Camera installation: how neural networks will search for children and the homeless
In the Novosibirsk region, they announced the introduction of a service for searching for missing children using artificial intelligence — even before contacting the police, it will be possible to upload a photo of the missing person to the Safe City system, which will allow them not to waste valuable time and immediately start searching. And in Kazan, they decided to use city surveillance cameras to search for people who find themselves in difficult life situations, including homeless people. Artificial intelligence will analyze uncharacteristic human behavior on the street. Neural networks are already taking over the functions of finding a person in trouble, but these systems are not well developed everywhere yet. For more information, see the Izvestia article.
How AI is used in Novosibirsk and Kazan
The government of the Novosibirsk Region said that the region was the first in Russia to introduce a service for searching for missing children using artificial intelligence. The program, developed by the regional Ministry of Finance, has already been tested by operators of 112 and 102 services and employees of the Ministry of Internal Affairs of the Novosibirsk region.
It works like this: when contacting 112, the operator offers to immediately upload a photo of the missing child. After that, the applicant receives a unique information exchange card number on his phone and an SMS with a link. The photo enters the secure circuit of the Safe City hardware and software complex. After that, it is necessary to register an application with the police department so that employees can quickly begin the search, including using biometric analytics tools.
Dmitry Vtorov, head of the association "Search for Missing Children", told Izvestia that the idea of the Novosibirsk region is unique precisely in terms of uploading photos to the Safe City system even before submitting a police report.
"The advantage is that the information will immediately flow through the system, but another question is whether there will be abuse of its functionality by applicants," he said. — On the other hand, it certainly increases the responsiveness to missing persons. We understand that one way or another, a person will need to come to the police and file a report, but this can shorten the search time, especially in emergency cases. Then such a system will be extremely effective.
And in Tatarstan, they decided to use city surveillance cameras and artificial intelligence to search for the homeless and people in difficult situations. This is part of the Uram project from the Center for Digital Transformation of the Republic of Tatarstan, which includes the Human Shelter charity project. Neural networks will teach scenarios for the behavior of homeless people and people in difficult situations: AI will analyze the image from surveillance cameras. The Uram project has been in operation for several years. With the help of AI, various events are analyzed: falling people, recognizing "pawnbrokers", fights, stray dogs, etc.
Alexander Sukhanov, director of Human Shelter, explained that the organization had proposed to the Ministry of Finance of Tatarstan to adapt artificial intelligence, which is already able to track stray dogs, to detect strays on the streets of the city.
"And we have developed such an algorithm: information from the camera will be processed and if the behavior of a particular person falls under certain parameters, then the data about this will immediately be sent to us, and a social patrol will go to the place," he told Izvestia. — We are also discussing how information, for example, that a person is unwell, should be immediately duplicated to the emergency services. The program should be launched before June of this year.
Sergey Ievkov, Executive Director of ANO Charity Hospital (St. Petersburg), notes that the idea of finding homeless people on the streets in this way is quite working.
— If a real system of assistance to people in need is built behind the technologies of the "Safe City", when a homeless person is taken to the street and offered help, routed to a shelter, shower, hospital, and not forcibly taken away, for example, to a police station or work homes, then it will work, — he said. Izvestia. — In order to reach large cities with millions of people, such as Moscow, St. Petersburg or Kazan, the participation of NGOs alone is not enough - no non-profit organization has so many resources: applications must be submitted to the state center for assistance to homeless people.
By the way, the Charity Hospital itself still has the Telegram bot ISeeHomelessBot, with which residents of St. Petersburg can transmit information about homeless people.
Is AI helping you find people now?
Artificial intelligence has been used in city surveillance camera systems for a long time, Dmitry Vtorov notes. In Moscow, the technology is used both on the streets and in public transport, and it automatically notifies services when they recognize wanted persons. Such systems greatly simplify the work of the police, the expert said.
Tatiana Izutina, Director of the Reksoft Public Sector business area, calls the development of AI technologies in Safe City systems quite mature. According to her, in 2025, the Ministry of Finance of the Russian Federation has planned to create a service for processing video streams from surveillance cameras from Russian regions using artificial intelligence, which will create a single data storage center.
In March, the head of the ministry, Maksut Shadaev, reported that in Russia, every third video surveillance camera out of more than 1 million installed in the framework of the "Safe City" is connected to a facial recognition system.
