Spot On Security

The good news is that most high quality video analytics systems these days work quite well with IR lighting for night time detection.

The one thing that often gets overlooked, however, is that IR lights attract insects. Apparently they like the light and the heat. And insects often attract bats and birds, and sometimes even spiders, who occasionally spin a nice web across your camera lens to catch the bugs.

When you are just recording video, this only means you might have an annoying bug or bird flitting near the cameras. Not a huge deal.

But with analytics, it can cause false alarms.

I first learned about this many years ago. Our analytics have not had many reports of bug problems, but we’ve seen a few. Then, I ran across a report by Raytheon that summarized years of experience, including deployments in Middle Eastern deserts, proving the problems created by mounting IR illuminators too close to the camera. Their suggestion: Don’t mount IR lights near cameras.

Sounds like a good suggestion to me.

I just searched to see if I could include a link to the Raytheon study, but it doesn’t appear to be on the web any longer. However, I did run across this document:

http://www.dvmd.com/downloads/ApplicationGuide.PDF

This article says:

“The ring of LED’s around the lens of “bullet cameras” attract amazing numbers of flying insects in summer months.”

Bullet cameras with IR LEDs have become quite popular, but I’d suggest avoiding them with analytics. It is better to mount the IR lights away from the cameras and preferrably closer to the area you are trying to illuminate.

The article also makes this comment:

“Bugs can produce approximately a thousand nuisance alarm events per hour…”

We’ve never seen anything like that. In fact, false alarms are rare from bugs with any good analytics technology, but they do occur and so the wise thing is to avoid the issue by mounting the light sources at a distance.

The reason this article describes thousand of false alarms per hour is because they are talking about Video Motion Detection, not Video Analytics. This is just another example of why VMD is not reliable outdoors.

One other thing to keep in mind with nighttime detection: If you are using a day/night camera, you should know that some video analytics technologies have problems when the camera switches from day mode to night mode, or vice versa. When the IR cut filter switches in or out, it creates a sudden change in all the pixels. This can create false alarms for many technologies.

VideoIQ’s technology doesn’t have a problem with this, and I know there are other good video analytics technologies that don’t as well, but some do. So, if this is an application you are working on, you might want to check with your vendor first.

I first got intrigued with the issues facing City Surveillance after reading a study earlier this year that San Francisco published on the results of their video systems. I commented on this in an earlier blog post:

http://spotonsecurity.com/2009/02/06/san-francisco-surveillance-study-shows-need-for-analytics-and-real-time-response/

Then I began researching further and discovered that public studies by cities around the world were reporting almost identical results.

Now a new report just released by the UK Home Office with researchers from Sweden and the USA summarizes over 40 studies from around the world, and their conclusions are talking about exactly the same things that I noticed. Here is a link to that new study called, Effects of Closed Circuit Television Surveillance on Crime:

http://db.c2admin.org/doc-pdf/Welsh_CCTV_review.pdf

They show that traditional CCTV surveillance systems produce a small reduction on crime in some areas – on the order of 7% – but it has little or no impact on violent crimes and has its best results in parking lots and preventing property crimes.

What jumped out at me when I studied all of these reports was that they were missing the underlying causes of such disappointing results.

We’ve been seeing a significant growth in what we call Remote Guarding, where video analytics detect breaches of security or potential problems and send video clips to remote security personnel who are trained to respond through audio down to the site. This actually prevents the crime, compared with just recording it as in most traditional systems.

The results have been powerful in the commercial sector, for three basic reasons:

1. The cost of pro-active monitoring is now practical for the first time. Where one person looking at a bank of monitors might at best keep an eye on 10-20 cameras, and do a fairly poor job of catching things, the Remote Guarding system that uses video analytics can have one person responding to 1,000 cameras or more and doing a much better job seeing potential problems.
2. Traditional systems have not been able to respond quickly enough, even if they can detect a crime in progress. The time it takes for police to get to the scene means that they are often there too late. This is one reason why existing systems have had almost no impact on violent crimes. If you can have an instant response, you can stop a dangerous situation from escalating. This is far more powerful than mopping up after the murder or attack.
3. The quality of the video was often poor because frame rates were too slow and resolution was scaled back in order to reduce bandwidth and storage costs. Storage in the camera and using analytics to intelligently control the recording solves these problems, as I’ve mentioned before.

