Is absolute zero cold enough?

Zero, the ultimate result of technology.

In my working life, one thing has caused me constant trouble. Finally, thirty-five years on I feel I am finally getting to grips with it. That animal is finally cowering and shrinking in size to a point where in just a few years it will be dead forever.

It sounds dramatic. It is not. Ultimately to most people, it is boring.

However, it is about doing things right, and that could save lives. How dull is that?

I am talking about false alarms (FA), false triggers, untrue positives. Whatever you call it, it has been a plague on the house of security ever since the invention of the first electronic security device.

For ages, the prospects of the reduction in FA seemed promising. 

SIA and Contact ID protocols were supposed to allow the writers of control room software the ability to, in real time, to display and track the progress of intruders through a building or discover if the FA cause is owner error. Adopted by one or two the concept gained high plaudits, but then, along came those with no or little engineering skills, and the protocol neutered. 

Instead of tracking in real time it changed it to be a double knock. We, those of us in the committees screamed and kicked but, no, the big companies knew best.

CCTV transmission has been around since 1985, first with Ibbsonscan, then Robot, then that grand master of the DVR, Dedicated Micro. It was supposed to enable control rooms to protect using CCTV visually. The transmission was so slow it was laughable, the same low skilled engineers not understanding that the control room was not seeing the same images as they were on site. Their inability to link even the simplest of detectors in a way that enabled the controllers to look at the same area as the detection a cause of frustration. It continues even today.

Good manufacturers responded, they produced DVR that sent snapshots and alarms in a professional manner, these tied to the output from the panel. Using the SIA and CID protocols to direct the DVR to the right camera and time. It worked, but not everywhere. The standards, now adapted to remove the intention of the original designers, had changed the programming of almost all intruder alarm systems. It was not even possible to send SIA or Contact ID correctly. We just went from 4*8*1 to a dumbed down process.

No matter how hard we in the sharp edge of control room design worked, it seemed political interference by those that could not or would not understand the technology would change it. It took the best part of 20 years for the ratification of EN50136. Possibly as the result of deliberate and concerted efforts of a major telephone company whose product could not, and never could, meet the requirements. Much to the frustration of our European colleagues in the committee stages.

Another reason? I have never been entirely sure, but I suspect that charging per hour at the control room has been a significant factor. It is possible to make more money by sausage factory processing instead of doing a good job: missing the occasional alarm event compensated by the charging culture. The security industry was paying out now and again because of its roomy pockets, lined by clients who need not have spent so much and could have had better service. The industry paid out, but people died, something I will never forgive.

Could I finally be winning? Possibly, maybe. If everyone with the same moral compass does not allow others to alter the work undertaken by the new standards committees (less political interference) and the insurers keep the promises they have made.

The technology exists, it is ready. We can get to zero or very close indeed and give users a significant and immediate change for the better. At the same time boosting the impression that customers have of our trade.

The standard? EN62676.

Careful crafted it has sneaked under the radar some significant changes (we learnt) that directly affect every control room. Proposing alterations to the form and function of CCTV that make it impossible for installers and control rooms to have an excuse. Insurers will be able to look at every claim and turn away those that do not comply, killing those businesses that do not precisely do things. Coupled with the changes in company law it will mean that those at the top will have to take care of the systems they sell - corporate culpability is no longer an idle threat.

Fool look out, there is no room for ill-concieved engineering any longer. 

I will not go into the standard in any great detail. If you want to know the aspects that affect your control room read it, I suggest carefully. We are advising our clients and have been for some time.

Look in particular for transmission time. Event processing is expected to be in seconds, compensation for network delay can be objectively taken based on industry standard, not the way you do it. Even at grade 01. Images have to be the right size and form, lighting correct and arrive promptly. Learn how IR is better to set looking at the area you want to surveil, not at the camera point.

However, this is not all about a standard. That just gives real teeth to the design and operation of the control room and CCTV. The technology is finally here to make "zero" a real prospect, or at least as close to it as possible.

Deep Learning. Cognitive Recognition. 

How good is this?

It immediately makes even the most false alarm prone CCTV derived security system better.

Let's get one thing straight. It is not analytics. It is artificial intelligence. It is not looking for pixel change. It identifies objects by comparing to pre-classified examples, what it has seen before, and the same algorithm applies to detection.

It works best on HD images. It works best on low compression pictures or frames. It works only on GPU based processing networks. It works best on historic data. It learns and adapts.

It is not, as yet, possible to use it accurately in the field, at the edge. Unless you understand the technology and are prepared to pay for the right equipment; most of all, you need to understand it.

Cognitive Recognition has one significant advantage over analytics. It learns. Just as we, as children learn to recognise "cat", "dog." The more you teach it about the scene it is looking for, the better it will be.

The robots look at images and classify objects based on what they have learnt. 

I recently read a review of one of the first edge versions. The discussion was wrong; totally, the point missed entirely. To use cognitive recognition you have to train it. So what if it identifies objects incorrectly, it will do that until you tell it what the "object" is. Once taught it would not make the same mistake again. Unlike analytics which will always make the same mistake.

Fools will always look for weakness in what they do not understand.

So how are we using it to reduce FA received in the control room?

Firstly we require no change to current field equipment. We still expect the installers to map detection to cameras, to link events to the cause. It takes more time but is infinitely better than relying on the video device to send an email. We still pull from the edge recording device video in real time, as it happens, as the trigger is received (immediately). We still process every image, even if obtained as H264/5, as a single frame, patching keyframes to create an animation of the event. 

Every CCTV device since 2002 is compatible.

These frames, now with no deblocking filters and with edges pre-processed pass to the robot where it classifies what it sees. Each image analysed, and the information returned to the control room where another robot takes the data and builds a statistic based threat analysis. If we see people, we increase the threat. If not we decrease it. The result is very accurate, better overall than a human processing images in a production line environment.

We can make it better by teaching it about the mistakes it has made. The object recognised as a human, might be a statue? Right or wrong?

We can also use the same approach to long-range detection and perimeter systems. The control room taking each frame, and based on time of day, the temperature along with other factors, teaching the remote detection what settings it should adopt and not necessarily be relying on the detector to make those changes locally. Or at least allowing the smarter control room software to make more significant changes to the engineering function of that remote detector.

Security breaches can cause loss of life. It's that serious.

Finally, by moving the logic from the panel and installing detection with its own internet identity, we can go back to the original thought of the 1980's and track intruders or users in real time. Taking away the bastard child of double knock that has caused so much pain.

Thank you for your time.

Christopher Berry