VideoIQ

Archived Posts from this Category

May 25, 2010

Getting Our Head Out of the Clouds

Posted by Doug Marman under Edge Storage, Remote Monitoring, Video analytics, Video storage, VideoIQ
1 Comment 

Cloud computing has generated huge press lately. It is being called the next big transformation in Information Technology. So, naturally, we would expect to see cloud computing being promoted in the physical security industry, especially with video surveillance, which is so compute intensive. But this is where the myths begin. Don’t get caught up in the marketing hype. Cloud computing is not going to be a big benefit to video surveillance.

There are major economic advantages driving cloud computing in the IT world, but those same economies don’t work with video surveillance. This is glossed over and never mentioned in all the articles and ads that tout cloud computing for video surveillance. Companies just want to use the “cloud computing” label to get attention, when, in fact, if anything, video surveillance is moving in the opposite direction – more and more towards intelligence and storage at the edge.

Why doesn’t cloud computing work for the video surveillance market? After all, video compression and storage are some of the most processor intensive functions in physical security, especially if you try to add video analytics processing. But there are big differences, and once you look at the math, it is clear that someone has their head up in the clouds to be suggesting it be used for surveillance.

There are three big cost advantages coming from cloud computing in the IT world:

First, it is more economical with virtualized services. This means that you can save money running your sales management software tools on the same server as another company’s manufacturing resource planning software, or whatever other kinds of software other companies might want to run. Servers have been doing this for years in data centers. Now, they can do it across the Internet for many companies in one location.

Second, you get added savings because you no one uses their software all of the time, it is usually only needed in spurts. So, you reap the gains by sharing servers with others, by maximizing shared processing power.

Third, bringing  lots of server applications under the roof of one specialist company means they can manage everything more efficiently and take the headaches away from lots of individual companies trying to do it themselves.

For the above reasons, cloud computing service providers are saving IT managers 25% or more. That is significant. This assures that cloud computing will grow fast.

But these benefits don’t fly with video surveillance. Here are the problems:

  1. You can’t virtualize NVR storage, because the video is recording continuously. This is the same problem that IP video systems have in the data room. IT managers would like to combine them onto their other servers, but they can’t do it because the video is streaming into the servers continuously. It is being recorded non-stop. In fact, video recording is so processor intensive that traditional servers fail, which is why the industry has introducing new server designs specifically for video surveillance. You can’t virtualize the NVR and run it on another server.
  2. Most software applications see people retrieving data about 50% of the time  and storing data 50% of the time, but those applications are generally idle and waiting most of the time. NVR storage is recording 100% of the time, with huge bandwidth demands, while people only play back the video less than 1% of the time. The reason you can only run about 30 cameras (this varies by resolution and frame rate being recorded) on a good server is because you run out of bandwidth and processing power. So, there are no benefits gained from idle time.
  3. In fact, it is far more expensive to centralize storage because of the above problems and the huge bandwidth needed by the video. This is why most hosted video services are finding that it makes far more sense to put a DVR or NVR at the site with the cameras, or to use cameras with built-in storage, rather than continuously streaming the video back to a central server. It is far less expensive.

As I said above, the real technology shift in video surveillance is in the opposite direction: Intelligent storage at the edge.

This becomes even clearer when you start talking about video analytics. Some companies offering a hosted video service are trying to stream video back to a central server. They are trying to make it sound as if cloud computing is going to save the end user money. In fact, their solution is more expensive, but if they don’t put in a DVR they can save up front costs for the customer. Unfortunately, they seriously cut back the video resolution and frame rate: You aren’t going to get anything close to high quality video this way. They won’t even be able to offer standard resolution, never mind megapixel video to their customers. It is more expensive, on an ongoing basis, but it does eliminate the cost of the DVR.

But when you try to offer video analytics, you can’t play that game: You need high quality, high frame rate video. That makes the so-called cloud computing video surveillance system a non-starter. You can’t get there from here.

This doesn’t mean that Hosted Video and Security as a Service (SaaS) aren’t growing fast. They are, and for good reasons, but none of those reasons have to do with cloud computing economics. They are growing because of superior ease of use for end users, friendly browser interfaces that they can access from any computer, and by providing remote services for end users. It is an ideal solution for small businesses and small retail stores that want to add security video.

The term “cloud computing” has been hijacked and is being treated as the next big thing for security. It is causing a revolution in the IT industry, so why shouldn’t it for surveillance? Well, because it ends up costing more, not less, to use it with video.

 

May 17, 2010

As the PTZ Camera Turns

Posted by Doug Marman under Cameras, Megapixel, PTZ cameras, Remote Monitoring, Video analytics, VideoIQ
Leave a Comment 

The world of pan, tilt, zoom security cameras has changed. Megapixel cameras and video analytics are shifting the role of PTZs.

PTZ cameras were once king of the hill. They represented the best possible technology you could get; giving you the ability to see in every direction  and zoom into the smallest details.

Companies like Pelco, Vicon and Kalatel, to name a few, started their businesses by producing high quality pan, tilt, zoom cameras. They rode the wave, as PTZs defined video surveillance, and grew into leading video companies.

But the problem is that all the power and benefits PTZs deliver are only gained when a person is sitting there actively panning, tilting and zooming. With only a few percent of cameras being actively monitored, and only a few percent of those being watched at one time, there is no one at the helm more than 99% of the time.

The other disadvantage of PTZs is that the moment you’ve zoomed in to see a license plate or to get a close-up of a person’s face, you lose the ability to see everything else. You can easily miss something more important.

