|
In this issue with you:
Oleg NIKULIN, manager of Pelco (Russia, Belarus)
Alexey KADEISHVILI, Technical Director of the company «Vokord»
Mikhail BYALY, General Director of «Aktiv-SB»
Andrey PIMENOV, Head of the Public Relations Department of the Scientific and Production Center «ELVIS» »
Evgeny MALIKOV, Director of the Sales Department of ISS
Question for discussion
Let's immediately separate the technological wheat from the advertising chaff: what real working functionality can video analytics give to the consumer today?
Oleg NIKULIN:
Real functionality — adaptive motion detector, object counting, direction of movement, abandoned object detector, object removal, sabotage, vibration compensation. In addition, the number recognition module definitely works. I know quite a few examples of successful installations, so we can safely say that this problem is being solved.
The biggest chaff is the face recognition module. In my opinion, this is pure marketing, which has no specific, workable application in real conditions (i.e. without imposing serious restrictions that reduce its effectiveness to zero).
Alexey KADEISHVILI:
Today, video analytics can be used to solve the following tasks quite reliably:
Detection of movement and abandoned objects (this is a classic of the genre).
Identification and tracking of objects of interest (people, cars, etc.).
Basic behavior analysis. The key word here is «basic». This is movement in a prohibited direction, crossing virtual barriers, staying in a prohibited area, etc.
Recognition of objects of interest (recognition of faces, license plates of cars, numbers of carriages, etc.).
In general, abstract terms, video analytics is effective where the event being sought has clear visual features by which this event can be identified, and where events that are not of interest have equally clear features by which they can be filtered.
Mikhail BYALI:
Video analytics can perform several functions. The first is to attract the operator's attention, the second is to organize a search in the archive for events of interest, the third is to work in a completely autonomous mode, independently managing the actions of people or mechanisms. In the first case, a lot depends on the operator's qualifications, therefore, even if analytics does detect most cases of violations in the behavior of monitored objects, an illiterate operator does not see most of them. It is important that the system monitors the operator's reaction, thereby exercising control over him. The video analytics — operator combination works well where a prompt response to the actions of intruders is required. These can be crowded places, such as: railway stations, airports, squares, other places of mass gathering of people, where video surveillance is carried out both as part of the fight against terrorism (abandoned objects, panic, etc.), and as part of the fight against theft, robberies, hooliganism (running people, aggressive movements, etc.). Also, video analytics, combined with security equipment, can provide good protective functions. For example, control over the safety of material assets (museums, galleries, exhibitions, showrooms). Working together with a fire alarm, video analysis can detect or confirm cases of fire at the facility. In the case of recording violations for use in further search or enforcement activities, video analytics is indispensable for identifying traffic violations. Driving into the oncoming traffic lane, sidewalk, making a U-turn in the wrong place, driving through a red light — the system can record and save this and much more in the archive.
|
Andrey PIMENOV:
Currently, security system manufacturers understand different things under the term «video analytics». Some call video analytics motion detection using IP video cameras, others — video surveillance systems with computer vision, which classify targets (people, cars) in real time and recognize dangerous situations, automatically providing information about them to the operator.
Today, computer vision video surveillance systems classify targets, recognize license plates, attempts to enter protected areas, detect abandoned objects, fires, smoke, camera obstructions, thrown objects, etc. Such systems eliminate the influence of the human factor and operate on the principle of «one object — one operator».
Modern video surveillance systems with video analytics also allow you to tune out false alarms, working stably when illuminated by car headlights, insects appearing in the camera's field of view, trees swaying, video camera shaking, precipitation, and cloud shadows.
Currently, developers offer video surveillance systems built on a modular principle. Their basic functionality can optionally be supplemented with intelligent modules. This allows you to create the functionality of the video surveillance system for specific objects and tasks.
Evgeny MALIKOV:
Today, we can confidently talk about systems for recognizing license plates, railroad car and container numbers, recognizing faces, receiving data on the movement of traffic flows, recording facts of traffic violations, monitoring cash transactions, and detecting abandoned objects, which are already in operation at many sites.
