printer friendly version

Video analytics: the definitive 5-point guide

Video analytics will not be able to tell you who is a shoplifter and who is not. It will never see around corners, but it is proving to be an essential element to CCTV for preventing crime. In my mind video analytics is the absolute key to video surveillance, it gives every camera a real job to do and that is powerful.

However, there is so much confusion and gobbledygook in the market place that I can understand people’s caution. As a result, video analytics has a poor image to live down. Previously providers overpromised and under delivered; I speak to many customers who just do not want to go there, or are so disappointed with what they have that they have completely abandoned it. Is it not sad that this technology has become so controversial when it can actually be so great? The intention of this article is to help you be more informed about analytics before making any choices with regard to using or implementing it.

A video analytics system is complicated and complex, but that does not mean it should be for the operator or the installer. As a general rule, the simpler the analytics software is to use, the more complex the work that goes on behind the tin. Of the many companies around the world that loosely claim that they can do proper analytics, I can honestly say that there are probably less than seven and of those, only a handful are available and supported in South Africa and Africa.

First, let us define it: 'Video analytics is a type of computer vision that analyses the behaviour of objects in the scene being captured. Video analytics is widely used in security applications in which a video camera is aimed at a scene where detection is desired. The analytics will pinpoint an intrusion by evaluating an object and its movement.'

The system can then send an alarm and manage recording. Here is one of the great advantages – video analytics will always do a better job than a human watching a scene. The human brain is not designed to do constant processing without a break in concentration and research has shown that this can be as short as a few minutes, especially with dull work.

Here are five things you should discuss and have confirmed with your supplier before you invest:

What am I paying for and how much?

It is a cliché, but you get what you pay for. The companies that do proper video analytics are generally from the USA with a couple in Israel, Europe and Australia. To do analytics properly and accurately involves a huge investment and enormous R&D work, there is no quick route to market. For true video analytics, and not motion detection or some algorithm that looks like it is putting boxes around movement in video, you will need to pay to cover the costs of the software. However, this may seem like small change once you understand how powerful this can be.

Using look-alike software to do analytics can cost dearly in the long run: high false alarm rates, inaccurate detection, wasted bandwidth are just a few of the problems. Most important in surveillance however, is the tendency for operators to become blasé due to overburden from too many alerts. This is almost always due to poor analytics detection.

So, it should not be dirt-cheap and it should not cost a fortune, but the technology has come a long way and done properly it may be the best thing you could do to combat crime or control costs in a business. Also, make sure you check on licensing and maintenance/upgrade costs, if there are any, up front.

Apples and pears. Analytics vs (advanced) video motion detection, what is the difference?

So many companies claim that they can do this and it is a dangerous claim. It is relatively simple to analyse video data from a CCTV camera, work out that something is moving and box it. This often looks great in a demo on YouTube or at IFSEC. In the real world things can work out very differently; during an average day any camera will at some stage pick up reflections, shadows, inhuman objects and other noise (like a poor camera connection). This is where analytics is really put to the test because if it cannot learn to ignore noise and filter out glare, shadows or irrelevant movement then it is not proper analytics.

Video motion detection (VMD) is just that, an algorithm that checks for motion in video and responds accordingly. There is really nothing wrong with this technology for very basic applications, like a low end system to use indoors at home for example. Where you will run into problems is if you try to deploy this commercially, especially outdoors.

So, video analytics systems should be able to understand what each object is, even if partially obscured (behind a tree or shelf for example) and as far as possible ignore anything that a camera sees that is not critical to detect. This is the difference between a system that is extremely useful and one that is useless.

Indoor or out? There is a big difference

It is far easier to deploy analytics indoors in a controlled environment with constant lighting and very little scene change. Here you are less likely to run into problems like sunlight and shadows, headlights and rain wind, foliage or animals. Now the system has less work to do to figure out what is what, objects are generally clearer and better framed against a very constant background. So good VMD can also work here and is often a viable option if it is going to give you the results you want.

Outdoors is another story altogether. Because of the real outdoors in Africa, we do have wildlife and storms and all the other challenges mentioned above. Now the system has to do a lot and this is where we can really sort the men from the boys in the industry. Also, here it is critical that you have chosen an excellent lens, camera/sensor and installation design combination to ensure that the analytics system can get the best quality video data to analyse. Here we are talking about quality ultra-wide dynamic-range cameras and very good light sources for dark.

Object classification and quick searches

Imagine the frustration of knowing there has been an incident and knowing you have got it captured on camera but not being able to find it quickly in order to respond. Proper analytics systems are able to learn the difference between objects, a person, a car, a motorbike, etc and even the characteristics of that object; a blue bakkie or a person wearing black pants and a red top. Now searching becomes easy.

So, what you should make sure of is that your system can track objects and people and keep a complete database of clips by object so that you can respond or find forensic data instantly.

What about hidden costs?

Sometimes analytics systems can be time consuming and cumbersome to set up. We call this calibration and it involves distance and object measuring. Outdoors this can take hours per camera and may need to be done several times a year as seasons change, so be careful because these call-out costs can add up in the long run. Also, many systems require centralised (server) processing unlike the newer edge devices. The problem with central processing is the same as in the old mainframe days – the server is down and takes all the cameras down with it.

Further, processing centrally means getting all the video from all cameras to one place live. This usually requires a sturdy network with huge bandwidth available. Often companies and estates put in a separate network just for these cameras and this can be very expensive. Any problems like degradation or outages across the LAN and your analytics is compromised.

Newer technology has clever self-calibrating software built-in which means that as long as any camera is on and running, it will continuously adjust, calibrate and learn. Even better if this is done inside the camera at the edge so that you are not dependent on a network to carry irrelevant video streams. By doing this you will reduce bandwidth needs by over 90%!

So, it is important to note that analytics should be done as close to the edge as possible and have self-calibration built in to save on rather unwelcome service costs.

In conclusion

If done properly, analytics is a winner. It reduces operator costs, guarding costs and for retail research, a lot of time and money. It is here to stay, but please make absolutely sure that you know what you are getting. Proper video analytics provide a fantastic return on investment, superior detection abilities to a human and a good system can dramatically reduce our dependency on guards to detect either at a post or in a control room. Today, video analytics finally under promises and over delivers.

Share this article:

Further reading: