Our proposed measurement methodology does not make use of cameras, Wi-Fi or Bluetooth and operates in all-weather conditions in which the influence of the physical presence of human individuals or animals on radio frequency (RF) signals in the environment is used to derive crowd size information.
We transmit radio waves in different directions through the crowd and use this in an advanced analytical model to accurately measure the number of people in a certain area. The hardware part consists of wireless nodes that will be deployed in a specific environment at approximately waist height. Additionally, these nodes are connected to a gateway. Each node is battery powered and can transmit and receive sub 1-GHz wireless signals. Because the data is processed in real-time, crowd estimations are updated every 10 seconds in a graphical dashboard which can be integrated with other systems from customers or partners.
It is possible to divide the full environments into subregions and perform crowd estimates within each subregion. This enables the detection – and in a later stage, prediction – of crowd flows, which is highly useful information in the context of crowd management solutions. We deployed 21 wireless sensors in the Langemunt and divided the environment into 3 different regions.
To evaluate our technology, a dataset is created with manually counted crowd sizes from each environment to train our real-time algorithms and to evaluate new estimations. In order to continuously keep our estimations accurate, the real-time algorithm will be retrained periodically.
To indicate the accuracy of our system, a CDF error graph is shown. This represents the probability that the error of each evaluated estimation is equal to or less than the values shown on the x-axis. In 80% of cases our estimate is off by at most 5 individuals.