Deep Learning Approach for Estimating a Human Pose on a Mobile Device

Abstract

Current deep convolutional neural networks (CNNs) approaches enable accurate marker-free estimation of body poses observed in an image. We think that this marker-free estimation approach develops all its potential in the field of pose estimation, if this is possible in real time and on a mobile device. In this work, we focus on the implementation of a modern CNN approach for the body pose estimation and a redesign of the CNN architecture, so that it can be applied on a mobile device. The results of our experiments show that even current smartphones can propagate our architecture in reasonable time.