Science traps aggressive train passengers

28 October 2010

Computer models can help catch aggressive train passengers from now on. Dutch researcher Dirkjan Krijnders has made it possible for computers to detect the source of sounds more easily. He tested his system on fake hooligans with considerable success. Krijnders gained his doctorate from the University of Groningen on 29 October 2010.

As you wander through the city you experience few problems in holding a conversation. You can almost flawlessly recognise the voice of the person you’re talking to from amongst the bustle of city sounds. Similar background noises are, however, disastrous for computers because a voice recogniser also views this as speech, with non-sensical results. Therefore we still have a long way to go before sound recognition can be used for alarm systems. However, Krijnders models signify a huge step forwards.

Sound consists of three elements: resonance, collision and noise. In Krijnders’s model sound is divided into objects based on various physical processes. Each method is processed differently and a number of crucial characteristics from each process are recorded. These characteristics can be used to further search for known examples with similar characteristics, thereby helping the computer to determine the source of each sound.

Detecting aggression

Krijnders collected various datasets to test the models. One of the sets was recorded on a normal train platform where four actors played out various scenes. These scenes varied from normal station situations such as a greeting, to less frequent situations, such as the aggressive behaviour of football hooligans.

The dataset was also used in combination with a video detection system from the University of Amsterdam to detect aggression based on the amount of movement on screen. This combined system used the outcomes of both the sound system and the video system to provide the best estimate of the aggression level on the platform. By using the two different types of sensors, the number of 'good' alarm signals appeared to increase from 45% for sound alone and 67% for just video, to 78% for the combined system without the number of false alarm signals increasing.

The research was carried out as part of NWO's ToKeN programme. ToKeN is an interdisciplinary research programme in which cognitive science and ICT are focused on fundamental problems of the interaction between the human user and knowledge and information systems.

About NWO

The Netherlands Organisation for Scientific Research (NWO) is the principal Dutch science funding body and its mission is to facilitate excellent scientific research in the Netherlands by means of national competition. Each year NWO spends more than 700 million Euros on grants for top researchers, on innovative instruments and equipment, and on institutes where top research is performed. NWO funds the research of more than 5300 talented researchers at universities and institutes. Independent experts select proposals by means of a peer review system. NWO facilitates the transfer of knowledge to society.

..............................

Further information:

last modified on 25 November 2010