A Lightweight Vision-Based Emotion Sensing Framework for Assistive Healthcare Robotics.

Area of Science:

• Robotics
• Computer Vision
• Human-Computer Interaction

Background:

• Facial expression recognition (FER) is crucial for assistive robotics but faces challenges like unstable landmark normalization and limited dataset variability.
• Existing methods often struggle with resource-constrained environments and privacy concerns due to reliance on raw RGB images.

Purpose of the Study:

• To develop a lightweight, privacy-preserving FER framework for assistive healthcare robotics using geometric facial landmarks.
• To enhance recognition robustness and suitability for low-power edge devices.

Main Methods:

• Proposed a novel nose-centered landmark normalization method, replacing expression-sensitive references with a stable nose-center anchor.
• Introduced an optimized Facial Feature Mapping (FFM-L03) combining anatomical priors and statistical methods to select 60 informative landmarks.
• Constructed a heterogeneous Freepik dataset to increase variability in lighting, background, resolution, and subject appearance.

Main Results:

• The proposed method consistently improved performance across 15 landmark groups, four datasets, and four classifiers.
• Achieved significant accuracy gains of up to 22.4 percentage points over the Ciraolo baseline and 22.1 percentage points over the full-landmark baseline.
• Demonstrated lightweight operation suitable for real-time, privacy-aware assistive robotic systems.

Conclusions:

• Principled normalization and targeted landmark selection substantially improve FER for real-time, privacy-aware assistive robotic systems.
• The developed framework offers a robust and efficient solution for emotion recognition in healthcare robotics.