CLUMM: Contrastive Learning for Unobtrusive Motion Monitoring.

Area of Science:

• Computer Vision
• Machine Learning
• Human-Computer Interaction

Background:

• Traditional human monitoring and motion recognition heavily rely on wearable sensors, causing worker discomfort and obtrusiveness.
• Camera-based systems offer unobtrusive sensing but require extensive data labeling and struggle with environmental noise, hindering deep learning performance.

Purpose of the Study:

• To develop a novel framework for effective human motion recognition using contrastive learning.
• To overcome limitations of existing methods by eliminating the need for manual data labeling and reducing environmental interference.

Main Methods:

• A contrastive learning framework is proposed to learn rich data representations from raw images without manual annotation.
• The model focuses on critical human joint coordinates, minimizing the impact of irrelevant environmental factors.
• A custom dataset simulating workplace tasks was utilized for training and evaluation.

Main Results:

• The contrastive learning approach successfully learned relevant human-specific features from images.
• Fine-tuning the model for motion classification achieved up to 90% accuracy.
• The method demonstrated effective real-time human motion monitoring with reduced environmental influence.

Conclusions:

• The proposed contrastive learning framework offers an effective, unobtrusive solution for human motion recognition in industrial settings.
• This approach significantly reduces the need for manual data labeling and enhances model robustness against environmental complexity.
• The high accuracy achieved validates the framework's potential for real-time worker monitoring and performance analysis.