These are issues that we’ve been solving in the commercial sector with Remote Guarding for the last few years, and are exactly the same underlying issues cited throughout all the city studies I read.

In other words, Remote Guarding using video analytics for detection and IP audio for response has the potential to transform city surveillance.

The commercial sector is seeing 50% to 75% reductions in costs compared to on-site guards, while improving protection and reducing crime dramatically. There is no reason I can think of that this same formula won’t have the same impact on preventing crime in our world’s cities, public parks and hotspots.

Security Systems News just ran an interesting article talking about the growing interest in private monitoring of public places:

http://www.securitysystemsnews.com/?p=article&id=ss200905c7kwET

In there, you’ll see the first case of Public City surveillance using Remote Guarding. Birmingham, Alabama, has contracted with ION Interactive, who has deployed VideoIQ’s iCVRs throughout the city, along with some commercial locations.

Richard Cruit, the COO of ION Interactive,  and I just ran a webinar this last week, sponsored by SDM Magazine, talking about the results in Birmingham and how Remote Guarding can have a significantly positive effect on City Surveillance.  You can watch this webinar from SDM’s archives by going to this link:

https://event.on24.com/eventRegistration/EventLobbyServlet?target=registration.jsp&eventid=144275&sessionid=1&key=CF33F1A5642D363ED9BDA7202D2EBFD8&partnerref=website&sourcepage=register

It is time for a new approach to city surveillance and how to use video to reduce crime. The technology is here, and the results are coming in every day that prove how effective it is. With all the money being invested, we shouldn’t be settling for small reductions in crime, when we can make our video systems pro-active and help break the cycles of crime before they happen.

If you’ve been working with video over the last decade, you’ve probably heard the myth that MPEG type compression has been rejected by courts because it records only the changes in a scene. This simply isn’t true. It’s one of the strangest myths, and I’m surprised it is still alive, but I just heard it again a few weeks ago.

I apologize for the long post here, but I hope to bury this bogus idea once and for all.

I first ran across this issue about 10 years ago when some of the rapid transit authorities who were interested in moving from analog to digital video systems in their buses were told by some competitors that the video would not stand up in court. Their legal departments were concerned.

So, I researched the matter to see what, if any truth there was to it. What I discovered was exactly the opposite. Courts had already dealt with these questions and ruled the video was admissable.

One of the most notable opinions came from a high level judge in the UK, who wrote her opinion as a summary for how all UK courts should treat digital video evidence. It came after years of working with cases and seeing the need for a conclusive opinion. Her comment was that even if the video was not watermarked, it should still be accepted as evidence. However, she highly recommended that all manufacturers watermark their recorded video. Watermarking adds a great deal of authority to the video evidence by verifying that it has not been tampered with.

Her other emphasis was on how the recorded video should be handled when used as evidence in court. Having a well documented chain of custody and clear processes and procedures for how to handle such evidence until the moment it ends up in court, was the most important thing of all. This was how all evidence should be handled to provide credibility that it is what it claims to be.

There were a number of other published papers by expert lawyers who had handled cases involving digital video. They made the matter even clearer. Courts regularly allow faxes, emails, photographs and audio recordings as evidence. All of these can be tampered with. Most are, or can be, modified in some way electronically to produce copies for use in court. The only real and valid concern is whether there is any indication that the evidence was in fact altered, or if anyone who handled it had a history of modifying evidence. Otherwise, most courts won’t even allow lawyers to suggest that the evidence might have been tampered with.

One lawyer, as an example, went on to explain how easy it is to fake photographs. People have done this for years. But simply the possibility that a picture could be faked is not enough to reject it as evidence. If it were, no photos or any other electronic documents such as emails or faxes would ever been usable in courts.

So, there is no truth to the myth.

However, besides the legal side, the whole technical reason for rejecting MPEG type video compression makes no sense.

One of the arguments I’ve heard was that if only the changes in the scene were recorded, then those changes could be overlayed onto a different background to make it look as if the person had been somewhere else. True, but you can just as easily do the same thing with JPEG video or even lossless compressed video. It is easy to extract moving people from a static background. Good watermarking, however, easily establishes whether this kind of alteration has taken place.

