Wednesday, December 24, 2008

Using Video for Detecting Perimeter Violations

I have mentioned John Honovich’s web site about video surveillance on a couple of occasions. One of the interesting things about John’s site is that he has a wide variety of industry experts contributing to the site. Yesterday, after receiving my weekly update email from John I went to a recommended discussion by Nick Grange of C3 Shared Services about setting up video analytics for perimeter intrusion detection. False Positives One of the problems with and intrusion detection system is the issue of false positive alerts. You certainly want a system sensitive enough to detect anyone crossing the detection perimeter. Setting the system to be too sensitive, however, results in alarms when there is no penetration, a false positive. Every time there is an alarm there has to be some sort of response. Depending on the distance between the perimeter and the actual security zone this could be a full tactical security team response, sending a roving patrol/guard to check out the situation, or reviewing the video record of the area. Nick noted that his firm tested a variety of video analytic systems at a site and found a wide range of false positives, from a low of 175 to as many 16,000 in the week long tests. Depending on what your alarm response is, 175, or about one per hour, may seem like a lot, but 16,000 is way to many for any security program. I would suspect that the equipment supplier never warned their customers about the potential for high false-positive rates on the 16,000 false-positive system. System Optimization The other good thing about Nick’s short article is the discussion about system optimization. I particularly like his explanation of how his firm selects the distance between cameras in a perimeter surveillance system. The one point that he does not emphasize is that those distances are for optimum situations, flat and clear terrain. If there are ditches or other obstacles those distances may be greatly reduced. Even so the 60m per camera and 3 to 4 hours per camera fine tuning and system adjustments gives you an idea of how long it should take a contractor to set up a system for your facility. Too much longer or shorter in the contract bid ought to be questioned. Nick’s comment about monthly maintenance of the system should also be considered when bidding out such a system. Intrusion Detection Systems There are a wide number of different types of systems that can be used for perimeter intrusion detection. Each of these systems has its strong points and weak points. John’s site concentrates on video systems, so it should not be used as the only source of information about choosing such a system. Unfortunately, I have yet to find a comparable single-point information source for other intrusion detection systems. One point in favor of a video surveillance system is that it provides a video record of the alarm situation. This frequently allows for a rapid determination of most false positives. This reduces the need for sending someone to the site of the alarm for investigation purposes. This is one of the reasons that video systems, without the analytics software, are used as a backstop for other IDS systems. Recommendation This short article by Nick Grange is certainly worth the time to read it. It will not make anyone a video analytics integrator by any means. It will, however, provide some additional information that will allow a facility security officer to be a more intelligent consumer of security services. While you’re there on the site, you might as well browse through the other offerings on John’s web site.


John Honovich said...

Hi PJ,

Thanks for examining this article and commenting on it.

My general recommendation for security managers is to be very careful when selecting video analytics. This technology suffers from incredible overselling. A lot of security managers have horror stories as a result.

I hope Nick's article helped to dispel the notion that video analytics are magic. You need to be very careful about selecting video analytic systems, and for managers of chemical facilities you may need to use multiple analytic technologies in combination.

If anyone has questions, feel free to email me at I am always happy to offer advice and feedback. It helps me as well better understand the challenges of security managers.



Anonymous said...


Nice summary on what I believe is one of the most the most critical elements of video analytics, the ability to filter out false-positives, without missing a real intrusion. After all, if it misses the real event or bombards you with too much noise, it is a bad investment.

Your comment,”I would suspect that the equipment supplier never warned their customers about the potential for high false-positive rates…”, is definitely true, but I don’t believe it is intentional in all cases. I believe the problem stems from manufacturers who overstate their capabilities, and integrators who don’t know (yet) that the products they are representing have had in many cases very little field testing in real environments. Anyone ever tested on melting snow? Doubt it, but I am right now and it is wreaking havoc on the analytics. It’s conditions like these that are unproven and weaken the confidence with the buyer when they end up making these types of discoveries about their investment on their own.

The second most important element of deploying video analytics is cost. Every organization will be faced with two cost categories; CapEx (capital investment costs) of the entire system and OpEx (operational and maintenance costs) to use and mainatain the system. CapEx can vary significantly, and need to be matched to the customer’s application: the budget for securing a nuclear facility is probably 100 or 1,000 times greater than that of a construction oe small/medium business. Also, you should consider infrastructure costs as part of buying a whole new system.

But OpEx is by far the most overlooked cost, as you say: “Every time there is an alarm there has to be some sort of response.” Imagine a Central Station receiving 175 false alarms per week from a single site: you can bet the monitoring company will refuse to service this customer. As consultants to the industry, we should insist that solutions meet MUCH more stringent false to positive ratios. A system we tested recently performed at a 1:38 ratio (yes, that’s one false positive for 38 true detection events—in a blizzard, no less!).

Overall we have had a significant amount of success with our clients providing pilot tests. During a pilot test you, as the buyer, get to experience the capabilities of the system before making a huge investment in technology, infrastructure, and operations. For little upfront investment you can see firsthand how (or if) you can utilize and benefit from the technology. To contribute to your research and others interested in the subject, I would like to provide a few more points that we’ve researched. We have discovered these through our testing and would encourage you to do the same:

-Determine where the analytics is performed. It will usually be at the central recording location, in a remote field appliance, or on the camera. If the camera is performing the analytics, determine if it is post-compression or right off the CCD/CMOS.

We have found that cameras performing the analytics onboard--before the compression of the video--are less susceptible to false positives.

Further, reducing the number of overall system components also serves to lower the cost, complexity, and reliability of installed systems.

-Review the installation and setup practice. Can you train some of your non-IT expert staff to perform the setup and configuration of the software or make changes to the system? Can your maintenance crew install the cameras?

To be really commercially-viable (e.g., non-nuclear plant), a system must have low labor costs. Some of the systems we have researched are quite plug-and–play capable from an infrastructure perspective. Some even have ex-cops installing, configuring, and operating them.

-Understand the remote support options.

We have found that many systems are moving to web-based software packages. This means you do not need to load any client PCs with software—thus reducing complexity and upgrade issues. Since it is web based, having remote access is crucial to fine tuning of the system. (Some tuning always occurs after initial setup to deal with unanticipated events such as reflection, swaying objects, etc.) These tasks must be performed 1) quickly, 2) remotely, and 3) by lower-cost operators in order to keep costs in line.

Taking 4-12 hours (over multiple days) to tune each camera is NOT a cost-effective practice.

Finally, I completely agree that John’s website, is incredibly enlightening and easy to source good information about advances in IP Video.

Best regards,

Ryan Taylor


/* Use this with templates/template-twocol.html */