Dmitry Golovin: “Moscow on top of innovative technologies of the development of video surveillance and video analytics solutions”

Dmitry Golovin: “Moscow on top of innovative technologies of the development of video surveillance and video analytics solutions”

A single operator can service a thousand courtyards, while the whole city is controlled by 30 operators”

By the end of 2020, Russia became the world's third-largest country by the number of video surveillance cameras, according to TelecomDaily. With its 13.5 million video surveillance cameras it is only behind China (200 million) and the US (50 million). Moscow's video surveillance system, according to last year's data, has a total of about 200,000 cameras connected to it. In 2020, it was used together with intelligent analytics algorithms to track compliance with social distancing and quarantine violations. Last September, the Moscow Metro started actively using video analytics, and in six months of use, it helped apprehend around 900 criminals and find 100 missing people.

Dmitry Golovin, head of Moscow City Video Surveillance Department, discussed with ICT.Moscow what challenges the capital's video surveillance system will address in 2021 and in what direction it is developing.

How is the video surveillance system in Moscow organized? How many cameras are used, which objects are connected to them?

Today, there are around 204,000 cameras in Moscow. The main projects to create a global video surveillance system were completed quite a long time ago, and new cameras are now installed as part of targeted projects - industrial or specialized.

CCTV cameras in Moscow began to be systematically installed in courtyards and entrances since 2001. Since 2015, video analytics algorithms have been developed as part of the Moscow video surveillance system. The pilot project was launched in 2017, then it included 1150 cameras in Moscow courtyards and entrances. In 2018, the system was expanded to 1,500 cameras, it was tested during the FIFA World Cup. At the same time, the cameras helped to apprehend about a hundred offenders in the criminal investigation database.

In its entirety, the system is divided into two parts. The first includes city-wide areas: courtyards, public places - that is, places that are unlikely to be sufficiently equipped with video surveillance systems by anyone else. This category also includes social facilities: schools, clinics.

The second refers to integrated systems that include video surveillance complexes of various public facilities: Moscow roads, public transport stops, and commercial, or private, facilities. For example, most shopping centers are integrated into a single video surveillance system, stadiums and public urban spaces.

In that case, what is the advantage for companies of connecting to a city video surveillance system compared to setting up their own?

First, they can access their cameras through the city system. Imagine: there are companies that have several outlets around Moscow, and they specialize in retail. They don't have to be experts in setting up large CCTV systems. They can simply put cameras, feed them into our system, and we will provide them with the capabilities of our entire powerful video surveillance core, the capabilities of a platform system: access to all their cameras from a single point.

Secondly, if something happens on their property, the police officers have no problem getting the data from the video surveillance system: the officers know exactly how to use the city system, it is all regulated within the law enforcement agencies. All the officers who have to work with video surveillance know how to do it and have the necessary access.

Thus, on the one hand, this tool helps the city to solve its problems, and, on the other hand, allows commercial organizations to increase security, convenience and, to some extent, even save money. In other words, Moscow video surveillance is a multifunctional 'Swiss knife', allowing a large number of tasks in various sectors to be solved simultaneously.

How can companies connect to a city video surveillance system? What criteria should the cameras and communication infrastructure at the company meet? How long does it take to connect?

The simplified process of connecting organizations’ cameras to our system looks as follows:

First, the organization submits an application to the Department of Information Technologies and attaches information about its surveillance systems. We will then investigate the feasibility of integrating the proposed cameras. If the decision is positive, we will send a recommendation to the organization for integration (list of work to be done in order to connect).

After that, the connection itself takes place and access is granted to authorized employees of the organization. For convenience, all technical requirements for video surveillance equipment and communication channels are in Appendix 3 of our, which can be found on the DIT website.

One of the global trends in video surveillance today is automation and the use of intelligent video analytics. What real-world challenges are being addressed by these processes?

In most of the processes where CCTV is involved, automation has enabled us to do away with the physical presence of inspectors. First of all, we are developing the most popular areas for the city: controlling snow removal, rubbish collection, and the condition of playgrounds. These are the most widespread processes that are clear and easily structured, which allows us to automate them effectively, including with the help of video analytics.

In the future, we plan to implement them in areas where the tasks are more complex and not so massive.

In Moscow's video surveillance system, is automation achieved primarily through software robotics (RPA) or are machine learning algorithms used?

Currently, it is much more efficient to solve some tasks using artificial intelligence (AI), because you can get an end result or enough data to make a final decision. But in our practice, there have also been cases of serious employee efficiency improvement through simple automation, robotization. An example is the inspection of the courtyard area's condition, which, until 2011, was performed manually by inspectors.

