10:00 - 10:30 | Mon 25 Sep | Ballroom Foyer | MoAmPo.28
Due to the great success of convolutional neural network(CNN) on image classification problems, several attempts have been made to train CNN for human action recognition problem. But since CNN is designed for static RGB images, it is not easy for it to learn temporal information from videos. To tackle this problem, temporal kinematics features are proposed, which compute the linear velocity and orientation displacement based on the skeleton data. The proposed temporal kinematics features are then encoded into images, which will be trained by a CNN. Experiment results show that the proposed method is fast to train and it performs quite well when compared with RGB frames.