If you want to look across a site, PTZs are valuable tools. But if they are sitting there, not being actively driven most of the time, they are expensive cameras. That’s why many PTZs are set for continuous tours, where they move from one preset location to another, auto-panning. This also partly overcomes the problem with missing things when it is zoomed into one area. They move from one point to another to cover a wider area. This allows one expensive camera to cover the whole area. That’s the hope, anyway.

However, belts, gears and motors don’t last long when the PTZ is set for continuous tours. The best quality models need replacement parts every year, when set for auto-panning. Cheaper model PTZs wear out even faster. When that one expensive camera fails at the site, you’ve now got nothing.

Megapixel cameras have taken a big chunk out of the once powerful PTZ. No motors, belts or gears to wear out. Even when you zoom in, you can still keep on recording the whole scene, so you won’t miss anything. This is ideal for recorded video. Megapixel cameras don’t give you anywhere near the full zoom capability of PTZs. You are limited to about 3X-8X for a megapixel zoom, not the 20x-30x you get with a PTZ, but in many cases that’s fine.  And, you can buy two megapixel cameras for the price of one PTZ.

Video analytics are also changing the world of PTZs. The applications where PTZs are most important – were live monitoring is needed – that’s exactly where video analytics provide the biggest bang. They enable security personnel to monitor far more cameras much more effectively, by proactively popping up cameras when the analytics see potential threats.

Why not use video analytics with PTZ cameras? You can, and we sell such systems all the time. But if you want to be sure you are going to catch an intruder, you can’t have that PTZ panning all over the place. Fixed cameras are the only way to assure you never miss a threat.

For example, let’s say you have a site where you want one PTZ camera to auto-tour across four different preset locations. The PTZ would move to one preset, watch for some time and then move on to the next. How much time will the camera spend watching any one area? Less than 25% of the time! That means more than 75% of the time you have no visibility on what is happening. With more than four presets, it is even worse. That’s a huge blind spot in your protection!

The alternative: You can put up 3-4 cameras with video analytics built-in for about the same cost as a PTZ camera with analytics, and you won’t have yearly replacement costs for the motors, belts and gears. Most importantly, the analytics won’t miss what is happening – so you get much better security. In fact, it is the only way to go, when you need surefire protection.

In other words, fixed, non-moving cameras with analytics now give the best site awareness. With active video analysis doing the watching for you, it is better to have more fixed smart cameras than a panning PTZ.

So, the new king of the hill is a megapixel camera with analytics built-in. This gives you zoom, a wide area of coverage that is never missed, and analytics to detect potential threats so that you know which camera to be looking at.

Even better, add a PTZ camera at the site if you want the ability to zoom in for a close-up. That’s where the PTZ excels. It’s a tool for extreme close-ups. This combination system has persistent protection because the megapixel cameras continue watching the areas they are trying to protect. They never look the wrong direction. Plus, the analytics watch for threats continuously, even when no one is watching the monitors.

The PTZ will be around for a long time and still plays an important role in surveillance. However, it is not quite as important as it once was. It once was king. Now it is a good soldier. The world has changed for PTZs.

 

April 27, 2010

When is Storage not Storage?

Posted by Doug Marman under Cameras, Edge Storage, Video analytics, Video storage, VideoIQ
Leave a Comment 

The latest buzz growing in the video surveillance world is storage in the camera. But storage, in this case, doesn’t always mean storage.

IMS Research placed storage in the camera at #2 of the biggest trends for 2010:

http://www.imsresearch.com/press_release_details.html&press_id=1224

Here’s a link on a recent review at the latest industry trade show in Las Vegas, last month:

http://www.experteditorial.net/securitysquared/2010/04/when-inteligence-moves-to-the-edge-1.html

As you can see, there is a lot of talk about storing video in the camera as a growing trend, and many of the leading companies are getting on the bandwagon. But what isn’t being said is that in almost all these cases, what they are talking about is not the ability to store enough video in your cameras to replace the DVR or NVR in your system.

What they are talking about is simply buffering the video storage. That means storing just a little bit of video, such as hours worth or sometimes a few days worth of video.

Anyone playing video from the Internet uses video buffering. That’s why it says “loading”, since it is creating a buffer in your PC, so that it can play the video smoothly without interruptions. That’s a buffer, not long term storage. You can’t go back the next day to find the video file on your PC, because it is erased. This is the difference between storage and buffering.

There are two main benefits for buffering video in surveillance: First, if your network goes down for a few minutes or a few hours, you don’t lose the video from your cameras. As long as you don’t wait too long, you can go back to that buffered video to see what happened during the network downtime.

Second, when transmitting video through networks, there can be short interruptions or delays. This is especially notorious with wireless networks and the Internet, which is the same reason video streams are buffered from the Internet. Buffering the video helps a lot in overcoming these kinds of issues.

But this is completely different from true storage in the camera, where you can get months or even up to a year’s worth of video storage right in the camera.

I hear strong interest for storage in the camera, since it eliminates the need for an external DVR or NVR, and it also reduces 99% of the bandwidth used by IP cameras, since most of the bandwidth is used while streaming video across the network for storage. Only about 1% of security video is actually used, so that means 99% of the bandwidth is wasted.

There are also in many cases significant storage cost savings, when it is built into the camera.

To get this kind of storage, we at VideoIQ include a full hard drive in the camera. That gives enough for months worth of continuous video storage. In fact, we now have models with up to a year’s worth of storage.

However, I rarely hear much excitement over buffering video. Most integrators realize that losing video during network down-times is a problem, but few want to pay the extra money for this, since it rarely occurs.

So, it looks to me as if most of these companies are jumping onto a bandwagon without realizing what they are jumping onto.

There is one application where cameras with a little bit of recording can make use of the storage in the camera: For applications where the camera is not recording continuously, but only with motion detection or some other kind of detected event. If events are rare, you can get by with a little bit of storage.