We have analyzed their work and worked out a conditional algorithm. For example, at 9 a.m. you have to use the cameras to see if the area has been cleaned, at 2 p.m. the cameras assess the condition of the playgrounds, and in the evening the removal of rubbish from the container sites. We have made a schedule according to which at a certain time the cameras turn to a predetermined point, take a screenshot, and in this way an archive is collected.

This approach has increased efficiency several times over: a single operator can service a thousand courtyards, while the whole of Moscow is controlled by 30 operators.

What about video analytics? For which applications is it used today, what are its special features?

We start with the principle "the onion should be worth peeling". It must be clear which processes should lead to which results. Because machine learning, which is used in video analytics systems, is still quite a structured task, with clear evaluation parameters, criteria, and so on.

We need quite a lot of training materials to train neural networks. And a lot of material is produced in large-scale projects. For example, know that everyday operators visually monitor the litter removed from container sites and the litter that was not removed. And there are many more examples of taken out and not taken out rubbish for training a neural network than for other tasks, where some are specific or quite exceptional.

Therefore, the first criterion for the application of video analytics systems is the availability of training material for the neural network. The second is the economic effect of implementing this automation process. It is clear that the housing and utilities sector is very large and concerns all Moscow residents without exception, and a huge amount of money is spent on this sector. Therefore, a greater economic effect can be achieved there through automation, and we are now prioritizing this area of application.

Of course, we conduct pilot projects in a number of areas that are less large-scale, but no less important for the city. For example, control of illegal construction. You have to determine that there is a structure in someplace that has not been approved, and it is a slightly more specific task because it is difficult to find clear training examples which demonstrate where there is an unauthorized building and where there is not. That's why it's more difficult to train the neural network for such a task.

Do you use only internal datasets or also employ third-party datasets to train machine learning algorithms?

Solving urban problems and similarly, narrowly focused tasks are simply not of interest to the market in and of themselves. There are not thousands of commercial companies, foundations or associations that create datasets for rubbish collection control. It is an urban or national task, which means companies can't make a product which they would be able to sell to all users.

So, if we are talking about some basic, comprehensive scenarios, then, of course, neural network developers use open datasets. There are more and more of these datasets being created, even the number of types of these datasets is growing, which is good. And for some tasks, there is no data at all.

Big Data collected through video analytics is now the main trend in the development of Moscow's video surveillance system. Such Big Data can be used quite widely, starting from simply collecting statistical data for managing and changing urban infrastructure, monitoring the sufficiency of social facilities.

How is the data shortage being addressed today?

Information from CCTV cameras is a huge source of data for training artificial intelligence in video analytics. The main challenge picking the necessary data for training specific neural networks. Depending on the priority of the tasks to be solved, the data can be tagged using internal resources, resources from partners or, if necessary, through government contracts.

Are the video analytics algorithms created in-house or outsourced?

Developing in-house would not be quite appropriate, because these are technologies that are constantly evolving in different directions, it is much more efficient to use market solutions than to develop your own narrowly focused ones that do not compete with anyone else.

We engage several developers, look broadly at the market, and take those products that are the most suitable. Either they are immediately ready to realize the tasks facing the city or the system, or they are finalized by those who find it easiest.

How does this multivendor approach work in practice? Are developments from different companies used for different tasks, or are developments from several companies combined in one system?

It depends on the task. If it is linear and straightforward, it will be one algorithm. If it is a more complex task, different algorithms can be combined there. Even the face recognition system has a separate algorithm for detecting faces from a video stream and three separate algorithms for recognizing faces. And these are products of different vendors, which solve the problem together. Moreover, there is a separate solution that links them together and creates a face recognition system itself. The best approach is worked out on a case-by-case basis, based on the solutions that are available on the market and meet our requirements.

Are other technologies, such as 5G or blockchain, planned as part of the development of the capital's video surveillance system?

5G is certainly of interest to us because video surveillance, especially centralized surveillance, is directly linked to data transmission. For this, communication channels are very important. And wireless communication can significantly reduce the cost of developing video surveillance systems and transmitting this data because providing a physical connection for cameras is a hard process. This is especially true as we use mobile surveillance systems in addition to stationary cameras.

We are studying blockchain technology and periodically come back to this issue, but so far we have not been able to find any concrete security solutions in terms of video surveillance systems. Therefore, there are no working solutions yet. We are looking into it but without fanaticism.

In general, we are constantly staying on top of innovative technologies, analyzing the possibilities of their application to improve the effectiveness of the development of video surveillance and video analytics solutions.

 

We use our own and third-party cookies to enable and improve your browsing experience on our website. If you go on surfing, we will consider you accepting its use.