ProxiTrak: A robust solution to enforce real-time social distancing & contact tracing in enterprise scenario

Abstract

In the wake of recent COVID-19 pandemic, contact tracing has turned out to be an indispensable technique to help administrative authorities contain localized infections efficiently. In the absence of a definitive and an official vaccine for the infection, practicing social distancing has proved to be an effective norm to prevent the risk of infection. In this paper, we present 'ProxiTrak', a smartphone based solution for an enterprise scenario capable of not only tracing the chain of possible infection transmission among a set of population, but also guiding the users towards following social distancing norms by alerting them in real-time about any possible violation of proximity norms on their smartphones. We devise an effective classification model to make proximity decisions on the smartphone itself using Received Signal Strength Indicator (RSSI) data of on-board Bluetooth Low Energy (BLE) module using multiple mobile devices in different environments, with novel addition of using temporal features from BLE data to boost the model's accuracy. We briefly discuss ProxiTrak's corresponding server-side framework for tracing a possible chain of infection and analysing social connectivities graphically. We also propose on-device decision aggregation and server-side pruning of proximity events to lower the false positive events. Our model is capable of making strong proximity decisions to an accuracy of up to 94% on the devices trained with the model.