Designing a Computer-Vision Application: A Case Study for Hand-Hygiene Assessment in an Open-Room Environment

Area of Science:

• Computer Vision
• Food Safety Engineering
• Machine Learning

Background:

• Hand-hygiene is crucial for safe food handling.
• Existing systems struggle with diverse scenarios in food environments.
• Need for robust automated monitoring of hygiene practices.

Purpose of the Study:

• To design and evaluate a hand-hygiene action detection system.
• To improve food-handling safety using computer vision.
• To address limitations of single-modality systems.

Main Methods:

• Iterative engineering process for system development.
• Baseline analysis using RGB-only Convolutional Neural Network (CNN).
• Development of a hierarchical system integrating RGB, optical flow, hand masks, and skeleton joints.

Main Results:

• Baseline CNN showed poor performance across different scenarios.
• Hierarchical system demonstrated effectiveness in detecting hand-hygiene actions.
• System validated using hand-washing videos from a commercial kitchen.

Conclusions:

• Multi-modal hierarchical systems are effective for hand-hygiene action detection.
• The developed system can improve food safety monitoring.
• Recommendations provided for real-world computer vision system design.