To help fight crime in trains, Zhenke Yang (MSc) developed a smart surveillance system that sets off an alarm when people act aggressively in train carriages. The system recognizes faces and walking patterns, and a microphone detects when someone screams.
On the international train from Amsterdam to Brussels, a man is asked to show his train ticket. He doesn’t have one. The conductor takes out his notebook to write a fine, but the man furiously slaps the notebook out of the conductor’s hands and punches him in the face.
A red light immediately alerts a security expert that there is something wrong. The expert is not on the train, but rather in an office in Utrecht. The emergency signal is relayed to this office via a smart camera in the railway carriage, which has recognised the punch as an act of aggression. The security expert then informs the other conductors on the train that their colleague needs help, while also alerting the police about the incident, so that they can arrest the man at the next stop.
This surveillance system was developed by Zhenke Yang, at the man machine interaction group (Electrical Engineering, Mathematics and Computer Sciences). He will defend his PhD dissertation on this subject on Friday, October 29.
Yang’s system uses several techniques to recognise aggression. The camera, for example, detects when a conductor stands still for longer than usual. “The normal pattern for conductors is to walk, stand still to check tickets, and then walk again,” Yang explains. “The camera recognises this and knows that it’s a conductor because of the uniform. If a conductor stands still for a long time, this could mean there is something wrong, so the camera sends a signal to the security expert in Utrecht, who then checks the live footage to see whether there is a problem or if the conductor is just chatting with passengers.”
Yang developed his system because security experts receive massive amounts of images from all the cameras in international trains: “They can’t check all the images at the same time; therefore, my camera only raises the alarm when it detects that something is going wrong.” Yang also developed a classification of alarms. If a conductor is standing still for a long time in one carriage, while someone else is being assaulted in another one, the latter incident is labelled as more important to the security expert. Yang: “This ensures that the security expert can act as quickly as possible when something really goes wrong and is not too busy solving minor incidents.”
The smart camera’s microphone detects when someone is screaming or talking loudly or in an angry manner. “I had to take into account that people talk louder when the train is making more noise when stopping at a station,” Yang says. “At the moment, the microphone only detects an increase or decrease in volume, but research conducted by others has shown that when people talk while they’re angry, there are certain cues that change compared to when they speak normally. I would like the microphone to pick up these cues as well.” The smart camera uses a special kind of facial recognition. At present, the camera recognises where faces are located in the train, but in future Yang hopes the system will also be able to detect the emotion in the face.
The aim of this smart camera is to reduce acts of aggression in public transport. “I hope the cameras will work preventively, once people realise that the camera detects everything and the security experts respond very quickly to crimes. After installing cameras in Zoetermeer, we saw that the damage done by vandals decreased dramatically. And 60 percent of the passengers felt safer in the train. Unfortunately, these were not smart cameras, and once the vandals figured out that there was no immediate reaction to their crimes, the rate of vandalism rose again. That’s why a smart camera is needed.”
It is not yet known when these smart cameras will be used in trains. “We tested the camera by having actors act out various scenarios in train carriages, but more testing is needed,” Yang says. “We now need to see if the smart camera detects all kinds of ‘real life’ aggression in trains, so that we’ll know for sure that the system works.”
De TU Delft publiceert acht procent van haar wetenschappelijke artikelen in samenwerking met de industrie. Dat is veel vergeleken met andere universiteiten. Van de 350 best presterende universiteiten ter wereld staat de TU op de vijfde plaats in een rangorde opgesteld door het Leidse Centre for Science and Technology Studies. Vooral binnen de vakgebieden elektrotechniek en de medische wetenschappen publiceren Delftenaren veel samen met het bedrijfsleven, respectievelijk 17 en 16 procent. Op nummer één staat de TU Eindhoven met 10,5 procent. De onderzoekers namen data uit het Web of Science voor de jaren 2002 tot en met 2006 onder de loep, en bekeken voor alle publicaties wie erbij betrokken waren. Web of Science verzamelt de wetenschappelijk publicaties van zo’n 14.000 wetenschappelijke tijdschriften. Drie Japanse universiteiten staan op de tweede tot en met vierde plaats in de ranking.
Comments are closed.