Research

One of the key elements in security management using surveillance cameras is to have information on suspicious people and their activities. Key is to have available an overview of the situation, certainly when it happens in public space. Do we know where this person went to? Did he talk to someone who should also be traced? Has he been in the neighborhood of crime scenes, or potential targets like government buildings or crucial elements of the rail infrastructure?

A problem in surveillance scenarios is the abundance of information or the absence thereof. With the current proliferation of surveillance cameras surely one may expect to have available 10-100 cameras observing people related to one case. This may constitute several hours of possibly relevant data to be scanned and evaluated as quick as one can. In times of serious threats reliably interpreting and cross- relating all data from all cameras is impossible to do for members of the observation team. What are needed are tools flagging the data automatically so that one person can manage the extraction of as much relevant information as possible. Hence it is crucial that the information be presented in a manner which supports the management of the information in quickly obtaining answers to burning questions of the officers, the government, and the public.

It is well known that identification of people from surveillance cameras is a non-trivial task. Even humans have a hard time doing it. Hence, relying on automatic systems to provide complete answers is unrealistic. The approach we take will reduce the data abundance by automatic analysis where the information is presented such that maximum speed to find answers is guaranteed. We define our goal: To develop a reliable set of software tools for the analysis and visual presentation of information from multiple surveillance cameras.