Besides, it isn’t true that only the changes are stored with MPEG type compression. Usually, once every second or every few seconds, a full copy of the scene is captured and recorded. These are called “I-frames”.

Secondly, each frame is re-assembled during playback. You can stop on any frame and see the whole picture. You don’t see only the changes, but the full scene.

I’ve had some odd discussions with the FBI about this matter. They seem to be one of the last strongholds for this myth.

You would think that the FBI would know. And I’ve seen a lot of people accept these concerns as valid simply because the FBI says it is a real issue. So, I’ve tried my best to speak to the experts at the FBI to find out their reasons. Whoever I have spoken to, however, has not convinced me they understand any better than anyone else, and its quite clear that a number of people in the FBI are quite ignorant about how compression really works. So, this is actually a technology problem.

What the FBI states that they want utlimately is lossless compression, so that no data is lost. They seem to think that would be the best, but even this is the wrong thing to be focusing on. On top of that, it is completely impractical. It could take over half a terrabyte of storage for a single day of lossless compressed video. No one is going to pay for something like that, or the bandwidth to transmit it, especially when there is no good reason to do so.

What makes the request for lossless compression the wrong goal is that they are willing to settle for fewer frames per second and lower resolution video as long as they could get the lossless compression. This is exactly the wrong compromise to make when you want video to be used as evidence.

I’ve explained to them and others that the last thing you want to do is throw away frames or resolution. This is value image data. You can’t identify people without enough resolution. You can’t see critical events without a fast enough frame rate. Why would you ever want to trade this data for lossless compression?

More importantly, compression technologies have been specifically designed to eliminate the data from video that is the least important. It is data that they eye can’t see or doesn’t notice, even if you study the video carefully. Good compression can only be recognized by experts, and for most people it often looks identical to the original. Of course, this is good compression I’m talking about, and many times we see recorded video that has seriously compromised the quality. But it doesn’t take lossless compression to get good video, which is why MPEG-2 has been used for years in DVDs.

The whole idea that MPEG type video is somehow modified while JPEG is truer is another weird idea that simply isn’t true. The people who start these kinds of rumors just don’t know how compression works.

If a system could accurately extract only the changes in a scene and record all those changes, you would have lossless compression. In fact, lossless compression uses this exact technique to shrink the size of the data. You don’t lose any information if you don’t send the same information over and over again with every frame. There is no need to send it more than once, except for the possibility of data corruption. Sending the whole frame over and over does make the data more immune to corruption, which is exactly why MPEG video sends I-frames so often, but that is the only reason to send it more than once.

JPEG, on the other hand, makes lots of compromises with the video data. The number of colors are limited, the edge information and detail is simplified, and contrast data is reduced. Look at it this way: A good JPEG image might be one-tenth the size of the original image. That means that JPEG is throwing away 90% of the data. But the end result with a good compressed image is that it looks almost identical to the original, to our eyes.

Since MPEG type compression, called temporal compression because it is compressing across time, only differs from JPEG because it throws away data that is being repeated, this actually throws away a lot less information than the spacial compression of JPEG. So, if anything, temporal compression thows away less information. In actual fact, the I-frames that MPEG sends use the exact same method of compression that JPEG does, but the added temporal compression focuses on reducing the same data that was already sent or recorded, since it adds no more information.

So, the whole idea that MPEG is throwing away valuable data while JPEG is not is simply wrong.

I’ve heard other arguments and concerns. The best one is that a lawyer might cast doubt on the validity of the video when they tell a jury that the whole frame isn’t being recorded. This concern is easily overcome by simply explaining as I did above: No information is lost when you can recognize which pixels have changed and only send the data on the changes. There is no reason to repeat the data that didn’t change. It gives you no more information. Nothing is lost by this. Lossless compression uses the exact same approach.

More importantly, MPEG type compression, which includes the latest H.264 technology, allows you to record 4X – 10X as many frames per second in the same storage space and with the same bandwidth – or you can record 4X – 10X higher resolution video. Using MPEG type compression, therefore, means you can include far more valuable and important data that can help you identify people and what has happened at the scene of the crime.

Clear pictures with higher resolution and close to full motion video is the most valuable recording you can provide to court for evidence. The video should be watermarked. And the video evidence should be handled using well established chain of custody procedures. All the rest is just a modern day myth based on misunderstandings of technology.

