Free Newsletter
Register for our Free Newsletters
Zones
Access Control
Alarms
Biometrics
Detection
Deutsche Zone (German Zone)
Education, Training and Professional Services
Government Programmes
Guarding, Equipment and Enforcement
Industrial Computing Security
IT Security
Physical Security
Surveillance
View All
Other Carouselweb publications
 
 
 
 
 
 
 
 
Editor's Blog and Industry Comments

Making intelligent choices when considering video analytics

19 June, 2008
With customers seeking added value from increasingly function rich and expensive camera systems, video infrastructure vendors are offering intelligence but do they pass the analytics IQ test to justify being called intelligent?
With CCTV surveillance traditionally having driven manpower with guards monitoring banks of screens to see what's happening or analysts sifting through hours of tape to see what happened, the arrival of video analytics technology came like a breath of fresh air, adding more value to the surveillance investment, reducing the dependence on manpower and increasing operational effectiveness in both prevention and detection. The vendor base latched onto the possibilities and the technology has now expanded into a wide scope of offerings, along with the inevitable marketing hype.

Part of this hype is the use of the term 'intelligence' to describe anything that involves the analysis of images. But where do we draw the line; and how do we separate out intelligent systems from those that aren't and if the analytics perform the functions you want, does it even matter whether it's intelligent or not?

The bulk of video analytics packages available on the market today perform a specific function. There are automatic number plate recognition (ANPR) systems, video motion detection (VMD) systems, biometric face recognition systems and retail transaction fraud detection systems with cameras linked to EPOS terminals, all of which perform some kind of analysis functions and can add value to the video surveillance investment. However, the decision to purchase any of these systems should be based on real expectations of what they can offer and what their limitations are.

Five key questions can help to determine some of these limitations and help one understand how smart the system really is.

1 â€" In what way is the analysis intelligent?

Making database comparisons or reacting to triggers doesn't constitute true intelligence. For example, biometric face recognition systems work in two modes, either authentication or identification. In both modes, a set of parameters are compared to databases containing the same set of parameters and the closest matches are selected as the best fit â€" a data processing task. The smart thing about these systems is in the definition of the parameters to make the recognition task more accurate. The resulting analytics however, don't make this any more intelligent than any other database query mechanism.

Trigger-based systems such as VMD are reactive and based on the information provided by the user. You can determine the area of interest, the direction of motion and certain filtering parameters such as size of object. With these kind of systems, filtering is crucially important in order to reduce the number of false alarms raised, but this filtering ability doesn't add intelligence, it refines the trigger.

2 â€" Is the system based on computer vision?

In determining the level of intelligence offered by an analytics package, this question is crucial since computer vision, a branch of artificial intelligence, enables the software to actively analyse the environment, seeing activities occurring within the field of vision. With decades of research having gone into computer vision, it is at a point where real time environmental analysis is achievable in commercial applications. To illustrate the difference this makes, a VMD software overlay may view a scene as 'a person has crossed the defined demarcation from east to west' whereas a computer vision application would view the same scene as 'a person has climbed over a wall from east to west and attempted to enter a vehicle'.

3 â€" Can it separate foreground from background activity?

Foreground activity is interesting, the background isn't. Therefore, the ability to separate the two is an important factor in reducing false alarms. In the natural world, there is always activity and movement â€" the weather changes, leaves fall, branches sway, light levels change and shadows track across the field of view. Understanding these kinds of changes in the field of view is a key component of computer vision.

4 â€" Is metadata generated?

Metadata is data about data, in this case providing a narrative of all the events which occur within the camera's view, irrespective of the nature of the event. With trigger-based analysis systems, an alarm is raised when the trigger is activated and so peripheral activities are ignored. Therefore the definition of the triggers is important in order to capture all events of interest. The generation of metadata enables a far greater depth of analysis.

5 â€" How is retrospective forensic analysis performed?

We can't always define what we're looking for before an event happens, and one of the advantages of surveillance systems is the ability to gather evidence about events that have occurred within the field of view of one of the cameras. However, this analysis often involves laborious sifting through surveillance images in an attempt to capture those few moments of interest. With intelligent video analysis, the metadata provides a short cut to performing retrospective forensic analysis. By interrogating the data rather than watching the film, specific moments can be isolated accurately and quickly.


Having asked these questions when making a choice of video analytics, the decision falls to the application. There is room for all levels of analytics within the market but it is important to see through the smoke and shadows and make informed decisions since it's easy to be blinded by science and impressed by demonstrations.

If, however, your choice falls into the area of intelligent analytics, how accessible is the technology and how can it be implemented? ObjectVideo is a supplier of intelligent analytics for surveillance systems and I recently spoke to Ed Troha of ObjectVideo about this topic. The technology is accessible through OEM partnerships with ObjectVideo OnBoard rather than selling directly to end users. However, this doesn't mean being locked in to certain surveillance system suppliers since ObjectVideo is keen on opening up accessibility to the technology through open standards. OVReady is the open interface developed by ObjectVideo enabling communication with standard video management platforms. Through OVReady and the partnerships ObjectVideo has forged with suppliers in the video surveillance market, the technology is accessible to the majority of users operating within any video surveillance system.
Bookmark and Share