Monitoring crowd condition in public spaces by tracking mobile consumer devices with wifi interface

Abstract

We present a systematic study and optimization of crowd monitoring methods based on tracking consumer devices with activated WiFi/Bluetooth interfaces using stationary scanners with directional antennas. To this end we have recorded a large scale, real life data set from a car manufacturers exhibition at the Frankfurt Motor Show IAA that includes data from 31 directional scanners covering a total area of 6000m2 running for 13 business days and providing nearly 90 million data points from a total of over 300000 unique mobile devices. For seven of the 13 days video ground truth has been recorded and extensively annotated. Questions that we addressed include the mapping from the number of detected devices to the number of people, the ability to generalize the calibration from a small number of ground truth points recorded on one day to other days and the ability to localize individuals in different conditions. Our methods show less than 20% error for the crowd density and less than 8 m localization error for individuals.