WiFlowCount: Device-Free People Flow Counting by Exploiting Doppler Effect in Commodity WiFi

Abstract

People flow counting is to count the number of people passing through a passage or a gate. Conventional vision-based approaches require line-of-sight (LoS) and impose privacy concerns, while most radio-based approaches require dedicated equipment and incur high cost. In this article, we propose to exploit commodity WiFi to count the number of people of continuous flows in a device-free way, requiring one pair of WiFi transmitter and receiver. Leveraging the Doppler effect induced by human passing, the proposed method, named WiFlowCount, first constructs the spectrogram of Doppler shifts from channel state information. Based on the spectrogram, WiFlowCount detects the people flows and the subflows according to the power distribution in the spectrogram. An optimal rotation and segmentation algorithm is proposed to segment the spectrogram of a continuous flow into the subspectrograms of its subflows. The number of people in each subflow is estimated from its subspectrogram via convolutional neural networks, which add up to the total people count in the continuous flow. Experimental evaluations demonstrate the effectiveness of the method WiFlowCount, able to detect people flows and subflows accurately and count the number of people in continuous flows with high accuracy, outperforming the existing work.