Amethyst - guiding CCTV intruder detection towards the future .

August '99 CCTV, Surveillance & Remote Monitoring

Intruder detection is vital in any high security site. There are many ways to detect motion, be it buried cable sensors, fence alarms, infrared detectors or video motion detectors. In some instances, more than one of these can be used together in an improved integrated alarm system.

The biggest problem with any of these options is false alarms. Certainly, some intruder detection methods reduce false alarms more than others, but extreme weather conditions, movement of animals and/or trees and bushes are just some of the factors that may affect performance in any of these solutions.

A security guard may be alerted to anything between 10-100 false alarms every day. Each false alarm has to be checked and cleared after thorough investigation, which not only wastes a guard's time, but also affects his concentration to the point where genuine alarms may be overlooked.

At top security sites that are manned 24 h a day, guards are usually provided with banks of monitors that they are expected to watch attentively. In reality, physical security devices are needed to alert a guard to look at the video monitors. A guard detecting an incident by means of video surveillance alone is low, due to his relatively short attention span.

In 1995, the Police Scientific Development Branch (PSDB, UK) presented a paper at the prestigious Carnahan security conference, looking at ways to develop a completely new kind of activity detector. A PSDB research team, led by Michael Horner, was looking into the possibility of a design combining the latest in video motion detection (VMD) and perimeter intrusion detection (PID), into one integrated system - a vision of security technology of the future.

In 1997, a prototype system was devised, and by 1998, Amethyst (AutoMatic Event auTHentication SYSTems) was delivered to the PSDB, where it is currently undergoing extensive and rigorous trials. This ingenious solution reduces false alarms by a factor of 10, and is designed specifically for high security perimeter protection.

The issues

What the PSDB were looking to achieve was a perimeter alarm system that reviewed and verified potential alarms, passing on only genuine alarms to the guard. Video Motion Detection (VMD) was a potentially viable candidate for development, although there were some problems associated with using VMD alone.

Basic VMDs use image differencing between successive video frames to detect motion. Changes in lighting, auto-iris setting and camera movement or vibration cause differences between successive video frames, generating false alarms.

In addition to the false alarms created by lighting changes and camera shake, there was the problem of 'nuisance' alarms caused by movement of tree foliage, ripples on water or heavy rain or snow. Also posing a difficulty were the false alarms caused by wildlife such as insects, birds and other animals.

More sophisticated VMD systems use complicated algorithms to analyse each camera frame, ruling out the majority of false alarms. These sophisticated VMDs are able to differentiate between the movement pattern of blowing trees, running water etc and a real alarm situation.

Image size sensitivity controls can also rule out certain alarms such as small animals and birds, but an alarm may still be generated if, for example, a deer or other large animal were in the field of view. Perspective could also be problematic too. When a VMD was used with a wide-angle camera, perspective could confuse a human being crawling far away from the camera, with, say, a rabbit in closer proximity.

Whatever solution was devised, it had to be capable of identifying and ignoring false alarms caused by:

 Unidentified camera shake.

 Birds, insects, foxes, dogs and cats.

 Wind blown debris.

 Rain or snow falling on the camera lens.

 Moving trees and bushes.

 Dazzle and distortion from snow.

 Shadows and changes in light.

In order to meet these demanding criteria, the final solution adopted state of the art technology in one integrated system.

The concept

The PSDB determined that the best solution was to devise a system for use in perimeter protection applications, combining VMD with another type of perimeter intruder detection system (PIDs). This would, in turn, improve the security guards' confidence in the alarm system, and lead to a general improvement in site security, easing the pressure on guards to accurately identify and verify every alarm. If for any reason the VMD was unable to perform correctly due to low visibility or other factors, the system would revert to PIDs operation alone.

An 8-channel prototype was decided upon, that would analyse recorded footage from a loop framestore and comparing and verifying this footage against preprogrammed 'failsafe' checks. Using data capture and an alarm management system from Roke Manor Research, alarms were to be grouped into three categories; true, false or unconfirmed, with any unconfirmed alarms sent to the guard as a precautionary measure.

An invitation to tender was issued, and in 1997, Primary Image was granted the contract to adapt and refine the algorithms already used in its existing video motion detector, VideoTracker, to produce the detection algorithm.

The difficulties and the solutions

Devising the technology to perform all of these functions was not easy. Major problems encountered were similar to problem areas faced by VMDs, such as coping with cloud movement, light changes and shadows.

It was imperative that the system must not be too complicated to either use or set up, with a target time for set-up of 30 s per camera. The PSDB also wanted the system to be able to zone mask certain areas of the cameras' field of view, such as public footpaths and adjoining roads.

For some alarms, there would be little or no motion of an intruder in the alarm sequence, for example, when an intruder was cutting a hole in the perimeter fence. Here, it was necessary to provide very detailed analysis. Also needed was a 'large target' algorithm that could successfully identify vehicles and humans, and differentiate these from birds and insects at close range to the camera.

The final Amethyst system has two main inputs: the alarm triggered by the PIDs and the CCTV sequences associated with them. These sequences contain pre- and post-alarm footage, which is analysed by Amethyst. The system typically operates in blank screen mode, only forwarding those alarms to the guard that are genuine and have supporting video evidence. The system can either be used as a stand-alone unit, or it may be integrated with an existing alarm management system.

Eliminating false alarms

In order to eliminate the false alarms, the system has a series of 'failsafe' modes where it compares each frame with the last and next frames in the series to check for signs of motion. Any camera footage that is of insufficient quality (i.e. due to poor visibility or loss of camera focus), is always forwarded to the guard for verification.

Firstly, basic failsafe checks are made. Any discrepancies will indicate camera bagging (where the camera has been covered over), poor contrast, loss of picture or dazzle.

