WiFi-assisted human activity recognition

This paper investigates the indoor activity recognition issue and proposes a novel recognition framework by exploring WiFi ambient signals. The key idea is to use data mining techniques to abstract footprints of different activities on the radio signal strength (RSS) data. Our experiments show that even using a single feature and the common k-NN classifier activities such as walking, sitting and standing can be recognized with a high accuracy, i.e. 75%. To further improve the performance, a new feature has been abstracted to represent the fluctuation of sampled data and a novel algorithm named fusion algorithm has been specifically designed based on the classification tree. Experiments show that the proposed fusion algorithm significantly outperforms the k-NN classifier in terms of both the average recognition ratio (from 75% to 92.58%) and the computational complexity. Compared to previous solutions relying on either special hardware or the cooperation of tested subjects, the proposed recognition framework is a passive and device-free solution that could be integrated into any WLAN network with low overheads.