Wearable IoT Sensor Combining Deep Learning for Enhanced Human Activity Recognition in Indoor and Outdoor Settings

Human Activity Recognition (HAR) plays a pivotal role within a broader framework aimed at continuously monitoring human behaviors across various domains, including medical diagnosis, elderly care, rehabilitation, entertainment, and smart surveillance. This paper introduces an innovative HAR system that leverages the capabilities of wearable devices in conjunction with deep learning techniques. Its primary objective is to accurately identify physical activities performed by individuals in both indoor and outdoor environments. The designed wearable sensor incorporates an Inertial Measurement Unit (IMU) and a Wi-Fi module, enabling data transmission to a cloud service and providing direct Internet connectivity. This sensor is integrated with a Convolutional Neural Network (CNN) architecture optimized for resource-efficient inference, making it suitable for deployment on cost-effective or embedded devices. Additionally, it enables real-time local activity prediction. The system is custom-tailored for physical activity monitoring and achieves an impressive accuracy rate of 98,8% in distinguishing various activities.