But the problem is that motion detection rarely works outdoors, since it cannot distinguish all the pixel changes occurring from changes in the light, trees blowing in the wind, etc., that are happening all the time. So, it will end up recording most of the time, which means you need a lot of storage. That’s why video analytics are needed – they can recognize what a person looks like, for example, and ignore all the kinds of pixel changes and movement in the scene.

One last thing worth mentioning on this subject that is rarely mentioned: Most of these companies selling cameras with what they call storage have no easy way of getting the video off the cameras. In some cases, if you want to retrieve the video, you need to go around to all of the cameras and manually collect the SD cards. Who would ever use such a system?

Frankly, I don’t think anyone is using such systems, but they are being advertised as storage in the camera.

There are a few companies who have software that enables you to access that buffered video, but there hasn’t been support for this with the large Video Management System companies (VMS). So, in large camera systems, where a VMS is usually a requirement, you are up a creek without a paddle, since there is no way to use that software to access the stored video.

Fortunately, this is changing rapidly. Genetec just announced a new update to their software that includes edge storage support. Milestone is going to be releasing something soon. OnSSI is also close.

One company not mentioned in the above video, but who is clearly way ahead of all the other VMS companies in this category is IP Vision Software. They’ve designed a system that was designed from top to bottom for distributed storage. So, they have full integration of storage in the camera capability. They are definitely worth checking out.

So, when is storage in a camera not storage?

First, when it is only buffer storage, for a few hours or days worth of video. It may help cover network failures, but it won’t replace the need for an NVR if you want continuous recording.

Second, when you have no easy way of accessing the video. It should be just as easy to playback video in the camera as it is from your NVR or DVR. It should be no different.

 

April 2, 2010

True Megapixel Analytics Have Arrived

Posted by Doug Marman under Cameras, Edge Storage, Megapixel, Video analytics, Video storage, VideoIQ
[15] Comments 

We put on the first demonstration of true megapixel analytics in the industry at our ISC West booth last week.

It was eye-catching. Lots of people stood there staring at the analytics detecting people, cars, trucks, motorcycles, sailboats, speedboats, etc. Here’s a picture:

Unfortunately, this blog can’t show the full resolution or video, which you really need to see to appreciate how incredible it looks.

When I say this is the first public display of “true megapixel analytics” I mean the resolution being analyzed is megapixel. There have been cameras with megapixel video that have had analytics processing before. CoVi is a good example, may they rest in peace. They sold a 1 MP camera that ran ObjectVideo analytics. However, the resolution of the analytics was only CIF (320 x 240 pixels), which gave hardly any detection range. It was silly to put CIF analytics on a megapixel camera.

Why hasn’t anyone ever demonstrated megapixel analytics before? Because of the sheer processing power that other technologies need to do this.

VideoIQ’s technology is different. We need about 1/8th the amount of processing power compared to other high quality analytics systems. So, we can run the whole thing in one of the popular low cost DSP processors. But all other analytics technologies need a lot more horsepower.

For example, ObjectVideo on their web site claims they can run up to 4CIF resolution video in the same DSP chip we are using. However, in most cases the users of OV onboard are only running CIF resolution, because there are serious limitations running 4CIF, such as only being able to have one rule running at a time and a limited number of objects that can be detected.

IOimage uses two DSP processors in their cameras to get high quality and avoid compromising detection.

The camera we demonstrated was a 1080p camera, which is 1920 x 1080 pixels. We demonstrated it live at the show, with the analytics all running in the camera. It provides 3X the horizontal coverage of a standard resolution camera, and more than 2X the anaytics detection distance.

For other technologies to run 1080p analytics, they would need more than 6 times as much processing power, compared to 4CIF video. That would mean 6 DSP chips, or some very expensive high end DSP chips.

If you try to run this on a server or a PC, you would need a full dual core processor to run one camera. So, you can see why it’s never been shown before. It is impractical for other technologies.

The other industry first we showed is something we call IQTrack. It uses the video analytics to automatically track and zoom on objects in the field of view. Here’s a picture:

This is different from PTZ camera tracking. If you look at the lower left of the picture, you will see that the whole field of view is still being recorded and it shows where in the scene you are zoomed into. So, you can always go back later and pick another part of the video to look into.

The other unique thing is, if many people are in the area, you can click on one person and it will zoom in on and track just that one person. That’s never been shown before either.

Watching it, you can immediately see that there is no comparison between watching video that is automatically zooming and tracking on important objects, versus static video cameras. It pulls your eyes to exactly what is important. I think this is going to be very popular for megapixel cameras.

The 1080p cameras we sold will also be the first cameras to ship with a new imager from Sony that has some amazing low light performance. We are still testing it, but it looks to be 2X-4X better than any other multi-megapixel imagers used in the security industry.

And of course, the camera we showed included a hard drive so that you can store 1-2 months of high quality 1080p video. This solves the bandwidth problem for megapixel cameras, since it needs no bandwidth to record, and eliminates the need for external storage in most cases.

Now that true megapixel analytics have arrived, I think it is going to set the standard, and I think it offers incredible visual value to megapixel cameras, even if you don’t want the analytics for detecting alarms.

 

March 5, 2010

The Emperor’s New Body Scanner

Posted by Doug Marman under VideoIQ
[4] Comments 

I guess we will all be wearing the Emperor’s New Clothes with the new millimeter wave scanners that the TSA is planning to introduce into airports.

What screeners will see is similar to people having no clothes at all. Of course it looks like they have no hair either, and it gives everything a weird metallic look.

It is not an erotic picture at all, but as you have probably heard, there is a huge reaction over privacy concerns.

