Deep-Learning-Based Stress Recognition with Spatial-Temporal Facial Information.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Growing interest in stress management has spurred research into stress recognition.
• Existing methods rely on physiological signals (requiring devices) or handcrafted facial features (limited discrimination).

Purpose of the Study:

• To develop a robust deep learning-based stress recognition method using facial images.
• To overcome limitations of existing physiological and handcrafted feature-based approaches.

Main Methods:

• Constructed a large-capacity image database for stress recognition.
• Employed a deep learning model incorporating temporal and spatial attention mechanisms.
• Integrated facial landmark information with the facial image analysis network.

Main Results:

• The proposed deep learning method demonstrated superior performance on the newly constructed database.
• Attention mechanisms effectively weighted salient temporal frames and spatial regions related to stress.
• Inclusion of facial landmarks further improved recognition accuracy.

Conclusions:

• The proposed deep learning model offers a promising, device-free approach for accurate stress recognition.
• The integration of attention mechanisms and facial landmarks enhances the model's ability to detect stress from facial images.