Edge-based Human Activity Recognition System for Smart Healthcare

Human activity recognition (HAR) is the method of detecting the physical activity of a person. It has a huge scope in the medical domain for supervision and health analysis. With the help of artificial intelligence, it can be performed using regularly available smartphone devices. For healthcare, HAR is often a part of an IoT framework. Using a cloud-based IoT system ensures maximum resource usage and data storage but comes with the challenges of high latency and bandwidth consumption. To avoid this, an edge-based system has been proposed in this work, which ensures minimum interaction of the device with the cloud and also makes emergency actions possible to be taken due to low latency of the system. A comparative study has been done on different state-of-the-art machine learning models. Also, to ensure minimum resource requirement, optimized neural network models have been used, and it has been shown how they require less storage as compared to traditional methods.