Transport mode recognition for smart eyewear using multimodal audio and accelerometer data

Wearable devices have enabled a wide range of sensing applications, including activity recognition such as walking, cycling, or running, among others. Despite the substantial advances in the research field of human activity recognition (HAR), the recognition of complex activities such as transportation modes (TM) requires more investigation. In this paper, we present a sound-accelerometer-based approach for transport mode detection through a wearable instrumented frame. Our experimental results demonstrate the robustness of the multimodal strategies compared to the monomodal ones that only rely on the accelerometer or sound data. Furthermore, we assessed the resource memory requirements of our proposed multimodal strategies to consider their integration within a microcontroller unit (MCU) for real-time detection.