printer friendly version

An end to complexity and frequent false alarms

No perimeter solutions are alike. But almost all approaches have two things in common: the combination and interaction between different technologies is complicated to implement and operate. And almost all perimeter solutions are dogged by a large number of false alarms, which must be checked by the security personnel and detract from its reliability.

Developments over the last few years promise to remedy this situation – providing the right technologies are deployed. A great deal has been done in two areas in particular. The first is the field of analysis: artificial intelligence (AI) is becoming more widely used in video analysis and promises to bring about a substantial improvement in detection accuracy. Consequently, for the first time, video analysis introduces the capability of partial automation of verification for perimeter protection. Secondly, in terms of the camera systems themselves: in combination with the modern analysis capabilities, some camera technologies are taking over more tasks of detection, that is to say the initial recognition of incidents.

A good camera system just for verification is wasted

In order to provide truly effective protection for a perimeter situation, a combination of several different technologies is needed. And there is no universal solution, but rather complementary technologies suitable for detection tasks must be chosen according the customer’s environment, such as taut wire systems, detection loops, radar equipment or acceleration sensors. Most systems are expensive to buy, implement (here infrastructure components such as wires and sensors usually have to be set up at intervals of a metre all around the perimeter), and finally to operate. The fact that no system is able to fulfil the role satisfactorily on its own just multiplies costs.

Multi-focal sensor cameras allow effective perimeter protection with a minimum number of systems.

So, would it not make perfect sense if the camera systems – which are often present anyway to provide visual confirmation of events – were able to take over the detection function, or at least a part of it? This is possible if the camera solution is designed for this purpose. To this end, of course the basic functions of video content analysis (VCA) must be present in the system and able to trigger an alarm as soon as an intrusion is detected – intrusion detection or line crossing represent just two examples of such.

Detection: As ever, the devil is in the details

But there are reasons why most of the camera systems used for perimeter protection are still limited to just verifying and do not have the larger role of detecting incidents. One reason is often linked to the difficulties associated with planning resolution quality: an analysis can only ever be as good as the data that is to be analysed. Users should therefore attach great importance to the fact that a provider offers capabilities for determining the precise minimum resolution density for successfully detecting objects and individuals with the minimum work and expense over the entire protected perimeter.

This is specified in the globally applicable standard DIN EN 62676-4, and depending on the scenario it is typically fixed at values between 125 and 250 pixels per metre for detecting and identifying a person. This capability to make a precise definition as early as the planning stage is a fundamental prerequisite for analysis, preservation of evidence and further detailed analysis by AI elements.

By using AI, the number of false alarms at the perimeter can be significantly reduced; the AI-based camera system only forwards the relevant alarms to the control centre.

Almost as important is the correct choice of lighting: if it is possible to use white light, one powerful camera system is usually sufficient for both daytime and night operation. Good systems are also able to detect objects and ensure that people can be identified satisfactorily in poor light with active infrared illuminators. If neither option is possible, it is recommended to add thermal imaging cameras to the solution for poor light or weather conditions. In any case, these do not present any problems in terms of data privacy law, but they cannot be used to identify individuals and the images are not usable as evidence, so it is still necessary either to rely on directed white light or security personnel.

Disadvantages of PTZ cameras for verification

For purposes of verification too, many mistakes can be avoided by choosing the right technology. Modern multi-focal sensor cameras such as the Dallmeier Panomera, for example, differ from the classic single-sensor camera plus PTZ camera combination, particularly in their consistent minimum resolution densities even over long distances, which means that fewer cameras are needed.

Another significant advantage is the capability to do multiple zooms when verifying, thereby making it possible to monitor even highly complex situations, such as intrusions by multiple individuals. Unlike conventional solutions, these zooms can be carried out in high-quality resolution both in live mode and in the recording, so all areas can be captured in detail as well.

AI reduces false alarm rate to practically zero

Apart from the smallest possible number of systems for perimeter protection, the area with the greatest potential for improvement is semi-automation of activities which currently still entail high costs in terms of personnel and time. AI is a crucial element in this issue.

Notwithstanding the justified wariness, quite specific solutions already exist precisely for perimeter protection, and these promise enormous benefits for the customer with regard to personnel costs and thus overall operating costs as well.

With the solution offered by Dallmeier, the pre-alarms triggered by the classic VCA – including the usual ‘error sources’ such as moving branches, animals or rapid changes in light conditions – are worked out in a second analysis step using neural networks. A certain detection probability is defined in the AI engine and only alarms which surpass this value are forwarded to the control centre for human verification.

The results are clear: experiences from the first customer installations indicate a reduction in false alarms to practically zero, and consequently substantially lower costs for verifying and investigating alarms. This immediately enables users to realise enormous savings, since much larger perimeters can be protected with the same personnel. Secondly, objective security is also improved because the operators are much more attentive. They know that when an alarm is triggered it is very probably caused by a significant incident rather than just one of many false alarms.

Share this article:

Further reading: