For this edition the author makes the first approach to Video Analytics, or software for video content analysis, which is available in a variety of forms.
Video analytics, or software for analyzing video content, is available in a variety of forms. Video Motion Detection (VMD), facial recognition software, automated license plate recognition software, and vision-based inspection programs are all forms of video analysis. To this group also belongs the intelligent video analysis, or intelligent video.
To put it in simple terms, intelligent video analysis is a software that extracts useful information from video images, turning video into data. The technology behind video analytics, called machine vision, dates back to artificial intelligence research in the '60s.
Only after several decades of research and funding by progressive organizations such as the U.S. Defense Advanced Projects Research Agency (which was also the entity responsible for funding the development of the internet), has this technology begun to bear fruit.
The discipline of computer vision encompasses sophisticated approaches to shaping the world from the way the lens of a video camera sees it. Environmental phenomena such as wind, rain, sunlight and shadows are detected, analyzed and rejected as immaterial natural events.
Objects of interest such as people and vehicles are automatically identified and recorded by camera view. Intelligent video software uses these machine vision techniques to detect and track, and then analyzes the information to identify specific behaviors. For example, smart video can detect when a person is crossing a perimeter or loitering at an ATM or when a car is parked in a forbidden place or for an unusual amount of time. There are three important criteria for determining whether a system, device or solution is truly intelligent:
First of all, it must be an application of artificial intelligence, specifically of computer vision. These are important because computer vision, the science that teaches a computer to see, is the foundation of all vision-based systems that are truly intelligent.
Secondly, the software must be able to accurately and reliably separate the objects in the foreground from those in the background. This is relevant because a truly intelligent system is one that can interpret and report on real events, not just react to movement.
Third, the system must be able to provide the user with important information about these events, ultimately in the form of metadata—a complete description of everything that is going on inside the scene at any given time. This is important because this data provides the user with the foundation to perform a highly efficient search and allows them to leverage the video infrastructure to gain valuable insights for their business.
The possibility of intelligently automating the video source monitoring process may sound like science fiction, and, considering the promises of some vendors, unrealistic expectations may have arisen about the real possibilities of the technology.
What can you do?
Faced with a provider's unrealistic promises about video analytics, a good rule of thumb is to ask the following questions:
"Could a human watching the video perform the same task?" If the answer is "no," then it's also not very unlikely that a form of video analytics can do it. Even if the answer is "yes", it is still possible for the possibilities of video analysis to try their best. Remember, human vision has been evolving for about five million years. Computers have only done it for about forty years!
But, even with the limitations of smart video, it is currently used in several sectors, including airports, ports, borders, manufacturing processes, facilities management, retail and in the banking sector. This technology is helping to improve the effectiveness of security in a smarter, faster and more efficient way.
For example, intelligent video is used in the rail transport sector to monitor the perimeters of railway centers, monitor large concentrations of people on platforms, reduce vandalism, ensure the safety of passengers with cameras inside cars, prevent people from sneaking in without paying and control mistakes in the entry and exit of passengers, among other applications.
The old, the new and the sensible
So what examples are there of technologies currently recognized as smart? Below we include two of them:
A newer technology is facial recognition with biometric technology, one of whose close relatives is Vehicle License Plate Recognition (LPR). Although both technologies are based on vision and in effect analyze what they are seeing, they are based on probabilities. That is, they rely on the relative quality of the viewing to do their job and, then, can only compare what they capture with images they already have in the database. So—unless you think a relational database application is 'smart'—facial recognition and LPR aren't smart to the same extent that machine vision-based video is. In addition, these technologies, which are touted as smart, are rarely compact, so be prepared to make a heavy investment in end servers (back-end) and enterprise software to use facial recognition or LPR in your video surveillance environment.
The other technology that is most often touted as intelligent is Video Motion Detection (VMD). It seems that from time to time this dinosaur is released to the market with new flavors and modifications, but do not be fooled, the VMD is definitely not smart. Why not?
The VMD does not analyze, it reacts to movement. It does not interpret events, it reacts to movement. It does not classify objects, it reacts to movement. So, video motion detection technology has a fitting name: in its most elementary form, any movement of any pixel at any time triggers an alarm.
How many times does the wind have to move the branches of the trees or the birds fly in the scene so that this system is turned off by all the false alarms it generates? Interestingly, a major camera manufacturer installs VMD technology on the camera (claiming it's a "cutting-edge smart device"), then asks for trademark filtering software to be installed on the end servers.
Filters must be carefully configured for the size of objects and according to a number of other parameters, all to reduce false alarms to a tolerable level. Despite all that the filtering software does, the camera's VMD is still merely reacting to movement, which is not smart at all.
A major U.S. security company, which makes a huge number of devices and systems, has announced another motion-sensing alternative. According to this company, the total system is the only way to achieve "intelligence". But what form of "intelligence" does this company actually offer? They even go so far as to promote the solution they offer as 'Motion Detection by Intelligent Video', a contradiction in terms that all it does is increase the confusion that reigns in this market.
Money is what counts
Software companies, device manufacturers, decentralized (end-to-end) solution providers all argue that their products are smart for one reason: money.
Hastily adding a VMD system to a device that didn't have it before, then calling it "smart" and finally recharging up to 30% on the price isn't going to provide the end user with the genuinely smart possibilities it requires. These companies were late to the game and are eager to catch up. And they try to achieve this by creating noise and confusion in the market, rather than offering products that actually offer embedded video analytics.
Such manufacturers like to partner with smart video technology providers, who have been in this market for some time; they do so because they know they will never be able to invest the enormous amounts of time and money in research and development that are needed to bring to market smart technology that is commercially viable. Other big firms have acquired smaller video analytics companies and, with them, their intellectual property, or they have given up and put their money into the VMD they already have.
* Rafael Ramírez is the regional sales manager of the company i3 International. If you want more information about this article please contact the author via email: [email protected]
