Development and validation of smartwatch-based activity recognition models for rigging crew workers on cable logging...

Area of Science:

• Forestry research
• Occupational safety
• Wearable sensor technology

Background:

• Traditional time studies offer limited detail in forestry work analysis.
• High-resolution inertial sensors and GNSS data from wearables present new modeling opportunities.
• Quantifying rigging crew activities is crucial for safety and efficiency.

Purpose of the Study:

• To evaluate smartwatch-based activity recognition models for quantifying rigging crew tasks.
• To assess the feasibility of using wearable sensors in cable logging operations.
• To establish a foundation for real-time safety notifications in forestry.

Main Methods:

• Collected accelerometer data (25 Hz) from smartwatches worn by choker setters and chasers.
• Utilized Random Forest machine learning to develop predictive activity models.
• Extracted features using sliding windows (1-15s) and overlap levels (0-90%).

Main Results:

• The best choker setter model (3s window, 90% overlap) achieved high sensitivity (76.95-83.59%).
• The best chaser model (1s window, 90% overlap) showed good sensitivity (71.95-82.75%).
• Model performance varied by activity and window parameters, demonstrating feasibility.

Conclusions:

• Smartwatch activity recognition is a feasible method for quantifying forestry work.
• This approach can advance the analysis of health and safety metrics in logging.
• The findings support the development of real-time safety alerts for high-risk forestry jobs.