Mikhail Bocharov, Deputy General Director for Scientific Work at SiSoft Development, notes that AI can be configured to search for and identify a large number of situations. Moreover, neural networks can analyze the degree of importance of the event.
— For example, artificial intelligence saw through the camera that a person had fallen in the surveillance area. If he got up and walked on, then on the video it can be marked as an incident and nothing more," he told Izvestia. — But circumstances can be determined that will help identify the incident as more dangerous. For example, the fall of several people at a subzero temperature — then it can be concluded about an area with increased ice, or the fallen person cannot get up on his own, or another person ran away when he fell, etc.
What is the situation with cameras in the regions?
Ilya Fomichev, head of the AI department of the IT company SimbirSoft, notes that over 270 thousand cameras in Moscow are capable of recognizing faces, in which algorithms for detecting uncharacteristic behavior are implemented. The capital remains the leader in terms of coverage and technology. Following Moscow, other regions are also starting to introduce and apply modern solutions.
"For example, in the Samara Region, there is a single AI—based video analytics service platform that unites more than 5,000 devices, from video surveillance cameras to smart intercoms and emergency communication consoles," the source told Izvestia. — In the Yamalo-Nenets Autonomous District, AI algorithms analyze data from cameras in courtyards and entrances, recognize faces and silhouettes, help search for suspects and missing people, and are able to reconstruct the picture of incidents.
Similar projects are being developed for implementation in the Ryazan, Yaroslavl, Tyumen, Nizhny Novgorod and Arkhangelsk regions.
The leading CV engineer of Softline Digital (Softline Group) Vladimir Valeev noted that the Moscow facial recognition system has already been tested in 10 major Russian cities, but there are limitations: there are significantly fewer surveillance cameras in other megacities, including Novosibirsk and Kazan. And if technically facial recognition is possible there, then tracking a person's movement around the city is already more difficult.
Alexander Didenko, head of the Artificial Intelligence Laboratory at the Skolkovo School of Management, notes that Russia ranks third in the world after China and the United States in terms of the number of cameras, but the coverage is very uneven. At the end of last year, this was confirmed by the GS Group analytical center, which conducted a study on the number of urban video surveillance systems and compiled a rating of 85 regions (available to Izvestia). In terms of the number of cameras per 1,000 inhabitants, the leaders were Moscow (22.13 cameras per thousand people), St. Petersburg (18.76), Moscow Region (15.14), Lipetsk Region (12.09) and the Republic of Tatarstan (11.99). The Republic of Dagestan (0.49), the Irkutsk Region (0.43), the Jewish Autonomous Region (0.24), the Kamchatka Territory (0.23) and the Trans-Baikal Territory (0.10) were among the five outsiders. Moreover, there are more than 20 regions in which the specified value is less than one.
Ekaterina Gerling, a leading analytical engineer at the Laboratory for strategic development of cybersecurity products at the Gazinformservice Cybersecurity Analytical Center, notes that Moscow and St. Petersburg are covered with cameras by 80-90%, while large megacities such as Kazan, Yekaterinburg, Sochi and Novosibirsk are less than 60-70% covered. In medium-sized cities such as Tula, Ivanovo, etc., coverage is 30-40%.
—And in small towns, outdated camera models and even analog ones are often used, which are categorically unsuitable for neural network analysis," AI expert Alexey Onosov told Izvestia. — For a detailed analysis of facial expressions, gait features, or other small details, it is advisable to have cameras with 4K resolution. Another problem is night shooting, not all urban cameras are equipped with high—quality infrared illumination, which means that the effectiveness of AI systems is noticeably reduced at night.
Alexander Didenko emphasizes that to monitor the main traffic patterns — and the search for homeless people can be attributed precisely to such tasks — high resolutions may not be required: modern neural networks can work with low and even ultra-low quality images.
"But for tasks like searching for a missing child using photographs, the requirements for systems are much higher," he emphasizes. — In addition to the quality of the camera and the type of task, it is important that the settings give a picture similar in characteristics to those used to train the model.
LUIS+ Project Manager Sergey Dyachenko notes that high resolution is not always crucial. The main thing is to install the cameras correctly and set up the shooting scene, which allows analytics to recognize certain events.
Alexey Onosov emphasizes that another problem is data transmission channels, server capacities for processing the video stream, and the bandwidth of data transmission networks. He believes that for the full implementation of AI in Safe City systems throughout Russia, a large-scale phased modernization of the video surveillance infrastructure will be required.
Переведено сервисом «Яндекс Переводчик»