The TSA is trying to minimize this problem by separating the screeners from the security lines, so they aren’t near the people who are being denuded.

However, this is a clear case where video analytics can make this privacy issue disappear – just as well as your clothes disappear with this scanner.

As I mentioned in my previous post on Big Brother versus Little Brother, analytics have the potential to improve privacy with video surveillance:

http://spotonsecurity.com/2009/02/13/big-brother-versus-little-brother/

In the case of body scanners: There is no need for screeners to see the whole body. The only thing they need to look at is the area where a gun or potential bomb is located. This can be displayed on a cartoon image of the body, to show them where it is located.

It turns out that most metal objects and bomb materials are easily distinguishable by their brightness or darkness on the scene. This is why the millimeter wave technology works so well. You can quickly spot the foreign objects.

Well, if a person’s eyes can quickly detect the differences, so can video analytics.

I’m surprised that the TSA is pushing this new body scanner without such a capability. I know they are aware of the benefits with adding analytics, because I brought the idea up for them about 5 years ago, when they first started getting interested in millimeter wave. They immediately saw the benefits of using automatic object detection as a way of protecting privacy.

Perhaps they felt there was an immediate threat that couldn’t wait. I don’t know, but they are asking millions of people to give up their privacy for something that may not be necessary.

I’ve read a lot of comments where people say they would choose being scanned over a physical search, because it is much less invasive. And, based on that, people say they will go along with their loss of privacy.

But what if they knew that wasn’t the real choice they had to make. What if their body parts did not have to be made visible unless they had something suspicious hidden in or under their clothes – and then it would only show that area of their body? Now, what would people choose? Would they still think giving up their privacy was worth it?

I don’t think many would think being scanned and viewed naked would seem so great any more. Of course, there will always be a few who enjoy the idea, I suppose.

The good news is that I understand L-3 may be introducing a system that uses video analytics exactly as I’ve described. I’ve heard they are testing it at the Amsterdam airport, where it highlights questionable objects on a “gumby like” figure.

I wonder if the TSA should try forcing the body scanner technology on everyone until a solution with real privacy protection is ready?

One thing is clear: Video analytics does, in this case, offer a chance to put real clothes back on the Emperor and all the rest of us as well.

This is a good example of the power of analytics to improve privacy. It also seems like a no-brainer to me as the right thing to do.

 

February 23, 2010

The Technology Acceptance Curve

Posted by Doug Marman under VideoIQ
[6] Comments 

I’ve been studying new technologies in the security industry for over twenty years. When estimating how successful a new product or company will be, there are many factors. One of the biggest is how easy is it for new users to accept the new solution.

This issue recently came up in John Honovich’s recent post:

http://ipvideomarket.info/report/objectvideo_poor_reputation_and_blaming_others

John takes ObjectVideo to task for blaming others for the slower adoption of their video analytics.

Brian Baker at ObjectVideo had written:

“Many in our industry correlate the need for trained users and the need to configure the analytics with the notion that analytics, as a whole, are immature and unreliable. Nothing could be further from the truth.”

Brian took the position that highly complex stuff requires training, and this is the same for SAP and Oracle systems. That doesn’t mean it is immature. It simply requires expertise.

Well, this is a perfect example of the the technology acceptance curve. If you have something different that requires new levels of training and new skill sets to be taught, you should expect sales of that product to grow slowly. There is no way it can take off rapidly or be adopted quickly. Why? Because you have to first build a whole new cadre of skilled workers before the system can be deployed.

In the late 1990′s when DVRs were still PC based software products, their sales grew incredibly slowly. They had been around for years, but so few integrators used them that many wondered if they would ever succeed.

The company I worked for then (Interlogix) and a few others in the industry took a different approach. We designed DVRs that were embedded appliances and worked exactly like the VCRs that everyone had been using for years. The sales of DVRs suddenly boomed.

The DVR business went from maybe $50M worldwide, to over $500M in a 2-3 years, and doubled again shortly after. That’s how quickly it changed once the technology acceptance curve was overcome.

The problem with the first PC based DVRs is that very few of the video integrators knew how to use them or how to fix them when they broke. It was a whole new world of computer software, and without extensive training they were afraid to adopt them. It was simply too much of a time investment, and too many opportunities to get caught in a hornet’s nest because you didn’t know what you didn’t know. So, most installers avoided DVRs unless they absolutely had no other choice.

Once they could pull out a VCR and replace it with a DVR that worked almost exactly the same – that’s when DVRs started selling like hotcakes.

A lot of people confuse this with “ease of use”, which is also important but is different. The technology acceptance curve is more about how familiar something is and how intuitive it is to use based on what people know and have used before.

This why I’ve long said that widespread use of video analytics would not begin until it was as easy to install as traditional video motion detection, which everyone is familiar with. Then anyone who knows how to install a camera would know how to install video analytics.

That’s what makes the adaptive analytics that we use at VideoIQ so significant. It is the only technology that gets rid of all that need for camera calibration and tuning required by other technologies. This means that anyone who knows how to use an IP camera can use our iCVR.

In fact, we’ve had quite a few installers who simply took our iCVR out of its box, plugged it in, and it was working. We rarely need to send someone out to help anyone understand how to use our analytics. Our number one tech calls are over IP related issues: How do you set up port forwarding on a router, or how to track down network problems…

It is still a good idea to understand video analytics, so you know where best to use it and what to avoid. So, we offer on-line training. But the skill set of knowing how to tune and calibrate a camera are not needed.

