Cognitive intelligence-based video analytics will benefit security efforts
- By Kurt Stoll
- Mar 02, 2009
Security directors face large problems in large facilities. Officials at airports, seaports, industrial facilities and other large installations deal with a unique set of security problems. They also have a unique set of limitations. They must protect against a variety of security threats, many of which are unknown, and they must address these issues with limited manpower. They also are dealing with creative enemies who are always adapting and enhancing their sly methods.
These are the challenges that large-scale facilities must overcome to ensure the safe operation of services on which consumers and citizens depend.
Historically, these facilities have used a variety of approaches to help meet physical security challenges. Many rely heavily on video surveillance systems, while others use police officers or security personnel as supplements to their video surveillance systems. The upside to having a lot of manpower to protect a facility is that officers are on site to stop criminals. But security personnel can’t be everywhere at once, and since it’s impossible for anyone to know exactly where or when the next incident will occur, trying to cover all of the bases can be cost prohibitive. In fact, one large U.S. seaport estimates it spends about $10 million a year for police protection alone.
Using a large number of cameras can help as well, but as the number of cameras goes up, video becomes increasingly difficult to manage. Having cameras does not ensure that they will be watched, and reviewing past footage does not guarantee future security. Recent successful terrorist attacks illustrate how forensic video analysis after the fact is simply not adequate—if there is no possibility of prosecution, there is no deterrent effect. Therefore, the need for real-time responsiveness to potential threats is critical. Large-scale organizations, in particular, need a solution that enables them to make the best use of available resources to pinpoint incidents and empowers them to respond to potential threats in a proactive manner before they evolve into actual disasters.
To meet these needs, security organizations are working to incorporate more advanced and effective technologies that offer improved visibility. More sophisticated forms of object recognition and motion tracking have evolved to provide a heightened sense of awareness in a variety of video surveillance environments. These rules-based systems have become highly specialized for different environments, whether they are focused on perimeter detection, surveying large crowds or watching for abandoned vehicles or dropped objects. However, these rulesbased systems also have limitations of their own.
Every environment and every scene is unique. No one is able to write enough rules to cover the infinite number of possibilities for any given environment. Rules-based systems also typically require extensive programming and calibration, making it difficult for users to quickly scale or achieve broad market adoption. Finally, rules-based systems historically generate too many false positives and have become labor intensive to set up and maintain. So, if rules-based video analytics is not the answer, what is?
The ability to create an interconnection between vision analytics and a system that emulates the cognitive process—using various machine intelligence and machine-learning technologies—represents a breakthrough for the video surveillance industry. This connection creates a system similar to the human brain; it is called a cognitive-based video analytics system because it can see better, as well as learn, remember and make observations.
Through its observation, a cognitive-based video analytics system assesses a given environment to build a mental model of the scene. It observes patterns of behavior— understanding the normal flow of traffic in and out of a given entryway, for example—to establish a standard of normal activity. Learning is achieved when the mental models adjust as the scene changes. The system interprets and alerts, if necessary, on new activities as they occur within the context of previous activities. Through an observe-and-learn paradigm, the camera creates an understanding of what it sees and establishes normal behavior for an environment. It is therefore able to alert on activity it determines to be abnormal.
Realizing the Benefits
In a vulnerable environment with hundreds of cameras all observing a variety of changing scenes, it is especially important to have a cognitive-based system that is able to learn what is normal for every unique environment and then alert when activities occur outside that normal pattern. Cognitive-based security observes and refines its model of a scene automatically, allowing it to detect, track and classify more efficiently over time.
A system of this kind minimizes labor and software upgrade costs and improves the effectiveness of operators and security personnel by allowing them to focus on events that have the highest probability of being actual threats. A learning capability also is an important component in order for the system to adapt to changes that may occur within any given environment over longer periods of time. Because these systems are able to learn behavior patterns over time, organizations can find out where the areas of greatest risk are and direct available resources to those areas.
These systems also provide real-time alerts, allowing staff to respond immediately to security breaches occurring out of sight.
These capabilities—to adapt to almost any scene or environment and to continue to improve upon its learning and alerting over time—are the most important distinguishing factors of cognitive-based systems over rules-based video analytics systems.
The benefits to businesses that adopt cognitive-based video analytics systems over rules-based systems can range from reduced costs due to less required coding and customization, increased effectiveness from reduced false positive alerting and increased return on investment on the entire security infrastructure.
This article originally appeared in the March 2009 issue of Security Today.