Enabling context aware data analysis for long-duration repetitive stooped work through human activity recognition in...

Area of Science:

• Occupational health and safety
• Biomechanics
• Machine learning for human activity recognition

Background:

• Repetitive occupational activities can increase injury risk due to changes in kinematics and neuromuscular control.
• Analyzing long-term biometric data for injury insights is challenging due to manual labeling difficulties and activity-dependent analysis.
• Accurate human activity recognition (HAR) is crucial for segmenting data and enabling biomechanical analysis.

Purpose of the Study:

• To develop a person-independent human activity recognition algorithm for sheep shearing.
• To enable the analysis of long-term biometric data for occupational injury risk assessment.
• To create a robust HAR classifier that accounts for inter- and intra-individual movement variations.

Main Methods:

• Utilized a Hidden Markov Model (HMM) for human activity recognition.
• Incorporated physical features relevant to spinal movement quality for classification.
• Developed a person-independent classifier to address inter-individual differences.
• Trained and validated the algorithm on sheep shearing activities.

Main Results:

• The developed HAR algorithm achieved a high F1 score of 96.47% for identifying the sheep shearing task.
• The person-independent classifier demonstrated robustness to variations in movement.
• The HMM-based approach effectively segmented activities within long-term biometric data.

Conclusions:

• The person-independent HAR algorithm is effective for analyzing biomechanical data in occupational settings like sheep shearing.
• This approach facilitates the study of kinematic and neuromuscular changes related to injury risk over time.
• Accurate activity segmentation through HAR is vital for understanding and mitigating occupational injuries.