Classification of human poses and orientations with deep learning
Anıl Atvar, Nazlı İkizler-Cinbiş · 2018
Within the scope of this study, we aim to classify human poses and orientations from the group activity images using deep learning. In the framework that we developed, the detection, pose and orientation classification steps are performed in a cascade fashion. Firstly, people in the images are detected, then, the detected people are classified as belonging to one of the classes “standing”, “sitting on an object” and “sitting on the ground” and finally classified into one of the eight different orientations of these three pose classes. To this end, an end-to-end trainable deep learning framework is used. The experimental evaluation show that the trained Convolutional Neural Network model produces successful results.