Reliable positioning-based human activity recognition based on indoor RSSI changes

Abstract

In this article, a human activity recognition system based on Wi-Fi signal strength variation (SSV) has been proposed. This strategy is built by exploiting the known fact that radio signal significantly reacts when it interfaces with the human body by causing fading and shadowing effects. Different irregularities in the received signal strength indicator (RSSI) propagation patterns indicate individual human activities. In the proposed method, utilizing the received RSSIs from various access points (APs) of known locations to the smartphone carried by a human, first, the position of the human is localized with the distances utilizing half the number of APs’ based on the strong RSSI values. Then, using the strongest RSSIs of the nearest AP, the activity of the human is recognized using the changing signal strengths. To accurately measure the monotonic distances by the RSSI values, the regression analysis technique (RAT) is used in the path loss model (PLM) to mitigate error significantly. Besides, to classify human activities, we calculate the deviation between any human activity and no human. Moreover, we arrange all activities in a successive order. With this infrastructure, we can develop a system where both human localization and activity recognition can be done within a single setup, which not only detects the position of a person on the floor but also produces the health condition of each person staying on the floor. In the existing methods, wirable devices are used to detect human activities, which creates irritations when they have to carry some heavy electronic device attached to their body. Moreover, these devices are expensive. On the other hand, channel state-based solutions have some advantages over wirable systems, but this technology does not support in major smartphones. So, in this work, to overcome such challenges, we have focused on an RSSI-based framework that does not need to wear electronic devices on the body as well as supports every smartphone. So, with a simple setup, the system can be operated. Our system can successfully recognize at most five activities simultaneously for the presence of the same humans in the experimental indoor premises. Such an approach enhances the interactions in intelligent healthcare systems.