It is time to put this one to rest.

We often get asked if we license our video analytics technology to other companies, so they can embed it into their products.

We have chosen not to license, which often raises the question: Why not?

There are actually quite a few reasons, but here are some of the most important:

1. The licensing approach to selling technology works well in mature markets with a lot of consolidation. This is what you find in the Telco and IT Networking markets, where one or two vendors dominate market segments. But this is just not true in the Security Market. There are boatloads of vendors, and even the largest generally have less than 20% market share. Most are a lot smaller. So, you have to license a lot of vendors before you command any significant share of the market. Some companies, for example, have licensed 30-40 companies or more, which sounds like a lot, but they still represent a tiny fraction of the total products shipped.
2. Licensing only works well when the technology is ready for widespread use and when it is ready for rapid market acceptance. This means the technology has to be easy to use and work well. Unfortunately, all of the companies trying to license are finding that they are too early for mainstream markets. It takes too much time to calibrate and tune cameras, and even after all of this effort, they still have too many false alarms or missed detections.
3. Every new company you license requires working with a different design to embed the technology into their products. If the technology is easily reduced to a simple chip you can plop in, then licensing makes more sense. But analytics technologies today still require a lot of design support – almost as much as designing new products. On top of this, analytics are still rapidly improving, which means that major improvements could force major redesigns.
4. If the technology is a real simple add-on feature that everyone wants to use, then licensing works. But the problem with licensing video analytics is that you end up with camera and DVR companies selling products that they can’t completely support and don’t fully understand. When the technology they are licensing is far more intelligent and complicated than the rest of the product, it makes a real mess of a problem for integrators and end users. You can’t get answers and can’t get problems fixed when you need them. The closer you are to the people developing the core analytics algorithms, the better your support will be.
5. The biggest issue of all, however, is that we never believed that “detection” was the only thing that mattered with video analytics. Analytics can make the whole video system smarter, and solve key issues, such as bandwidth and storage. It is not just about detection. This is a radical new idea that we felt needed to be demonstrated first.

In other words, the power of analytics to make your cameras smarter, to control bandwidth and improve storage quality, along with many other benefits, are just as valuable as the ability to detect intruders early to prevent crimes before they happen.

But that’s a whole new product idea, not just an add-in technology.

The other vision we had was to sell intelligent IP cameras for about the same price as dumb IP cameras. Wouldn’t everyone want the smart camera instead? This makes it an easy choice to start using analytics when it doesn’t cost any more, and solves your bandwidth and storage problems. Isn’t this what the market wants? But camera manufacturers paying license fees to add in analytics will always have to charge more to cover their added costs. So, the licensing approach makes it harder to reach to ultimate goal.

That’s some of the reasons we haven’t licensed our technology.

Image sensor technology is outpacing lens designs, which is leading us into a strange world.

You can now buy 5 megapixel surveillance cameras from a number of vendors. However, there are no 5 megapixel lenses!

Sound strange? Unfortunately, it seems to be true. If anyone knows of a real 5 MP lens for surveillance, I’d love to hear about it.

We’ve talked to all the major lens manufacturers. They all advertise megapixel lenses, but they don’t say how many megapixels their lenses are designed for. When you push them for an answer, you find out that none have 5 MP lenses.

We even asked the lens manufacturers who make the lenses that the megapixel camera vendors recommend for their 5 MP cameras. They told us that their lenses were good for 1 MP or sometimes up to 2 MP. We haven’t found a 5 MP lens yet, except for the fisheye lens from Theia.

This means that you aren’t going to get the full value of all those extra megapixels. You might still be able to cover a much wider area than a single lower resolution camera, but it won’t have all the detail you think it will.

Our camera expert at VideoIQ, Steve Lefkowitz, who has been working with lens vendors for over 20 years, asked them a good question:  If they can design lenses for the 8 MP and 10 MP point and shoot consumer digital cameras, why can’t they make one for surveillance?

Apparently, the situation is no better for consumer cameras: The imagers may be capable of 8 MP or 10 MP, but the lenses fall far short. They don’t even come close.

This is bizarre.

