Yawning Detection using Earphone Inertial Measurement Units

Yawning is a key and reliable indicator of fatigue, and detecting fatigue is vital in scenarios ranging from safety-critical situations for preventing performance impairment, to work environments for promoting timely breaks, leading to enhanced worker healthiness and productivity. State-of-the-art yawning detection studies face several limitations, such as privacy concerns, high costs, and lack of portability. In this paper, we conduct a feasibility study on enabling a privacy-resistant, low-cost, and portable solution to detect yawning by leveraging earphones equipped with inertial measurement units (IMUs), with the aim of benefiting future fatigue detection methods. We employ a range of preprocessing methods and develop 5 neural networks along with 3 classical machine learning (ML) approaches based on our initial research into the patterns within earphone IMU data from yawning and various activities. We collect data from 10 participants wearing a headphone with an IMU and evaluate the performance of our models on both the collected dataset and a public dataset. The results show Fi scores of up to 0.90 on the collected dataset and 0.71 on the public dataset, which indicate the feasibility of yawning detection from earables.