Non-intrusive Stress detection based on temporal emotion analysis in videos applying machine learning

Abstract

Stress has become an embedded part of daily life in modern work environment. Continuous exposure to stress affects physical and mental health of an individual. Early detection and treatment of stress is the only way to avoid long term health issues. This work proposes a non intrusive method for stress detection from videos captured in work environment. A model for correlating stress to the facial and posture clues is proposed. Visual features extracted from facial clues and posture features are mapped to emotion state using machine learning classification. From the sequence of emotion observed over a period of time, stress level is classified using deep learning LSTM classifier.