Infant Facial Expression Analysis: Towards a Real-Time Video Monitoring System Using R-CNN and HMM

Area of Science:

• Biomedical Engineering
• Computer Vision
• Pediatrics

Background:

• Manual monitoring of infants is crucial due to their inability to articulate discomfort.
• Diseases like gastroesophageal reflux require careful observation of infant states and expressions.
• Objective, real-time analysis systems can significantly aid clinical diagnosis and care.

Purpose of the Study:

• To develop and evaluate a video-based system for real-time infant expression and state analysis.
• To correlate detected infant discomfort with potential medical conditions like gastroesophageal reflux.
• To improve the accuracy and efficiency of infant monitoring for healthcare.

Main Methods:

• A system combining Fast R-CNN for expression detection with Hidden Markov Models for compensated detection.
• Utilizing object tracking and analyzing information from previous frames for robust detection.
• Evaluating the system on both clinical and daily datasets for expressions (discomfort, unhappy, joy, neutral) and states (sleep, pacifier, open mouth).

Main Results:

• Achieved a mean average precision of 81.9% for 4 infant expressions and 84.8% for 3 states.
• Demonstrated high accuracy for discomfort detection, reaching up to 90% average precision.
• The system showed effective performance in real-time analysis across diverse datasets.

Conclusions:

• The proposed video-based system offers a promising approach for objective, real-time infant monitoring.
• Accurate detection of infant discomfort can support the diagnosis of conditions like gastroesophageal reflux.
• This technology has the potential to enhance infant healthcare and parental support.