Mastering Computer Vision with TensorFlow 2.x

Action recognition using the 4D method

4D action recognition means that the three-dimensional actions of volume representation work as a function of time. Think of this as volume-tracking an action. Quanzeng You and Hao Jiang proposed a novel 4D method titled Action4D: Online Action Recognition in the Crowd and Clutter. The details of this paper can be found at http://openaccess.thecvf.com/content_CVPR_2019/html/You_Action4D_Online_Action_Recognition_in_the_Crowd_and_Clutter_CVPR_2019_paper.html.

This method tracks humans using 4D representations and recognizes their actions in a cluttered and crowded environment. The concepts of the paper are as follows:

• A three-dimensional point cloud is created for each scene using multiple RGBD images. ...