Sam Pfeifle, from Security Systems News, and I will be talking about this and other key technology differences that matter when using video analytics, along with the most popular uses for it – outdoor protection – in a webinar this coming Thursday. You can watch live, or you can watch it later if you can’t make the time:

https://presentations.inxpo.com/Shows/UnitedPublications/02_10/Registration/UnitedPublications_Feb_Registration_Page.html

The need for training can seem like a small thing, but this ends up being one of the biggest factors in how quickly a new technology spreads. That’s why the most important advances are often the ones that seem almost invisible, because people gain all the new benefits without having to change their lives.

 

February 5, 2010

The Mystique of BRS Labs

Posted by Doug Marman under VideoIQ
[11] Comments 

Mystique can be a great advantage in lots of professions. For example, movie stars seem more exotic and interesting the less we know about them. The attraction of traveling to distant countries and strange cultures is also boosted by the desire to see something obscure.

But I’m not sure mystique is a good thing when it comes to security systems, and that’s one of the problems I see with BRS Labs. They have generated market interest by their amazing claims, but there is very little known about their technology. That may be their biggest advantage – at least for creating marketing buzz – but it raises quite a few concerns when it comes to actually using their systems for security.

Let’s take their central claim that they can detect out of the ordinary behaviors without any rules being set up or configured. It automatically observes the area, learning what is typical. Then it alerts you when something happens that is abnormal.

It’s a truly fascinating marketing concept. Who wouldn’t want a system that you could just plug in and was smart enough to know when a potential threat was occurring – even warning you about things you would not have thought about before. At first it seems like the perfect answer to security.

However, the more I think about it, the more concerned I get. The problem is that we don’t know what it is going to detect or why, and we don’t know what it might miss that could be important. In other words, it is mysterious about how it works.

How do you know how effective the system will be unless you know what it is detecting and how it works? How do you know it is going to catch real threats that matter to you, if you don’t know its enigmatic methods of detection?

Being curious, I decided to do some research. I tracked down the first patents that BRS Labs filed to get a better idea of what was under the hood. And just like in the Wizard of Oz, once I pulled back the curtains and understood what they were doing, it lost a lot of its mystique.

The most important thing I learned: Their system isn’t smart enough to work without rules. Their system requires rules just like all analytics systems. The big difference is that they simply aren’t telling you what those rules are that the system is using. It can’t detect anything out of the ordinary, it can only detect the types of things they program it to look for. How can you ever judge how well the system is going to work unless they tell you what those rules are?

From what folks at BRS have said, their system watches where objects enter the field of view and where they go in the scene, including their direction of travel. I believe they also detect where objects stop and about how fast they move. They probably distinguish people from vehicles and seem to be able to filter out ordinary background movement. This is actually all stuff we do as well, as do many other analytics systems.

What is different, is that they monitor these specific activities over time, and if some pattern of actions happens that is different from previous activities, they consider that a potential threat. In other words, if a car parks in an area where the system hasn’t seen a parked car before, it generates an alert. It will do this whether you care about that or not. If you get the alarm and don’t care about it, then you can tell the system to stop sending alarms like that.

But here’s the problem: when you are telling it you don’t care about that kind of alarm, you can never be sure what you are saying you don’t care about. You might think you are telling the system that you don’t care about someone parking in that spot, but in fact it might have alarmed because it was a truck and it had never seen a truck in the scene before, or the truck might have taken a different path than usual. You are telling the system to stop sending those alarms, but you don’t even know what it is you are turning off – because you don’t know the rules it was using in the first place. So, you might be making the system worse.

You might think, then, that you should not tell it to stop sending alarms, but the system needs you to, because when the system starts up, it generates large quantities of alarms, because lots of things look abnormal at first. The false alarm rate would never be manageable if you didn’t teach it what was not important. This process takes weeks of training, from what I’ve heard.

If you were trying to protect a high risk facility, such as a nuclear power plant or a place were dangerous chemicals were stored, how secure would you feel if you never knew what your system was detecting and what it was ignoring? And how secure you would feel if you told the system something was unimportant, but you didn’t know exactly what you were turning off?

If I want to detect someone in an area that is off-limits at night, with a rules based system you define the detection you want, and you can easily measure if it is missing real threats or sending you false alarms. But how do you measure or judge the accuracy of a system when you have no idea what it is detecting or avoiding? I don’t think this is a place for mystique. I think this is a place where we need to know what the system is doing. Otherwise, how can we ever know if it was going to provide the kind of protection we need?

That’s the first big problem that concerns me about this idea. However, as I thought about it, another problem became clear as well. This is something that is important in security: How easily could someone defeat your system? In this case, it becomes clear that if you do something repeatedly, the system is going to start ignoring it, because it is no longer abnormal. So, if someone wants to defeat the system, they just need to do something over and over again, and they can be sure the system will stop alerting on that behavior.

For example, you might want to be warned whenever a car enters a parking lot at night. Well, if a smart terrorist or criminal knew you had one of those mysterious behavioral detection systems, they would simply make a habit of driving into the parking lot and turning around and driving out. The first few times it would generate an alarm and anyone looking at the video would probably think a person just came in by mistake and left. But even if you wanted to keep an eye on such behaviors because it could potentially be a problem, you would not be able to, because the system would start ignoring it once it happened often enough.

Hopefully BRS has a way for the user to tell the system that although an alert was not important, that it still wants to keep seeing them – and not to start ignoring those kinds of things. But the problem is that you don’t know exactly why it generated the alarm, and so you don’t know what it is you are asking for more of, or saying you don’t want to see any more.

The BRS Labs  systems would be a lot more useful if they told everyone exactly what their rules are for detection. However, this would probably rob them of the great mystique their system has, which has certainly created a lot of good marketing for them.

