Options
2018
Conference Paper
Titel
Learning Optical Flow for Action Classification
Abstract
Action recognition is the task of assigning labels to human actions. It is particularly important for service robots because they need to react to human actions. For instance, if the user is cooking, the robot can offer to bring him some tools such as a pan or knife. In this work, the possibility of learning action recognition and optical flow extraction simultaneously using 3D Convolutional Neural Networks is analyzed. The preliminary results show that it is possible to learn two tasks together, but the proposed architecture needs further improvements.