Free-Weight Exercise Activity Recognition using Deep Residual Neural Network based on Sensor Data from In-Ear Wearable Devices
Sakorn Mekruksavanich, Anuchit Jitpattanakul · 2023
Human activity recognition (HAR) is a popular study area in the current era of the Internet of Things and artificial intelligence. HAR approaches have been successfully applied in many real-world scenarios, such as sports tracking and assessment and remote healthcare for the elderly. Over the past decade, studies on HAR employing wearable sensors have been explored operating various wearable devices, including smartwatches, smart shoes, and other intelligent wearables. The advancement of sensor technology has directed the combining of inertial sensors into ear-worn devices, making it possible to record physical movements privately. In this study, we proposed a deep residual network named ResNeXt, for identifying free- weight exercises using sensor data from smart in-ear devices. To consider the performance of the proposed model and other deep learning models, we conducted experiments using a publicly available dataset called ERICA, which collected free-weight exercise activities from inertial sensors of both an in-ear device and a dumbbell. Our experimental results demonstrated that the suggested ResNeXt network surpassed other deep learning models, gaining the highest F1-score of 96.67% and 99.75% using inertial data from in-ear sensors and dumbbell sensors, respectively