Because the footage being assessed is not realtime, camera shake can be a major cause of false alarms, as the scene may change markedly between two frames that are not directly adjacent. The detection algorithm has built in compensation to remove any discrepancies.

Then the focus, range, possibility of floodlight failure and deep shadows are all double-checked, before objects that differ from, or are not in the background are identified. Any discrepancies are counted and sized.

Then, further elimination of animals and small objects are removed from the images, before a series of perspective checks compare the height of a human from different distances.

Next, user defined failsafe criteria are checked, to eliminate things such as animals, insects and flying debris. Once these three stages have occurred, images from each stage are compared to identify discrepancies. Finally, an illumination filter removes light changes from the image, before checking again to remove any final false alarms from the images.

At this point, stationary objects can now be detected by first applying failsafe processing to ensure that the images are suitable for automatic assessment. Then, the images are processed to see if an intruder can be detected by searching for motion between images.

On detection of an intruder, the series of images are sent through to the control room, accompanied by an audible alarm, for the guard to investigate.

The final product

Using this state-of-the-art technology, the false alarms outlined above can be successfully eliminated, while Amethyst is capable of successfully identifying the following attack profiles:

 Crouching, with little or some motion.

 Walking or running.

 Climbing.

 Vehicles.

 Crawling or rolling.

By having the ability to differentiate and filter out any other unwanted information, the guard is now able to identify the cause of any alarm immediately.

It is developments such as these, which are leading towards the 'virtual guard' scenario, where a CCTV security system can be automatically manned 24 h a day. Such automated machines eliminate the problem of a guard missing events due to his relatively short attention span (thought to be a maximum of just 40 min), while improving the chances of incidents being detected accurately, and virtually instantaneously.

Such state-of-the-art systems are now sufficiently reliable to be used in critical applications such as airports, prisons, nuclear sites and high security site 'sterile zone' monitoring.

The future

These, and other similar systems, may be used to improve external security for buildings and other facilities that are not protected by perimeter fences, as well as being used to improve public safety in communal areas and transportation systems. The other benefits of such systems include reducing CCTV guard manning levels, minimising vandalism and loss of property and improving deterrence by catching perpetrators 'in the act'.

It is envisaged that such systems will have numerous applications such as public safety in commercial buildings and transportation systems, 'perimeter' protection of commercial and government buildings and protection of industrial and utility sites.

Easily integrated into existing public CCTV installations, these systems would further improve the 30% reduction in the crime rate associated with similar projects.

For details contact Jo Mclean, Marketing Executive of Primary Image (Vision Systems) on tel: (0944) 181 339 9669 or e-mail: joannem@primary-image.com.





Share this article:
Share via emailShare via LinkedInPrint this page



Further reading:

Body-worn cameras transforming security
October 2019 , CCTV, Surveillance & Remote Monitoring, Security Services & Risk Management
Police Service Northern Ireland now has over 7 000 officers using 2 500 cameras covering approximately 173 000 incidents each year.

Read more...
Camera Selection Guide 2019
CCTV Handbook 2019 , CCTV, Surveillance & Remote Monitoring
The Camera Selection Guide 2019 includes a range of IP, analogue, thermal and speciality cameras aimed at a broad range of surveillance functions.

Read more...
Intelligent analytics and the brains to match
September 2019, Bosch Building Technologies , CCTV, Surveillance & Remote Monitoring, Integrated Solutions
What if the brains behind our security cameras could be trained to improve their cognitive ability to pay attention, learn, and problem-solve according to specific rules and situations?

Read more...
AI-powered autonomous Drone-in-a-Box
September 2019 , CCTV, Surveillance & Remote Monitoring, Integrated Solutions
Organisations in the mining, energy and industrial, oil and gas, ports and terminals sectors can optimise security and business operations, whilst reducing risks and operational costs

Read more...
Cybersecurity for video surveillance systems
September 2019 , Cyber Security, CCTV, Surveillance & Remote Monitoring
Video surveillance systems are increasingly accessible over any IP network, which has led to the rise of potential cyberattack.

Read more...
A platform approach to innovation and value
CCTV Handbook 2019, Technews Publishing , Editor's Choice, CCTV, Surveillance & Remote Monitoring, Integrated Solutions, IT infrastructure
Moving to the platform model of doing business holds tremendous advantages for end users and smaller developers, but also for the whole technology supply chain.

Read more...
Back to manufacturer uniformity?
CCTV Handbook 2019, Dallmeier Electronic Southern Africa , CCTV, Surveillance & Remote Monitoring
Decision makers often have to choose between an all-inclusive, complete solution from one manufacturer and the products and services of several different providers.

Read more...
Open does not always mean easy integration
CCTV Handbook 2019, VERACITECH, Technews Publishing , Editor's Choice, CCTV, Surveillance & Remote Monitoring, Integrated Solutions
Customers who opt for best-of-breed solutions will have to rely on their integrators to develop customised integrated solutions for them.

Read more...
The impact of AI on the surveillance industry
CCTV Handbook 2019, G4S South Africa, Hikvision South Africa, Myertal Tactical Security, Technews Publishing , Editor's Choice, CCTV, Surveillance & Remote Monitoring
What the impact of AI will be on companies, the services and solutions they supply, as well as on the jobs people do.

Read more...
Video analytics and AI
CCTV Handbook 2019, Axis Communications SA, Dallmeier Electronic Southern Africa , Hikvision South Africa, Technews Publishing, Dahua Technology South Africa , Editor's Choice, CCTV, Surveillance & Remote Monitoring, Integrated Solutions
Artificial intelligence has the potential to deliver real benefits in the world of video analytics and many companies are already delivering customer benefits.

Read more...