Here’s a little extra info on lenses. When evaluating the quality of a lens, there are three main specs to look at:

1. Resolving power or resolution: This means how many horizontal lines the lens can accurately distinguish. But you need to be careful that you find out the resolution not just in the middle of the lens, but check the edges and corners as well, since they generally have much better resolution in the center. A good lens is sharp to the edges.
2. Geometry or distortion, which means how round a circle will look anywhere in the scene. A circle will often look perfect in the middle, but looks like an oval in the corners. This kind of distortion is common. A smart digital camera can actually correct distortion like this, but as far as I know the only MP company doing this today is Mobotix.
3. Flatness of field, which means that when the lens is in focus at its center, it is also in focus in the corners. Often this isn’t true. No, it isn’t your eyes, it’s the lens.

Another little know fact that our expert Steve told me: Everyone looks at the F-Stop spec, but this only shows the amount of light coming through the iris. The real spec to know is the T-Stop, which tells you the amount of light coming through the iris and all of the lens elements. A good 1.4 F-Stop lens might have a T-Stop of 1.5, which means that the lens elements aren’t blocking much of the light. But a poor 1.4 F-Stop lens might have a T-Stop of 1.8.

If you want more info on megapixel lenses, check out John Honovich’s recent column:

http://ipvideomarket.info/report/the_importance_of_megapixel_cameras

I think it would be a big help if the lens manufacturers started being more up front about their full lens specs. I can’t imagine why the good lens manufacturers wouldn’t lead this change, since the specs will show how much better their lenses are. Why not show why it is worth paying more for good quality?

Unforunately, dealers are left to figuring this out by trial and error. Dealers come to learn the hard way that some lenses are a lot better than others, but why put the burden on them to figure this out? Hiding the specs doesn’t help anyone.

We’ve seen the IP camera manufacturers coming together to establish standards. Why not the lens manufacturers?

I ran across the following article from John Honovich at his IPVideoMarket.info web site, which is chock full of valuable information about video surveillance:

http://ipvideomarket.info/report/do_video_analytics_work

I highly recommend John’s site.

In this article he raised his concern about the gap in perception between manufacturers of video analytics equipment and the users. Integrators who have tried using analytics products often say it doesn’t work, while all the manufacturers that John talked to said their technology does.

How can this be?

John writes:

Manufacturers generally have a significantly lower standard for determining what works than customers or integrators.  This is not an accident yet it is generally not an issue of malice.  Most manufactures, especially at the senior management level, possess little domain knowledge, resulting in routine underestimation of the needs of their customers.

This is certainly true, and it is a big factor. Many video analytics companies have come from out of University labs or from Computer Vision developers. They saw the security industry as a great place for their technology, but had no real experience in what security professionals have to face in the real world.

As John points out there is also a gap between what a new technology is capable of. When people see a breakthrough, the imagination can immediately jump to Star Trek tricorders and dylithium crystals.

However, I think there is another very big factor at play in this case. It has to do with calibration and tuning.

The best of the Video Analytics companies have tried to simplify this process and make it as straightforward and repeatable as possible. Only VideoIQ has eliminated the need for calibration and tuning, while providing accurate detection. As a result, we’ve seen a much smaller gap from our customers. Our end users are generally surprised at how well the technology detects. They, of course, hope we continue to improve, but rather than being disappointed they seem to be pleasantly surprised.

Here is what I think is happening: Manufacturers are familiar with their technology and their products, so they have learned how to tune and calibrate their systems to make them work well. However, integrators and installers, even after training, will never know how to program their product as well as the engineer who designed it.

I was talking to an engineer, at our VideoIQ booth at a trade show, who ran tests at Cisco when they were evaluating the leading video analytics technologies. He no longer works there now, but he told me about one of the products they thought was the most promising. He said it worked great.

I asked him why they hadn’t adopted the products into the Cisco line, if they worked so well. He said:

“When the head engineer installed the products, they just plain worked. It was remarkable. He knew exactly how to set it up. But, even after our engineers were trained how to calibrate and tune the system, we could never get it to work reliably. Only the head engineer could make it work.”

In other words, a big part of this perception gap comes because manufacturers know how well their products act when installed properly. That’s their perception and that’s what they mean when they say it works. But integrators see only how well the systems work when they try to install them.

What matters is not how accurate and reliable a technology is when a trained specialist installs it – but how well it performs when any video installer puts it in.

That’s an important gap to keep in mind.