Efficient Real-time On-the-edge Facial Expression Recognition using Optomyography Smart Glasses

Understanding emotional states through facial expressions plays a pivotal role in various domains, from mental health assessments to human-computer interaction. In this study, we present an algorithm tailored for efficient, real-time, on-the-edge monitoring of key facial expressions, utilizing the data from novel optomyography (OMG) based smart glasses - OCOsense™. The expressions targeted in this study include low-and high-intensity variation of: smile, frown, eyebrow raise, and winks, facilitating a comprehensive understanding of key facial expressions essential for emotional assessment. The algorithm, developed and evaluated on a large dataset of 40 participants, prioritizes low power consumption and fast inference times. After performing optimization steps that include feature selection and model hyper-parameter tuning, our final prototype achieved a compact code size of 71.3 kB and an accuracy of 93.8%, all while maintaining power consumption at 0.43 watts.