Senior Marketing Manager

Runar Finanger's profile picture

XeThru sensors count inbound and outbound commuters at Gangnam Station

PSY's 2012 hit song 'Gangnam Style' surely put Seoul’s exclusive neighborhood "Gangnam" on the map and instantly transformed it into a global tourist attraction. The Gangnam neighborhood is a newly constructed upper class part of Seoul and being the HQ of Samsung and many other corporations, it has the busiest subway station in the country.

Seoul Metro transports on average 5 million people a day, and as part of a development strategy to provide subway users more comfortable services, their plan was to measure the actual number of inbound and outbound commuters at the Gangnam subway station without invoking any privacy issues.

A dedicated team from the South Korean Hanyang University, led by Professor Sung Ho Cho, was assigned the project back in 2014 and has been working with the XeThru X2 radar SoC.

XeThru Radars at Gangnam Station

In order to cover the whole pathway, the team installed a radar box (50x20x4.5cm) at multiple neighboring pillars. Each radar box consisted of two embedded XeThru X2 radar modules and one Raspberry Pi module. Each module was set to approximately cover a 5~6m range.

The challenge here was clearly to maintain the 90% accuracy requirement in identifying the direction (left or right) of massive human traffic, as well as counting the number of people coming from each direction. Despite these stringent requirements, the modules succeeded in meeting the level of accuracy with values varying from 90~99%.

Another challenge was to balance the radar algorithm to correctly count during peak and off-peak hours. As confirmed by Professor Sung - there will be a tendency to have ghost images when few people pass by the sensors and thus too high numbers. While in a busy environment, non-line-of-sight phenomena will cause observed numbers to be on the low side.

The current results of the project are impressive but in order to further improve them, the Hanyang team is now in the process of developing a machine learning algorithm to make the measurements even more stable and accurate.