for characterizing human actions. In this study, we try to identify and recognize the human actions using 3D joint data collected with Kinect devices. Instead of using raw joint positions, we use temporal and spatial joint relation information to recognize human movement. To accomplish this, we define the bag of words classification as joint features. We utilize popular human action datasets as well as our own dataset collected with Kinect 2 as a part of this work.
Figure 1: Overview of our work