Integration of TinyML-based proximity and couch sensing in wearable devices for monitoring infectious disease's social distance compliance

With the advent of artificial intelligence (AI) and Internet of Things (IoT), there has been a rapid increase in the use of sensors to intelligently monitor the environment and movement of objects. Smart solutions have been widely used for monitoring infectious diseases by limiting the transmission of contagious diseases using proximity sensing systems. This is an alternative to conventional social distancing technologies like Bluetooth and cameras which uses machine learning (ML), image processing to identify trespassers, and multiple object detection in real-time. This paper leverages the emerging Tiny ML technology to design and develop a wearable device that can prevent infectious diseases from spreading. The device senses the cough sound of the nearest person within a limited distance and then identify the nearest objects such as humans, animals (dog, goats), and wind-blown vegetation, based on patterns of PIR signals bounced back from different objects. By using machine learning algorithms, the device can be able to notify the user when they are in a safe environment or not. This solution is a wearable device that has the potential to be used in monitoring the transmission of contagious diseases by detecting and identifying moving objects and alerting people to keep their distance when they are in an unsafe environment with a high risk of being exposed to the disease. This work-focused research project will particularly focus on monitoring the risk environment to prevent infectious diseases between humans and between humans and animals, reminding users to keep their distance for their safety and the use of the Convolutional Neural Network (CNN) algorithm on the device for identifying moving objects and for detecting cough. The system has been evaluated, and the experiments have shown a performance accuracy of 92.1% for object detection and 68% for cough detection, promising for detecting a safe environment. This accuracy could be increased over time via reinforcement learning.