The Sounds of Violence
IT’S NOT EASY to monitor numerous security cameras at once, and many sites can’t devote enough manpower to the task. In recent years, technology has attempted to solve the problem by helping officers sort out only what’s important within the video. One solution has been analytics, software programs that detect an action and sound an alarm.
The most common analytics are visual ones, like objects left behind or motion detection. But a U.K. company called Audio Analytics has developed an analytic that focuses on sound. It not only detects sound, but it also can tell if the noise is aggression, a cry for help, or something more benign.
The work began as a Ph.D project of Audio Analytics CEO Chris Mitchell. The company plans to have audio analytic software that can be added to a camera or DVR/NVR backend system, just like other types of video analytics. Many of the video surveillance systems or analytics on the market now either don’t use sound at all or only detect sounds but don’t necessarily classify them.
Audio Analytics would provide various sound packs that could be installed on the hardware, such as a package for aggressive noise or a package for car alarms. Any sound that comes into the system would then be analyzed for pitch, tone, and other information, and compared with the sound package and classified. Mitchell points out that amplitude is not what the system is looking for, because amplitude varies so much depending on how close to the camera the sound is.
The analytic could be set to route back to the control room so that officers could pull up the adjacent video, or it could just send an alarm back.
Mitchell says that “there are a number of natural advantages for audio over video for certain situations.” For example, he explains, “audio works irrespective of lighting conditions. It doesn’t matter if it’s low lighting; less than one lux doesn’t make any difference to our system. It just works. It doesn’t matter if the event is slightly off-camera; our system is still going to detect it.” Mitchell adds that in a crowded situation where a camera might not be able to see aggression, the analytic could still hear it.
Tjeerd Andringa, who coauthored a 2007 report on a project called CASSANDRA, which fused audio and video sensors to detect aggression, agrees that sound is an important analytic. “Sound goes in all directions. And our ears work in all directions. Our vision works only in a limited scope... and what we can really see is an even smaller bit of the visual field. ...Our hearing system is [also] very good in detecting the unexpected.”
The CASSANDRA system that Andringa reported on went beyond just using sound. It coupled an audio aggression sensor with video aggression cues in a project in a public train station. It successfully differentiated aggressive screams from passing trains and nonaggressive sounds.
The combination of the sound and video was particularly effective, says Andringa, because someone may have Tourette syndrome and make aggressive noises but not be harmful, or two people may be yelling to each other without creating a security incident.
Mitchell says that Audio Analytics has obtained sufficient funding to get through development, and it is now approaching vendors about adding the analytic to their products. He says the software
has been pilot tested in a public environment, although he could not reveal exactly where. Mitchell added that the false-alarm rates are lower than the industry averages for video analytics in similar environments, “So they’re well inside of manageability.”