For today's megalopolises, subway systems are a sort of bloodstream, keeping the city alive and active with the proper movement of resources, a vital network that shuttles riders to and fro. When the subway stops working, so does the city. A broken-down subway line means that employees will be late to work and students late for school. The remainder of the transport network is overloaded and city-dwellers become more stressed, which in turn reduces their trust in the city's administration and their overall satisfaction with city infrastructure. So when talking about a city's underground transport arteries, we must pay close care to a number of factors. The biggest of these consists of disorderly conduct, misbehavior, and – as experience sadly shows – terrorism. This threat makes it incumbent on subway operators to install innovative security systems, which reduce the human factor in threat management and promote rider safety. Innovative security is a multifaceted issue that requires cooperation between multiple city agencies. Security systems must be effective and reliable, even as they ensure broad coverage for the needs of diverse customer agencies within the city government.
In early 2013, the management of the St. Petersburg subway system decided to launch a pilot project for rolling out a smart video surveillance system with facial recognition. Ladozhskaya station was chosen to host the trial installation. If successful, the system would be expanded to another 19 stations.
The Axxon Intellect PSIM was chosen as the software to tie the system together.
Axxon Intellect had already been deployed with great success in subway systems in London, Rome, Ekaterinburg, and Novosibirsk, giving AxxonSoft's product a substantial advantage in the procurement deliberations.
For the purposes of the St. Petersburg project, Axxon Intellect was to be integrated with the Concorde security system, which had been developed specially for the St. Petersburg subway. The system installer was TekhnoOkhranServis LLC.
The main goal of the project was to ensure that the photo of every person entering the station is recorded, with information about the passenger's further movements to be made available to law enforcement.
The first task was to install facial recognition, which requires that image quality be sufficient for identification and comparison with an image database. The conditions for recognition are far from optimal: low lighting, backlighting during the daytime, and enormous passenger traffic, just to name a few.
The next logical step is to compare the image with those in a database. And of course, the parameters to be compared need to tweaked to ensure accurate matching.
It is necessary to account for a number of details, in light of the complexity and importance of the task: the height and angle of the cameras (finding the tradeoff between the mounting height and angle of incidence), camera downtilt, indoor lighting, outside lighting, and vibration caused by escalator movement. By solving this problem, the developers and installers effectively cracked the whole nut, as it were.
So in the entry area at Ladozhskaya station, cameras for passenger monitoring were installed at a height of 170 centimeters (67 inches). Neither a snug scarf or hat brim will save passengers from identification. According to St. Petersburg subway executives, daily passenger flow at Ladozhskaya is approximately 65,000 people – an amount that the new video surveillance system is handling just fine.
The cameras are located out in the open, but hiding from them is not possible: images are captured, and recognition is started, by multiple cameras as soon as a passenger enters the station.
The operator workstation is located at Ladozhskaya station as well, in a special monitoring area. The surveillance system supports multiple levels of notifications, different types of operator notifications, and configurable methods for displaying information on the operator's screen. Fewer than four seconds are needed to notify the operator when the face of a criminal is recognized in the database.
There is practically no limit on the capacity of the photo database used for comparison. The current storage limit for recorded video is one month, but actual retention length is limited only by the expenses necessary for data storage. Up to one month of subway trips can be "played back" for any person whose photo is found in the database.
According to Fontanka.ru, the system currently recognizes 98% of faces, and if the recognized photo is in the database, there is a 92% chance that the face will be identified. These are fantastic results that give cause for optimism.
The next step is to install similar systems at another 19 subway stations in St. Petersburg: currently, when a fugitive is identified by the system, it is impossible to track further movements since cameras are installed at only one station. Thus the most impressive results of surveillance of the St. Petersburg subway system are yet to come, after expansion is completed – something we hope will be soon in coming!