Mystery is a great attractor. But when it comes to security, I think we need to know how a product works before designing it into a system. Spice is nice, but it doesn’t make a good main course.

 

January 8, 2010

Comparison Testing is Hard to Find

Posted by Doug Marman under VideoIQ
[6] Comments 

It is not easy finding public information about how video analytics products actually compare. We have been through side by side tests with quite a few companies, but they have all turned down our requests to publish the results for a variety of reasons. That’s one of the challenges of the security industry. There are often good reasons not to make what you are using public.

We were happy to see John Honovich step up to provide some open testing. As soon as he asked if we were interested, we sent him our iCVR. He ran a test and published it last year:

http://ipvideomarket.info/report/testing_videoiq_video_analytics_icvr

Unfortunately, from what we’ve heard, no other video analytics companies have stepped up and agreed to John’s requests. It’s hard to compare a technology when there is only one.

We were invited to join in on a side-by-side comparison at ISC East last October. Three video analytics companies were invited and agreed to run a test: IOimage, BRS Labs and ourselves (VideoIQ).

It was sponsored by Government Security News and about 100 people attended.

At the last minute, BRS Labs cancelled. I never heard why. So, it ended up only being IOimage and ourselves.

I will try to be objective about the results, but of course that’s not as easy as it sounds.

The demos weren’t long. We each had about 10 minutes to show our products working. Not exactly an in-depth test. Both systems had to detect someone walking from behind the curtains across the room. With our system they also tested “Object Missing” and with IOimage they tested “Object Left Behind”.

Both products worked fairly well. Some of the differences were minor but interesting.

IOimage needed to calibrate their system before the event. They used two people and it took them about 30 minutes. However, since it was indoors, they could cut their process short. In general it followed the videos that they show on how to calibrate their products, but these guys were clearly experienced and didn’t have much time, so they moved fast. One person sat at the PC and marked the head and feet of the other person at four different places in the room. They drew a line on the floor, showing a predetermined distance (10′ in this case). They only calibrated one axis, not two. And they only calibrated in the one area that was going to be tested, not the whole room.

Our system doesn’t need calibration, as I’ve mentioned before. So, we had plenty of time. Picture the Maytag repair guy with his feet up, while waiting. <G>

However, what we found interesting was the way IOimage positioned the whole process of calibration and tuning. They claimed calibration is what makes detection better. This is why they could detect someone crawling. However, as I’ve said before, this is wrong. While calibration clearly makes their system work better, in our case the calibration runs automatically. So, it is really a question of manual calibration compared to automatic calibration. As I’ve pointed out, there are lots of big disadvantages to manual calibration:

http://spotonsecurity.com/2009/11/05/the-tuning-and-calibration-controversy/

The issue about detecting crawling people is bogus. Perhaps they don’t realize it. They might think that calibration has something to do with this, but it doesn’t. It is simply that we have not yet developed an object classification type for people crawling. Quite frankly, we haven’t had any demand for it yet. The fact that we can still detect people crawling is easily demonstrated by setting our detection to “suspicious objects”, which easily captures people who crawl.

On the other hand, IOimage worked hard to make sure that they were getting a full view of the people in the area they wanted to detect, and they were especially concerned about moving chairs or tables out of the way. We didn’t have to worry about that, because we have an object classification type for head-and-shoulders detection of people. That’s pretty important indoors, because chairs and desks are quite common. IOimage apparently doesn’t have that object classification type, which is why they needed to see the whole person. In other words, this has nothing to do with calibration; it is all about the types of objects the system has been trained to detect.

You can read my previous post about the down side of manual calibration. But here are some other questions: How do you calibrate boats? Does someone have to walk out on the water? If  you are setting up a system to watch the tarmac at an airport, do you have to shut the runway down while you walk out there to calibrate it? What about hazmat sites? Do you really need to send people in there to calibrate the camera? Besides all the other problems with having to manually calibrate a system, as I mentioned in my previous post, it isn’t always practical.

The real test comes down to how well do the systems work. That’s what really matters. Unfortunately, this wasn’t a stringent test. We wished it could have been outdoors with trees blowing in the background, to make it a tougher comparison. In general, both systems seemed to work well. We spotted some minor false detects on their system when the curtains moved, high up in the air where calibration should have ruled detection out, but for some reason didn’t.

Another noticeable difference was that they were only streaming the video from their camera, while our camera captures and records the event. Therefore, it appears to be more difficult with their system if you missed what just happened. In our system, you get a video clip that you can playback whenever you are ready. You don’t have to be there watching. IOimage would have to add an NVR to their system to get storage playback.

What IOimage added instead, that we don’t have, is mouse trails of where a person has been in the scene. This helps when it isn’t as easy to see where someone has been. You can just look at their previous trails. That works. I would rather watch the actual pre-alarm video so you can see what they were actually doing, but both systems offer something that works.

Apparently, from what we could see, IOimage can adjust sensitivity, but it seems to change it for the whole camera. We can individually set sensitivity for each detection rule. So, our missing object can be very sensitive, if we want, but people detection does not need to be.

One of the best things about their product is their web interface. Very well done. An integrator can do all of their calibration through the web, and it worked smoothly. If you have to calibrate, that’s a big plus.

However, they had to switch their camera from people detection to object left behind during the demo. Apparently it doesn’t detect both at the same time. Ours does, and in fact can run quite a few different types of detections with different types of objects, in different regions of interest, all at the same time on the same camera.

Also, once our rules are set up, our system keeps detecting even when you are changing or adding new rules. We were surprised to see that the IOimage system shuts down when you are setting up rules. It’s not a big issue. It is more important to detect a number of different things at the same time. That’s very useful.

