Leland, Jake Mitchell (2019-05). Recognizing Seatbelt-Fastening Behavior with Wearable Technology and Machine Learning. Master's Thesis.
In the case of many fatal automobile accidents, the victims were found to have not been wearing a seatbelt. This occurs in spite of the numerous safety sensors and warning indicators embedded within modern vehicles. Indeed, there is yet room for improvement in terms of seatbelt adoption. This work aims to lay the foundation for a novel method of encouraging seatbelt use: the utilization of wearable technology. Wearable technology has enabled considerable advances in health and wellness. Specifically, fitness trackers have achieved widespread popularity for their ability to quantify and analyze patterns of physical activity. Thanks to wearable technology's ease of use and convenient integration with mobile phones, users are quick to adopt. Of course, the practicality of wearable technology depends on activity recognition--the models and algorithms which are used to identify a pattern of sensor data as a particular physical activity (e.g. running, sitting, sleeping). Activity recognition is the basis of this research. In order to utilize wearable trackers toward the cause of seatbelt usage, there must exist a system for identifying whether a user has buckled their seatbelt. This was our primary goal. To develop such a system, we collected motion data from 20 different users. From this data, we identified trends which inspired the development of novel features. With these features, machine learning was used to train models to identify the motion of fastening a seatbelt in real time. This model serves as the basis for future work in systems which may provide more intelligent feedback as well as methods for interventions in dangerous user behavior.