Question for discussion
Most experts believe that the need for systems capable of replacing humans in video signal analysis has grown and is recognized by most security practitioners. The demand for automated video signal analysis is huge. How does the market respond to it? What are the new trends in the development of such technologies (implementation of complex analysis, integration with data from other sensors, etc.)?
Oleg NIKULIN:
There is certainly a demand, and it is constantly growing. Especially with the advent of IP cameras and doubly so with the advent of megapixel cameras. Today, the use of analytics on the camera is especially relevant. Why on the camera? 2 Mbit/sec is a standard average flow given by an IP camera. And if the camera is megapixel and you need to get “live video” from it, then flows of more than 10 Mbit/sec are already operated. And what if there are 100 or 1000 cameras at the facility? Also take into account that network equipment cannot operate with 100% load (for video streams, a maximum of 50%). And you will understand that building a functioning network for a modern IP system is not so simple and not so cheap. Therefore, the use of analytics on the camera will significantly reduce the network load, giving out only critical information in maximum resolution. Video analytics not only increases the level of security of the facility, but also reduces the cost of the system.
Alexey KADEISHVILI:
Unfortunately, the enormous demand is largely offset by the fantastic expectations of customers. Therefore, if we leave only those needs that are truly meaningful from the entire enormous demand, and leave only those that can be effectively solved at the current level of development of video analysis tools, it turns out that the rumors about the enormous demand are greatly exaggerated. Nevertheless, there is demand and it is growing from year to year. The market responded to the growing demand as usual: all manufacturers of equipment for video surveillance systems immediately added the word «intelligent» to the names of their products. Only the connecting wires remained non-intelligent. All this has led to a literal flood of systems with so-called souvenir video analytics, which is present in the description for show, but is absolutely useless in practice.
Nevertheless, despite individual excesses in places, video analytics is developing. We can highlight some promising areas of development:
use of stereo cameras to obtain and analyze three-dimensional images;
analysis of human behavior;
highly specialized niche solutions (for example, an automatic drowning detection system for swimming pools).
Mikhail BYALY:
The human brain is a fairly complex device. Different decisions can be made on the same event in each case, and the basis for this will be small details that were previously unknown or seemed insignificant. It is too early to talk about artificial intelligence. Many video analytics functions will only work in tandem with an operator. It is like a dispatcher at an airport: he has all the necessary software and hardware tools to monitor the situation, but only he makes the final decision.
Evgeny MALIKOV:
Video analytics technologies in themselves, no matter how complex and unique they are, are of no practical importance to the end user. What is vehicle number recognition, for example, for a warehouse security officer? But if you present him with a ready-made solution for automatic registration of vehicles entering and leaving the protected area, then the interest in its implementation will undoubtedly be high.
In other words, we need to talk about the variety of user tasks that systems with implemented video analytics functions can solve; the main thing here is to convey to users what opportunities and advantages such systems provide them with. Automation of the processes of accounting and registration of rail and road transport, cargo at enterprises or at transport complex facilities, containers at port transport terminals, implementation of city projects for transport control and security, identification of people — this is an incomplete list of applications in which one of the main components will be a system that provides analytical processing of the video stream.
As demand creates supply, so the user's tasks dictate the vector of development of such technologies. As has been said, their advantages are revealed in the composition of ready-made solutions, therefore, they should be easily integrated with technological equipment, with traditional means of ensuring security (such as, for example, ACS and OPS), with actuators, with information resources used at facilities.
It seems to me that the most correct thing today is to implement complex projects on the platform of the released video analytics systems, within the framework of which various installed equipment, document management systems, databases — all the components necessary to solve the tasks set by the customer — will interact in the necessary way.
For example, by installing a license plate recognition system at an enterprise checkpoint, the customer can receive not only automatic registration of vehicle license plates (with saving of date, time, direction of travel), but also saving of weighing data, commercial inspection, preparation of a unified reporting form, data export to the document management system, notification when alarm events occur, a user-friendly interface, and a customized response of barriers.
Question for discussion
Video analytics and IP technologies for video signal transmission: gains and losses.
Oleg NIKULIN:
Only the advent of IP devices made video analytics a reality, a part of our lives. It is foolish to expect good results from video analysis of a video frame with low resolution or with poor characteristics. Today, video analytics without IP is no longer conceivable. The old categories of TV lines look like an anachronism. Only working with each pixel will allow us to obtain an adequate result.
