Description
Facial emotion recognition (FER) has attracted the attention of many researchers for its promising applications. In the factories of the future (FoF),
human operators may be supervised over large periods, not only to predict and avoid misuse conduct but also to predict and stop malicious actions.
Thus, an intelligent system is needed for safety and security purposes.
This thesis investigates the use of facial emotion systems in real-life situations. Computer vision and deep learning have become extremely popular
technologies for mapping facial expressions to emotional states. Therefore, we propose two deep learning methods to analyze the performance of facial
emotion systems in controlled and uncontrolled environments. The first one is concentrated on supervised learning using convolutional neural networks (CNNs)
with a transfer learning technique. The second method is a self-supervised approach inspired by contrastive learning of visual representations. The
suggested FER techniques are verified on the three facial image datasets: KDEF, RAF-DB, and JAFFE. The evaluations of the experiments reveal that the first
FER approach achieves better accuracies than the second one. However, both methods show satisfactory results and demonstrate the prospect of using FER
for real-life applications.
|
Thesis topic in co-operation with the Fraunhofer Institute for Applied and Integrated Security AISEC, Garching