It wasn’t much of a detailed competitive test, as I said. Both systems use high quality video analytics, not advanced video motion detection. And the results from both were pretty good.

And of course, this would be more objective if someone else were reporting it. However, it is so hard to find any comparison testing that I thought I would share it, anyway.

The CEO of BRS Labs, Ray Davis, has apparently refused to send their equipment to John Honovich for testing:

http://ipvideomarket.info/report/brs_labs_layoffs_and_skepticism_mount

I hope we aren’t the only ones who are pushing for more open comparisons. It helps everyone to see how technologies stack up.

I encourage our competitors to step up and send their products to John. Or let’s find some other approach to public testing.

 

November 5, 2009

The Tuning and Calibration Controversy

Posted by Doug Marman under VideoIQ | Tags: |
[2] Comments 

When we first introduced our iCVR, I heard from a number of other video analytics companies. They offered their compliments. Many of them said that when a new market is gaining recognition, all successful products help bring growth to the whole field, so it was good for everyone. I was glad to hear them say that, because I feel the same way. It is nice to work in a field where there is open communication and friendly compliments, even though we are competitors.

But the one area that they thought we’d gone too far with was our marketing message that our product needed no tuning or calibration.

I said, “Well, it’s true.”

They looked at me  in a way that it made it clear they didn’t believe it was even remotely possible. They were thinking this was going to be another case of a company over-exaggerating their claims. That’s one of the worst things for new technologies, since it hurts credibility and people start doubting any of it works.

I reassured them that our claims were true, no matter how hard it was to believe. In fact, I said, even if someone wanted to, there is no way for an installer to tune or calibrate our iCVRs. That’s when they realized we really meant what we said.

For some people, our claims sound too good to be true. In fact, only a few months ago, Itsik Kattan, CEO of Agent-VI, said in an article: “Beware of ‘low-touch’ and ‘no-touch’ systems and vendor claims that their analytics self-learn. There’s no such thing.”

http://www.experteditorial.net/securitysquared/2009/08/video-analytics-business-intelligence-security.html?p=5

Lately, the language we’ve been hearing has created even more confusion. Some sales people are saying that calibration and tuning are required to get high accuracy, and any system that doesn’t use calibration and tuning are on the order of traditional video motion detection.

We’ve also heard from some who promote the idea that the more tuning knobs the better, and some products have more than 50 different adjustments that can be made. They feel they have the best accuracy as a result.

There are some serious flaws with these comments. However, the real problem here is that this kind of talk focuses on the wrong thing.

It isn’t about whether tuning and calibration can help make a system more accurate. We agree that it does. It’s hard to argue with that. The real question is which is better: Automatic self-tuning and calibrating systems, or manual systems?

We’ve got plenty of tuning going on in our iCVR. The moment it starts up it learns the environment, adjusts the filters is it using, and continues to get smarter over time. As the environment changes, it re-tunes and continues to adapt.

The calibration process also happens automatically.  The iCVR watches for people and vehicles and boats, and when it sees them it automatically calibrates what the proper height for these objects should be. This does indeed reduce false alarms, although it doesn’t help our systems as much as it helps other technologies.

The reason we can do the calibration automatically is because our accuracy before calibration is far better than any other product. If accuracy is poor to begin with, then you have to use manual calibration, because the system doesn’t know what objects are people and which objects are false alarms, like tree branches or birds or bushes blowing in the wind.

In fact, the worse a technology is to begin with, the more important calibration and tuning is.  Take advanced video motion detection (AVMD) systems that don’t even know what a human looks like: They are far more dependent on calibration and tuning than true video analytics systems. Why? Because AVMD doesn’t have any way of knowing it is a person except by the size of the object. Manual calibration is the only way of teaching the system what a human should look like.

Of course, AVMD even with calibration and tuning doesn’t even come close to true video analytics systems, as I’ve written about earlier on this blog.

But let’s get to the big issues that make this issue so important. Here are the advantages of a self-tuning and self-calibrating system:

  1. It saves hours of set-up time per camera. We’ve heard the average is between 3-8 hours per camera. That’s significant.
  2. You don’t need specially trained technicians to install it. Anyone who knows how to install a camera can do it. The more tuning knobs a system has, the more important it is to train every installer to get the best results. No tuning knobs means that anyone can install it.
  3. You don’t need to re-tune or re-calibrate every time a camera is moved. The system adapts automatically (note, if you set up detection for a specific region of interest, you may need to change the region of interest, which is why our system automatically tells you if a camera has been moved). For manual systems, if you don’t re-tune and re-calibrate, you are assured that the system is set up wrong after the camera is moved.
  4. If the environment changes, the system automatically re-tunes itself. When the leaves fall off the trees or when snow appears on the ground, you can be assured that re-tuning is needed to improve performance. But with manual systems, unless you manually re-tune it when the seasons change, you will have degraded accuracy and more false alarms.
  5. Self-tuning and self-calibrating systems are always at their best. Manual systems are at their best the minute after you tune it and calibrate it. After that, performance deteriorates. With our iCVR, performance actually gets better over time.
  6. Lastly, we are continuing to improve the performance of our analytics. Every time we do, we provide a downloadable update that can be sent to all the iCVRs in the field, thus improving their accuracy. But with a manual system, how do you send new algorithms and use the old tuning settings? With new algorithms, you need new tuning adjustments. So, it is a much more time consuming process to update to the latest improvements.

When you add all of these issues up, I don’t know how you argue that manual adjustments are better.

But the whole thing still comes down to how accurate the system is. Having an automatic technology that has poor detection and high false alarms is not going to be an improvement.

So, where does our iCVR stand?

