A computerized image processing system is being used in public places in the UK to identify criiminal activity and other problems in crowds:

EU policy to encourage more people off the roads and onto public transport, as well as addressing the needs of public service operators, has led to a series of collaborative research projects between universities and industry for several years. Leading UK universities in this area include Kingston University, Reading University, University College London (UCL), and Kings College London (KCL), among others. The research groups in the Digital Imaging Research Centre (DIRC) at Kingston University, led by Dr. Sergio A Velastin, formerly of KCL, and the Centre for Transport Studies (CTS) at UCL, where the key researcher is Dr Maria-Alicia Vicencio-Silva, are working collaboratively in a number of related areas directly aimed at alleviating some of the public transport issues highlighted.

Under EPSRC and EU-funded projects, the teams have developed a distributed pedestrian monitoring system, based on image processing techniques, that brings suspicious images to the prompt attention of staff. The system consists of a set of networked modular components using open standards for communications and annotation along with a number of dedicated supervisory PCs, called MIPSA (Modular Integrated Pedestrian Surveillance Architecture). It can be connected to an existing conventional CCTV system (16 cameras/MIPSA), collecting and storing the video streams, routing them to image processing devices, and feeding processed data back to the CCTV control centre using a unified interface.

In order to ensure the system is reliable and robust, thus reducing the possibility of false alarms, simple image processing techniques have been used. Video data is first processed to separate motion from the fixed background. One of the key enabling elements of this project is the DSP video processing based hardware, designed and manufactured by UK firm, Sollatek Ltd (www.sollatek.com), which is used to provide the fast calculation of motion data from the video streams. While UCL researchers have identified the features that need to be extracted, the Kingston University team has concentrated on the image acquisition and processing requirements.

In operation, the system takes conventional video inputs, combined with site-specific information from the CCTV operators, to detect unusual motion patterns which can arise due to situations such as congestion, trespassing and threatening behaviour. Excessive permanence of an object that is known not to be part of the background, for example, could indicate loitering, an injured person or even a suspect package. Similarly, large areas of motionless objects could indicate congestion, while an object moving against the flow could indicate an incident requiring further attention, such as gate jumping. In this application, extracting the basic image properties is sufficient; the system does not need to refine the nature of the incident further, but simply alert the relevant CCTV monitor operator.

It can detect patterns of movement characteristic of pickpockets:

He fends off concerns by explaining that it's very difficult to recognize people in these images, and also that no one is going to be targeted for features like unusual clothes or an identifiable walk. Instead, he says, more universal situations will be programmed into the system, such as people walking against a crowd, which can be a sign of pickpocketing. "They're only going to be stopped and investigated if there is sufficient reason to do so," he says.

This technology of course will become increasingly sophisticated. Science fiction author David has argued that pervasive monitoring by computer cameras and other sensor systems is coming in our future. See his book The Transparent Society: Will Technology Force Us to Choose Between Privacy and Freedom?. You can also find out more about his book at this page on his web site.