Activity Recognition in Smart Homes using UWB Radars

Kevin Bouchard, Julien Maitre, Camille Bertuglia, Sébastien Gaboury
2020 Procedia Computer Science  
In the last decade, smart homes have transitioned from a potential solution for aging-in-place to a real set of technologies being deployed in the real-world. This technological transfer has been mostly supported by simple, commercially available sensors such as passive infrared and electromagnetic contacts. On the other hand, many teams of research claim that the sensing capabilities are still too low to offer accurate, robust health-related monitoring and services. In this paper, we
more » ... e the possibility of using Ultra-wideband (UWB) Doppler radars for the purpose of recognizing the ongoing ADLs in smart homes. Our team found out that with simple configuration and classical features engineering, a small set of UWB radars could reasonably be used to recognize ADLs in a realistic home environment. A dataset was built from 10 persons performing 15 different ADLs in a 40 square meters apartment with movement on the other side of the wall. Random Forest was able to attain 80% accuracy with an F1-Score of 79%, and a Kappa of 77%. Those results indicate the use of Doppler radars can be a good research avenue for smart homes. Abstract In the last decade, smart homes have transitioned from a potential solution for aging-in-place to a real set of technologies being deployed in the real-world. This technological transfer has been mostly supported by simple, commercially available sensors such as passive infrared and electromagnetic contacts. On the other hand, many teams of research claim that the sensing capabilities are still too low to offer accurate, robust health-related monitoring and services. In this paper, we investigate the possibility of using Ultra-wideband (UWB) Doppler radars for the purpose of recognizing the ongoing ADLs in smart homes. Our team found out that with simple configuration and classical features engineering, a small set of UWB radars could reasonably be used to recognize ADLs in a realistic home environment. A dataset was built from 10 persons performing 15 different ADLs in a 40 square meters apartment with movement on the other side of the wall. Random Forest was able to attain 80% accuracy with an F1-Score of 79%, and a Kappa of 77%. Those results indicate the use of Doppler radars can be a good research avenue for smart homes.
doi:10.1016/j.procs.2020.03.004 fatcat:z2nywuid3vdo7nvwwrw62qi7mu