Fast and Robust Stationary Crowd Counting with Commodity WiFi

TL;DR

This paper presents a WiFi-based method that passively estimates seated crowd sizes by analyzing natural body fidgeting behaviors, using signal bandwidth as a robust indicator, and validated through extensive experiments.

Contribution

It introduces a novel bandwidth-based crowd counting model leveraging body fidgeting, with an anomaly detection module for robustness, outperforming existing methods in speed and accuracy.

Findings

Achieved a mean absolute error of 1.04 in crowd estimation

Reduced convergence time to 51 seconds, faster than prior methods

Validated scalability to larger crowds through simulations

Abstract

This paper introduces a novel method for estimating the size of seated crowds with commodity WiFi signals, by leveraging natural body fidgeting behaviors as a passive sensing cue. Departing from prior binary fidget representations, our approach leverages the bandwidth of the received signal as a finer-grained and robust indicator of crowd counts. More specifically, we propose a mathematical model that relates the probability density function (PDF) of the signal bandwidth to the crowd size, using a principled derivation based on the PDF of an individual's fidget-induced bandwidth. To characterize the individual fidgeting PDF, we use publicly available online videos, each of a seated individual, from which we extract body motion profiles using vision techniques, followed by a speed-to-bandwidth conversion inspired by Carson's Rule from analog FM radio design. Finally, to enhance robustness in real-world deployments where unrelated motions may occur nearby, we further introduce an anomaly detection module that filters out non-fidget movements. We validate our system through 42 experiments across two indoor environments with crowd sizes up to and including 13 people, achieving a mean absolute error of 1.04 and a normalized mean square error of 0.15, with an average convergence time of 51 seconds, significantly reducing the convergence time as compared to the state of the art. Additional simulation results demonstrate scalability to larger crowd sizes. Overall, our results show that our pipeline enables fast, robust, and highly accurate counting of seated crowds.