Motion Capture Sensor-Based Emotion Recognition Using a Bi-Modular Sequential Neural Network.

Area of Science:

• Computer Vision
• Affective Computing
• Human-Computer Interaction

Background:

• Gait emotion recognition (GER) is crucial for applications like smart homes and robotics.
• Current deep learning methods for GER often suffer from large model sizes and slow inference times.
• There is a need for efficient and accurate GER techniques.

Purpose of the Study:

• To propose a novel, lightweight bi-modular architecture for gait emotion recognition.
• To develop a computationally efficient method that overcomes limitations of existing deep learning approaches.
• To achieve high accuracy in inferring human emotions from gait patterns.

Main Methods:

• Developed a lightweight bi-modular architecture incorporating handcrafted features.
• Employed the RMSprop optimizer for model training.
• Utilized stratified data shuffling to enhance model robustness.
• Evaluated performance on the Edinburgh Locomotive Mocap Dataset.

Main Results:

• Achieved a micro-mean average precision of 0.97 on the Edinburgh Locomotive Mocap Dataset.
• Outperformed recent deep-learning methods in gait emotion recognition accuracy.
• Demonstrated a significantly reduced inference time of 16.3 milliseconds per gait sample.

Conclusions:

• The proposed lightweight architecture offers a highly effective and efficient solution for gait emotion recognition.
• This method provides a viable alternative to complex deep learning models, reducing overfitting and inference time.
• The findings have implications for emotionally aware robotics, adaptive therapy, and surveillance systems.