(+34) 926 29 53 00 - Ext 6494   Spanish version
grupo.air@uclm.es

Research Lines :: Applied AI

Intelligent Surveillance

What we offer

We propose creating advanced surveillance systems based on using independent normality components which can easily be reused. Each normality component specifies how each entity capable of acting in the environment as regards an aspect of surveillance should behave.

Our systems is composed by different layers. The first one corresponds to the sensorial layer where the sensors entrusted with capturing any change in the state of the environment under surveillance are located. The information from the sensors can be sent directly to the intermediate layer (if previous processing is not required), or it will be processed by different types of algorithms for generating space-time information, such as, for example, segmentation algorithms, tracking, sound analysis or classification of objects in motion.

Additionally, there is a continuous communication flow between the lower and intermediate layers. The latter, at all times, receives the space-time information as regards each of the objects detected in the scene which have been processed in the lower layer. A clear division of these layers enables the analysis of events to be independent from the algorithms used to process the signals as well as to ease reuse of expert knowledge. That is, the change of algorithms to process the signals does not need to influence the way of analysing or determining if a situation is normal or not.

The main objective of the middle layer is to interpret what is going on in the environment being monitored and classify each of the situations as normal or anormal to sound the alarms and offer support for decision-taking. To do this, normality in the environment and the most common anormal situations are defined. The worst that can happen with this approach is that there can be an anormal situation which has not been previously defined, but at least, the system is aware that "something" suspicious or unusual is happening at the present time. Defining normality generally for the whole environment may be a task that is too complicated or even infeasible. For this reason, it has been developed by breaking the problem down into lesser complexity problems: the environment is divided into smaller areas (specific areas of the environment observed by certain sensors) where the number of possible situations which may arise is delimited, and the normality analysis is divided into specific analysis by using normality components.

Finally, in the last layer, the information from the intermediate layer is used to offer help to security staff with decision-taking in critical situations. In this layer the monitoring tools are located where the security staff can control the current state of the system and the environment under surveillance.. At present, there are no solutions which offer enough guarantees for completely substituting human workers but, this type of system can be really helpful for lightening the load the human workers have to cope with and for offering advice or suggesting solutions in complex situations in which a succession of anormal situations has occurred.

The development of surveillance systems based on normality components has some of the following advantages:

Reuse

Construction of independent components where normality is specified in accordance with an aspect of surveillance, enables the expert knowledge to be reused.

Reduction in the development cycle

As a direct result of the previous advantage.

Improvement in the surveillance quality

Reuse of previous experiences avoids making the same mistakes made in the past.

Reliability

If a component has been previously tested in a variety of environments and this has enabled potential mistakes to be debugged, the inclusion of this component in a new security system ensures a high probability of success.

Flexibility

The combination of some components or others enables different types of surveillance systems to be built. It could be said that the components are the "ingredients" of the application and they combine in one way or another to reach a specific goal, here, an analysis of specific situations.