Start Date: April 22, 2019
Ioannis Vernikos
This thesis aims to explore, implement and evaluate new methodologies for the recognition
of human behavior. These techniques are based on novel representations of visual data, as
well as models of deep neural networks. Techniques of fusion of color information, as well as
depth information, will also be investigated.
The novelties of the proposed thesis are as follows:
● Novel deep learning neural network architectures will be presented and will be
applied to human action recognition problems
● Novel 3D skeletal information representation methodologies for the training of neural
networks will be proposed
● An assessment in a real-like environment, will take place which will aim at identifying
daily activities