And another important point: video analytics should be installed on the video signal source (on an IP camera or on an encoder). Any compression worsens the quality of the video image. And only pixel-by-pixel work with the frame before its compression-decompression will allow using the capabilities of analytics to the maximum.
Alexey KADEISHVILI:
Since the computing resources of video cameras are still limited, it is not yet possible to implement serious analytics in the video cameras themselves, and separate powerful computing servers must be used for video analysis. Moreover, this limitation is quite fundamental in nature with today's technological standards (~50 nm); approximately 30-100 W per camera is required to implement reliably working video analysis algorithms. You can't squeeze such an iron into the camera body — it will overheat. According to Moore's law, the same computing power will squeeze into the coveted 10 W in about 3-5 years. From this point of view, nothing has changed with the transition to IP technologies.
The main gain is the convenience of building the infrastructure of video analysis systems. You can use any suitable computing platforms and place the equipment in any convenient place, since the distance to the camera is no longer a limiting factor.
However, one has to put up with the fact that the compressed image coming from IP cameras contains compression artifacts that impair the reliability of video analysis algorithms (recognition algorithms are especially sensitive to this) and, in addition, part of the computing resources of video analysis servers has to be spent on unpacking compressed video data. The special irony of the situation is that camera manufacturers are trying their best to increase the resolution of video cameras (which is an absolute benefit for video analytics), but at the same time they are trying with no less zeal to compress the video stream, using algorithms with interframe compression such as H.264/MPEG4, which, from the point of view of video analytics, kills all the advantages obtained from high resolution.
Mikhail BYALI:
Obviously, for high-quality work, video analytics requires a signal that is not compressed. In this case, using on-board analysis of the camera may be preferable. However, there is a nuance regarding the power of the processor that will perform this analytics — after all, simple processing does not require as much power as complex calculations. And if each camera is equipped with such stuffing, its price can increase several times. IP cameras also have undeniable advantages in the form of progressive scanning and megapixel cameras. The first allows you to receive non-deinterlaced frames, the second produces a picture with a high digital resolution, significantly exceeding analog. However, you will most likely have to forget about the MPEG4 or H.264 compression formats — the entire image will go in the «heavy» MJPEG.
Andrey PIMENOV:
In CCTV, there is a transition from analog video surveillance to IP systems. Open standards are used in IP video surveillance equipment, which simplifies the installation and upgrade of equipment, and it becomes possible to use products from different manufacturers in one video surveillance system. One of the main advantages of IP video cameras is the presence of built-in motion detectors. A motion detector is a module whose main task is to detect the movement of objects in the camera's field of view. Motion detection is a simple form of video analytics, which is usually used indoors.
Evgeny MALIKOV:
The development of IP video surveillance is strategically important for many software development companies today: every day, more and more integrator companies use IP equipment when implementing projects, and, consequently, as was said earlier, the customer sets a wide variety of tasks that can only be solved using special software.
So there is no talk of losses. Now, when installing modern IP equipment at a facility, you can freely use video analytics systems. Their combined use provides a qualitatively new level of video surveillance organization, while the «native» software of IP cameras is often insufficient to solve the problems of intelligent processing of the video stream.
Question for discussion
What, in your opinion, is the practical effectiveness of video analytics?
Oleg NIKULIN:
Public places, transport arteries, special objects – today there can be no serious video surveillance system without video analytics. Even parking lot security, control of entry to the shopping mall parking lot will become much more effective with the help of electronic brains.
The simplest analytical modules – such as “vibration compensation”, which will help compensate for wind vibrations of the mast on which the camera is installed, such as “sabotage”, which will issue an alarm when the camera lens is covered or painted over, or when the box camera is deliberately turned – make life much easier for the security service.
Analytics has already entered our lives following IP.
Alexey KADEISHVILI:
Well-made, correctly installed and configured video analytics is able to quite effectively solve the problems for which it was made. But this statement is true subject to the following conditions.
