Line B and line C are two major lines of the Paris Regional Express Railway (RER). Line B, which carries an average of 165 million passengers each year, joins the northeast to the southwest via the center. It is the second busiest line on the network with its 80 kilometers and 47 stations. Line C is more than 100 kilometers long and has twice as many stops. It joins the northwest to the southwest through the center. In total, these two lines represent more than 260 km of track, 131 stations and transport more than 300 million passengers every year.
In 2018, the SNCF launched a tender for lines B and C of the Paris RER. They want to know in real time, 24 hours a day, 7 days a week, the number of people present on its station platforms and in its stations. The goal is to have a global overview of network occupan-cy rate. This allows the SNCF to charter additional trains or to detect the over-load of a platform for example. For this, the SNCF already has a large came-ra network that has a view over the platforms. More than 750 cameras are available for carrying out the project. The mission is therefore to offer an ef-fective solution that stands out from the competition and at a competitive price.
Among its products, ACIC has a solution to count people in a zone. This solution, called CrowdNum-bering, has been proposed to SNCF. In concrete terms, this product makes it possible to determine a virtual zone and to instantly count the number of people present in this zone. The counting frequency can be customized (every minute, every 10 seconds, …). With the right settings, the error rate is rela-tively low and can be below 10%. The video analytics which allows the counting is based on artificial intelligence models (deep neural networks). They have been trained to detect the appearance of a person or a head, regardless of their pose or orientation. This product runs on an analytics server and requires a GPU.
In competition with other companies including IBM, ACIC and its CrowdNumbering solution were chosen by SNCF. We therefore equipped more than 786 cameras in the 57 stations of the two lines with CrowdNumbering® licenses. Currently, the 786 video streams are processed by four analytics servers, each equipped with a GPU. In addition to CrowdNumbering®, each camera has been equipped with an option called «Full Anomaly Detection». This option allows the operator to be notified in case of camera displacement, blur, camera occlusion, loss of signal and under or over exposure.
The data collected is processed internally at SNCF by the TCO Voyageur platform.