Stand Up (or Sit Down) and Be Counted

There are times when it is important to know how many people are in a given area, but physically counting may be either inconvenient or impossible. Some scenarios where it is necessary to count crowd size are in energy management, monitoring crowd size during a pandemic, business planning, and security.

The lab of Yasamin Mostofi at the University of California Santa Barbara has been working on an interesting solution to this problem. Their solution is able to do the job wirelessly, without any action on the part of the people being counted. It will even work through walls.

To estimate the size of a crowd, an off-the-shelf Wi-Fi transmitter and receiver are positioned nearby. Signals sent by the transmitter interact with the environment and are then captured by the receiver. When people are in motion, they interact with the Wi-Fi signals on their way to the receiver in a way that can be measured. However, when people are sitting still, there is no motion for the system to detect… or is there?

Mostofi’s insight in this work is that, even when stationary, people in the crowd will periodically fidget a bit, whether that means adjusting their seating position, scratching, checking their phones, or otherwise. They mathematically described these periods of fidgeting, and also the periods of silence. As you might expect, when a crowd size is larger, the periods of fidgeting are longer, and periods of silence are shorter. With their mathematical model, the lab was able to leverage this information to estimate crowd size with high precision.

The team tested their method in a series of 47 experiments, which were conducted in a variety of locations, and with different seating configurations of those being counted. Overall, the algorithm produced the correct result 96.3% of the time when the equipment was in the same room, and 90% of the time when operating through walls. An error rate of zero or one person was observed in 44 out of 47 trials.

These results are highly impressive, however, it should be noted that the system was tested with a maximum of ten persons in each trial. It is not clear how well the method would perform with much larger groups — presumably with a large enough crowd, fidget time would reach 100%. That may prevent this method from providing an accurate count estimate. Without results from trials that involve larger groups, it is not clear how useful this technique will be in real-world applications.