Sensing Room Occupancy Levels with Signal-to-Noise Ratio and Signal Phase and Multiple Antenna Configurations

We propose a novel machine learning based method for estimating the number of people present in a room (e.g, in a shared office space) based on WiFi signal- to- noise ratio and signal phase data provided by WiFi Channel State Information compatible hardware. We apply random decision forests machine learning and show that the precise number of people can be estimated with a score of 0.66 and the occupancy levels (empty, low, high) with a score of 0.87 at an affordable cost. We evaluate our approach in two settings: one small room with 0–2 and in a medium-sized office space with 0–8 people performing their usual office desk work. Beyond determining maximum recognition rates we systematically investigate the impact of different design choices (antennas, training data) on system performance. The proposed method outperforms a statistical baseline method significantly.