Researchers hope the smart cameras will have a dramatic impact on crime detection and prevention by cutting response times and ensuring that more incidents are caught on tape.
A team at the University of Portsmouth has already developed software which enables cameras to spot visual clues to anything from violent crime to vandalism, by looking for tell-tale signs such as someone raising their arm suddenly or even a snapped car aerial.
Now the artificial intelligence software is being taught to recognise sounds which are associated with crimes, including breaking glass, shouted obscenities and car alarms going off. Cameras which 'hear' the sounds will automatically swivel to the direction they have come from, and will alert the person monitoring the system to a possible crime in progress.
Dr David Brown, who is leading the team at the university's Institute of Industrial Research, said: "We have already developed visual recognition software, but the next stage is to develop audio recognition software to listen for particular sounds.
"We can teach the cameras to listen out for things like a swear word being shouted in an aggressive way, or for other words which might signify a crime taking place. The camera will be able to swivel to the direction of the sound at the same speed someone turns their head when they hear a scream, or about 300 milliseconds.
"People monitoring CCTV images have banks of screens in front of them, and this system helps them by alerting them to something the system has spotted. The person looking at the screen can then quickly identify if it is a crime taking place, or whether the camera has simply picked up on something innocent, like a child screaming, and act on it accordingly."
Although the new system will inevitably raise concerns about the unstoppable march of the "surveillance state", with one CCTV camera for every 14 people in the UK, Dr Brown said there were no plans for the system to record conversations.
"The system would not be sensitive enough to record individual conversations," he said. "We are just looking for certain trigger sounds and visual anomalies."
The software will also dramatically cut the amount of time police officers have to spend reviewing CCTV footage for evidence of crimes, because they will be able to use it to search automatically for a specific object, action or noise.
Dr Brown's team is one year into a three-year research project, in partnership with British CCTV company Neuron Systems, and is confident the system will go into production at the end of that period
