Dynamic Facial Stress Recognition in Temporal Convolutional Network

Stress is a major problem that infiltrates our society in countless ways. We cannot eliminate stress, but can recognize stress and manage it. Automatically recognizing stress through facial expressions has been extensively studied in the past decades. Recent research indicates that certain architectures can reach state-of-the-art accuracy in stress recognition. However, they recognise facial stress in view of static expressions, while only a few papers identify the fundamental limitations of static facial expression. This paper adapts ANUStressDB database in dynamic and develops a Temporal Convolutional Network to recognize continuous facial stress problem. We further apply Bimodal Distribution Removal to improve our result. The experimental results show that our system achieves 67.56% classification accuracy.