Deep Learning Approaches for Unobtrusive Human Activity Recognition using Insole-based and Smartwatch Sensors

In this digital age, human activity recognition (HAR) has been playing an increasingly important role in almost all aspects of life. Healthcare systems for monitoring the activities of daily living (ADL), security environments for automatically detecting abnormal activity and notifying the pertinent authorities, and increasing human contact with computers are all examples of HAR applications. The data collecting tools used in HAR research can also be categorized (sensors or cameras). In sensor-based HAR, various wearable sensors have been investigated for their usefulness in effectively detecting both simple and complex human activities. In this work, we studied HAR using signal data acquired from insole-based and smartwatch sensors. To achieve our research goal, we proposed a ResNeXt inspired deep learning (DL) network called HARNeXt. The proposed model was evaluated on a public data set named 19NonSense data that collected activity signal data recorded from insole-based and smartwatch sensors. In addition, we compared the proposed model to baseline DL models. The results showed that the HARNeXt outperformed other DL models and achieved the highest F1-score of 97.03%.