Firstly, it is necessary to correctly formulate the tasks that we want to solve using video analytics. If the task is to «identify suspicious human behavior», then 99% of the time such a task is not solvable, since «suspicious behavior» is too broad a concept. If the task is to record the faces of all people entering a protected building, then video analytics solves such a task by 99%.
Secondly, it is necessary to realize that effective video analytics requires considerable investment. It is necessary to ensure a high degree of coverage of the object by cameras (as a rule, the number of cameras should be greater than in the case of overview video surveillance), the cameras should provide high-quality images (which means these are not budget cameras), it is necessary to ensure sufficient computing power of the video analytics servers (which means there should be more servers than for overview video surveillance).
Thirdly, it is necessary to keep in mind that the implementation of complex video analysis, as a rule, requires quite painstaking work on setting up the system, and this is due not only to the imperfection of current implementations of video analytics, but to the very formulation of the problem.
|
Mikhail BYALI:
In addition to the above, I would also note the joint use of video analytics not only by security services, but also by marketing, logistics, and production departments of the company. For example, by counting people, shopping centers can provide accurate data on traffic. Merchandisers get the opportunity to identify hot spots at points of sale during the most intense visiting hours from the office. Logisticians monitor the movement of goods in real time by reading train car numbers. In production, video analysis can be used to control manufactured products and regulate the implementation of technological processes. But, probably, out of 100 users of video systems, in addition to primitive motion detectors, video analytics is used by a maximum of five.
Andrey PIMENOV:
In video surveillance systems without video analytics, a person is an observer. It is he who analyzes the video stream. For high efficiency of video surveillance in this case, it is necessary for one operator to monitor the video information received from one camera. Otherwise, the video surveillance system is limited only to the properties of archiving huge flows of information without the ability to process in real time.
Video analytics has made it possible to assign video stream analysis to the video surveillance system, and humans have begun to make decisions.
A question for discussion
According to some experts, one of the most serious problems is the shortage of highly qualified video surveillance operators who constantly maintain their level. Should the players in the TSB market participate in solving it?
Oleg NIKULIN:
Video analytics is designed to reduce dependence on the operator's qualifications and condition. Proper system design will allow for automatic connections and system response without taking into account the operator's adequacy.
Alexey KADEISHVILI:
I think that the problem here is not specific to video surveillance, but is common to all areas of human activity and is related to the fact that there are few people in nature who are capable of doing their job conscientiously and responsibly. As for video surveillance, with normally functioning equipment, any person who has independently graduated from high school is able to learn how to switch cameras and control them in one day, the only requirement from manufacturers is that the equipment works.
Mikhail BYALI:
We do not have such a profession as a video surveillance system operator. How to teach them, what to teach them – we do not have this in our country either. This work is not considered qualified and is financed on a residual basis. Until these issues are resolved, no participation of the TSB market players will help. But this participation is not needed. The task of video surveillance system suppliers is to ensure that the customer's requirements for the security system are met, and to provide all the necessary tools for this. Formulating these requirements is the prerogative of the heads of security services. And training video surveillance system operators in their work is their responsibility.
Andrey PIMENOV:
The use of video surveillance systems with video analytics allows us to solve the problem of the shortage of highly qualified video surveillance operators. The main problem with existing video surveillance systems (analog, digital, IP) is the need to connect an operator to almost every camera. The task of video analytics systems is to free the video surveillance operator from analyzing the video stream, offering him only the information that is necessary for decision-making. A person is a bad observer, but he makes decisions perfectly.
The ability of video surveillance systems with video analytics to recognize targets and situations in real time will not only reduce the operator's workload, but also reduce the number of operators, save funds allocated during the life cycle and operation of the system. Thanks to video surveillance systems with video analytics, even the largest protected area or facility equipped with hundreds of video cameras does not require a large staff of security personnel. In addition, video surveillance systems with video analytics record all operator actions, which eliminates the influence of the human factor, collusion between operators and intruders.
Evgeny MALIKOV:
In our opinion, improving the skills of video surveillance operators is not the task of developer companies. Companies offer products that significantly facilitate their daily activities, allow them to solve many specific problems as efficiently as possible thanks to automatic image processing, a user-friendly interface, and pre-configured work and response scenarios.
|