When compared against the best of the best video analytics technologies, even if you bring in an engineer from their factory to do the tuning and calibration so you know it is absolutely adjusted the best it can be, from what we’ve seen and heard, we are about on par and maybe slightly better. In some cases we’ve seen that we beat them, but we’ve seen a few cases were they edge us out by a hair.

But that is the moment after tuning and calibration. Our system then continues to get smarter and more accurate, while all the manual tuning and calibration systems get worse. So, over time, we have the clear advantage in terms of accuracy.

We’ve had plenty of people test our stuff and have confirmed this, but it is always best to try this out for yourself and come to your own conclusion.

However, I think the rest of the benefits make the point pretty clear. The cost to install and maintain, and the ease of use are all in favor of systems that are smart enough to calibrate and tune themselves.

In other words, there really isn’t any controversy here.

 

October 20, 2009

Why Baggage-Left-Behind has Been Such a Failure

Posted by Doug Marman under VideoIQ
[10] Comments 

It continues to amaze me that people list Baggage-Left-Behind detection as one of the shining examples of the power of video analytics. In fact, it has to be the biggest disappointment in the market.

However, the reason it has failed so often is not a reflection on bad technology, but the dangers of complex systems. It is a good lesson in knowing where and when to use new technologies.

First, it is important to say that there are hundreds of integrators now using video analytics with very positive results. They are going after the sweet spot, where analytics produce great returns on investment and solve real problems that would be difficult to solve any other way: detection of people where they aren’t supposed to be, perimeter protection, security for outdoor assets, remote guarding, storage and bandwidth savings, etc.

But we have heard so many stories of failed projects as a result of trying to make baggage-left-behind work that we have intentionally left it out of our product, and we regularly warn integrators of the dangers. I thought it would be worth sharing a little more information about it, to help others from falling into this trap.

If you want to see cases of proven failures, you can check out the lawsuit between Lockheed Martin and NYC:

http://www.nytimes.com/2009/04/29/nyregion/29mta.html?_r=1

Their problem wasn’t just about baggage-left-behind, but that was one of the applications that they simply couldn’t get to work. This doesn’t mean the technology couldn’t detect bags left behind. It means that even though it could, the system was unusable.

IBM made a study of baggage-left-behind years ago. I don’t see a public posting of the study that I can link to, but they went back over the last decade to study terrorist bombings. The question they asked was how much benefit would baggage-left-behind have delivered if it were installed at the sites that were bombed.

Their conclusion: Baggage-left-behind would not have helped a single one of the cases. No benefit at all. It wouldn’t have helped even prevent a single bombing or saved any lives.

We also had a recent discussion with someone from the I-LIDS group in the UK, which is a UK Home Office sponsored testing group for video analytics. They set benchmarks for a variety of video analytics requirements. Many have been effectively deployed. However, when it comes to baggage-left-behind, they told us that no vendor has come even close to meeting the needs of supplying the functionality needed to make the technology work as it would need to work to be successful.

The problem isn’t detecting a bag being set down or left behind. That’s the easy part. It is the complexity of the whole system and what is needed to make it practical.

Think about how often you might get an alarm if you detected any bag put down in a crowded airport. How often do people let go of their bags for a moment? How often do they walk away from their bags to throw a newspaper away in the trashcan, or leave their bags with a family member while going to the restroom? How often do people leave a newspaper behind or a lunch bag?

So, the first problem is that the system produces a ton of false alarms. Who wants to continue responding when false alarms are continuous and far outnumber real warnings?

But this is only one part of the problem.

The second part of the issue comes down to how often is a terrorist bomber going to leave their bomb out in the open in the middle of the floor? Even if they were crazy enough to do this, how long would terrorists continue leaving bombs out in the open once they knew baggage-left-behind was deployed everywhere?

Not many. Especially when it is so easy to put a package into a trashcan, or hide it from open view.

It is true that there have been cases of people leaving bombs behind under a seat or in a corner, but the moment word gets around that someone was caught by baggage-left-behind, the technology would be useless, as terrorists would soon stop leaving bombs in any visible place.

On top of these issues, you have the problem of accurately detecting objects that are left behind when you have a crowded area. When people are continuously walking in front of a bag, how long will it take before you detect it?

If the false alarm rate were really low, I’m sure some transportation facilities would still like to use it, even if the ability to detect a bag was low. However, when you add in the low, low likelihood of it actually catching anyone who is intentionally leaving a bomb, and the lower likelihood of it catching terrorists in the future, then the whole program simply becomes impractical.

This doesn’t mean it can’t be used some places effectively. I’ve heard of a couple. They were not crowded areas, and they were not places where people waited, but where people simply walked through, so they shouldn’t be leaving anything behind. But this doesn’t deliver the great promise that people imagine when they think of baggage left behind. They think of catching terrorists planting bombs in crowded areas, such as train stations or airports or bus terminals.

It’s such a tempting use of analytics, especially when it is so easy to demonstrate. It really isn’t that difficult to detect a bag being left in the middle of an empty platform. But when you start looking at the overall complexity of the application, how difficult it is to reduce false alarms, the problems with recognizing bags in crowded areas, and how easy it is for people to defeat the system by not leaving bombs in open areas, then you start to realize that this really isn’t such a great application for video analytics.

What is such a shame is that this application has gotten so much prime time attention, while the uses of video analytics that produce huge real-world benefits get much less discussion.

I think the big lesson here is the importance of thinking the whole system through. Detecting bags is the least of the problems. There are way too many bags and it is way too easy to plant bombs in areas that can’t be seen. The system turns out to be far more complex than it seems at first.

To counter my note of caution, I’d love to hear of examples where baggage-left-behind has been used effectively, and the key requirements needed to make systems like this successful.

 

Next Page »