Video Analytics: Ready For Prime Time?

More than 40 years ago, the founder of Intel, Gordon Moore, first predicted that the number of transistors placed on an integrated circuit would double every two years — leading to regular, fairly predictable advancements in computer-related technology.

Moore’s Law has proven to be pretty accurate for processing speeds, memory capacity and the number and size of pixels in digital cameras. And it has held up well when discussing advancements in video analytics, a technology where computer-based algorithms are used to sort through video for noting people or vehicles in restricted areas, unusual behavior, object tracking, identification and counting, an abandoned bag in a train station or airport and a host of other tasks.

Nearly two years ago, my colleagues from SecurityNet and I were asked by STE to bench test 13 of the then-leading video analytic offerings on the market. What we found were largely expensive solutions — most often built into a DVR or stand-alone box — that needed frequent tweaking and were not ready for the mass market.

Since our tests in 2007, video analytic technology has changed, mostly for the better. The engineers have improved at writing programs that can distinguish between people and a stray dog or a tree blowing in the wind. Some of the offerings can now automatically tune and adjust themselves. And the cost per camera, or channel, is coming down. The technology may still not be quite ready for the mass market, but it is getting closer.

Moving to the Edge

One of the biggest changes I have seen is analytics moving from a stand-alone box to residing in the camera. That is the direction that always seemed to make the most sense. The analysis is done on the pixel level and the pixels start in the camera.

Using analytics in a box with analog cameras, end-users must transmit the video back to a central location, convert the data to digital format and then analyze it. With IP cameras, the video is transmitted to the station, uncompressed and then analyzed. Both of these scenarios take extra time and are not really efficient.

One reason analytics began its life sitting in boxes or on a DVR is the tremendous amount of computer power required to do the job of processing the images. But here is where Moore’s Law plays a role. Most current CCTV cameras are built around digital signal processors. DSP’s are also the tool of choice to compute the algorithms at the heart of all video analytics. As the power of these DSPs continue to grow exponentially, it has become possible — maybe even preferable — to add basic analytic functions into the camera.

Some of the big camera manufacturers are now building some fundamental, basic analytics into their products. Pretty soon, we will see the rest of the major camera manufacturers doing the same. They are going to be able to offer basic analytic functions for a couple of hundred dollars per channel.

The proprietary analytic box makers are generally charging anywhere from $1,000 to $3,000 per channel depending on the number and scope of the functions built into the box. There is a big difference in price and that does not bode well for the box makers. When you can lower the cost by a magnitude of this size, as the camera manufacturers have done, that is a big deal. That can change when and where people are willing to use analytics.

So, it is going to the edge. That writing has been on the wall for a while. It just takes the big guys — the camera manufacturers — a little longer to get their act together, but when they do, they will likely capture most of the market.

The IP cameras are going to have the power for analytics built into them, or at least the option to do so. This will work well for the end-user starting a new video system or expanding an existing one. Where a legacy system is involved, it may be best to replace analog cameras one at a time as they wear out with an IP model.

Thinking Inside and Outside the Box

Let’s take a look at some of the pros and cons of having the analytics in the camera or in a box. By going with the analytics in a box, DVR or NVR away from the edge, all video from all cameras has to be transmitted. That is a tremendous bandwidth hog that can tax almost any network to the breaking point. By handling the heavy lifting of the analytics at the camera, only vital information, sometimes called metadata, is put into the video pipeline.

Another company is now building cameras with multiple digital signal processors (DSPs). These cameras, intended for high-end, outdoor use, can cost about $10,000 each. But they do offer some pretty interesting features. For example, one processor is dedicated to stabilizing the image in a windy environment. Other processors look for different types of motion. This is all great stuff, but for the price, it is pretty much left to protecting government facilities, ports and international borders.

The idea of using multiple processors will become commonplace in the not-too-distant future. Manufacturers will start with a basic camera and then allow end-users to add features as required. The power of the DSPs is growing rapidly, but they are still nowhere powerful enough to provide analytics for three- to five-megapixel images.

Megapixel cameras provide a big challenge to analytic providers. Some camera manufacturers are getting around that by providing multiple video streams — high and low resolutions. An end-user can use the low-resolution stream to find targets and then view images at high resolution. That could be a major advantage.

But the sales of IP and megapixel cameras still have not gone much above 20 percent at best and most of those sales are going into new installations, not upgrades. It is going to take some time before all the legacy systems are updated. So the analytic boxes have life — at least for a few more years.

Software Only

There is another model for analytics that is catching on with some users — the emergence of software-only companies. They do not sell boxes or cameras; instead, their software is designed to work on servers and PCs using common operating systems such as Microsoft Windows. This approach requires the end-user to pay an upfront cost and then make annual software support agreement (SSA) payments.

This is a little different for an industry used to buying hardware, such as cameras and DVRs. It is not necessarily a bad model. The IT people have been using this type of agreement for years to acquire and keep their software up and running.

But security is different from IT. We are not a profit center for an organization. Security people are used to buying something and then not putting much more money into it for five to 10 years. The idea of paying for something every year requires a different mindset.

In the long range, this arrangement will probably represent a major piece of the analytics business, but security people just are not buying into it in a big way right now. It is going to be up to the software manufacturers to show that SSAs provide value.

Analytics for Operations

Another way we may see analytics adopted sooner rather than later goes outside of traditional security requirements. More organizations are seeing the value of analytics for operational purposes. Retailers may put a camera on an end-cap promotion and then use analytics to count how many people walked past the display; how may stopped and looked; how many held an item and then put it back; and how many decided to make a purchase. This helps stores to sell.

Banks are using analytics to see how many people are coming into the lobby and at what time of the day. They want to know how long the lines are getting so they can make sure to have enough tellers present to handle peak times. This is a customer service issue.

In time, you may find an organization’s human resources, legal, sales and operations departments may have as many uses for analytics as the security staff. If so, you may be able to spread the cost over several budgets, making the technology more affordable. While the security director is under pressure to cut the budget, the marketing person might have some money to help sell more products or increase customer satisfaction.

What End-Users are Looking For

In 2007, the SecurityNet testers reached a few conclusions about video analytics. One was that the quality of the camera image is very important, with high resolutions and good optics critical for intelligent analysis. That is probably still true for license plate or facial recognition applications, but there is a company I visited recently that uses very low-resolution sensors for its analytics. You cannot really see a person’s face, but the folks there said, you only need to be able to distinguish the shape of a blob of pixels as being a human being. Since megapixel cameras provide far more data than the analytics can handle, that extra resolution is going to waste — as least as far as identifying basic alarm events.

Almost two years ago, we decided that camera placement was critical, and an end-user should expect to use more cameras installed in unusual locations to optimize video analytics. That is still true.

Also, we concluded that because of lighting and other environmental factors, outdoor analytic applications were more difficult to set up than indoor ones. That is still true, too.

So, analytics are getting closer to mass-market adoption. But most security directors still have to be convinced that the technology does more than wake up his guards to look at the monitor. They want to know if analytics will enable them to employ fewer guards. Will they be able to do things without guards that they could not do before?

Especially in the current economic situation, budgets do not have the extra money for special software, proprietary boxes or high-tech cameras.

But as the prices for analytic solutions continue to drop, it will be possible to make a case for a real return on investment. Then analytics will become a must-have.

In any case, let’s see what Moore’s Law and another 18 to 24 months bring us.

Jim Coleman is president of Operational Security Systems of Atlanta. Coleman is also a past president of SecurityNet, a 24-member group of independent system integrators dedicated to providing a single source of electronic security for government and corporate installations.