Figure 1.

System overview. The first phase is the training phase, while the second phase represents the testing phase. In the training phase, 3D poses are normalized first by removing orientation and translational information from each pose, thus developing the normalized pose space. The joint features are extracted from these normalized poses and are given as input to the deep neural network in order to learn the model. The testing phase includes normalization, keyframe extraction and then the extraction of joint features, on the basis of which the action is recognized with the help of the learned model.