Passive Crowd Speed Estimation and Head Counting Using WiFi

Abstract

In this paper, we propose a framework to sense occupancy attributes of an area, such as speed of a crowd traversing through the area, the total number of people in the area, and the rate of arrival of people into the area, using only the received power measurements (RSSI) of two WiFi links, and without relying on people to carry any device. We first show that the cross-correlation between the two WiFi link measurements and the probability of crossing a link implicitly carry key information about the occupancy attributes and develop a mathematical model to relate these parameters to the occupancy attributes of interest. Based on this, we then propose a system to estimate the occupancy attributes and validate it with 51 experiments in both indoor and outdoor areas, where up to (and including) 20 people walk in the area with different possible speeds, and show that our framework can accurately estimate the occupancy attributes. For instance, our framework achieves a Normalized Mean Square Error (NMSE) of 0.047 (4.7%) when estimating the speed of a crowd, an NMSE of 0.034 (3.4%) when estimating the arrival rate to the area, and a Mean Absolute Error (MAE) of 1.3 when counting the total number of people. We finally run experiments in an aisle in Costco, showing how we can estimate the key attributes of buyers' motion behaviors.