Abstract
Availability of real-time location sharing devices that use global navigation satellite system (GNSS) positioning paired with radio frequency (RF) transmission (GNSS-RF) and wearable devices equipped with inertial measurement unit (IMU) and other sensors provide opportunities to improve occupational safety in forestry using new techniques and methodologies based on human activity recognition. Forestry is among the most hazardous professions in the United States. In particular, logging and wildland firefighting occur in remote, off-grid environments that often lack traditional cellular communications infrastructure and involve frequent interactions among ground workers, heavy equipment, and dynamic terrain- and weather-related hazards. GNSS-RF location sharing, geofencing, mesh networking, and wearable-based human activity recognition modeling can increase situational awareness (SA) among these workers. The overall goals of this dissertation are to assess the factors affecting the performance and accuracy of a variety of location sharing networks and to demonstrate the feasibility of using wearable sensors to quantify forestry work activities. The work consists of four chapters. In the first, I develop the concept of GNSS-RF mobile geofences and model the intersections of mobile and stationary geofences in order to characterize the factors affecting the timing of intersection alerts. This field study provides the basis for evaluating the feasibility of using real-time safe work areas and incident avoidance alert systems for person-to-person and machine-to-machine interactions in dynamic forestry environments. The second chapter evaluates the effects of forest stand characteristics, topography, and line-of-sight (LOS) obstructions on radio signal propagation, positional accuracy, and geofence alert timing using a network of GNSS-RF transponders. The third chapter assesses the overall performance of smartphone-based GNSS-RF mesh networks and develops Dirichlet regression models to predict network connectivity using lidar and satellite remote sensing data. In the final chapter, I present the first use of wearable sensors to develop human activity recognition models that quantify occupational work in forestry using machine learning. Collectively, this research provides the basis for using real-time GNSS-RF based location sharing, geofencing, and wearable-based human activity recognition systems to improve SA and inform smart alerts to reduce fatal and near-fatal incidents among natural resource professionals.