Technology

Surveillance system searches million of faces per second, looking for a match

A new video surveillance system is able to search through data on 36 million faces per second, looking for a match for a specific individual (Image: DigInfo)
A new video surveillance system is able to search through data on 36 million faces per second, looking for a match for a specific individual (Image: DigInfo)

Japan’s Hitachi Kokusai Electric has developed a surveillance system that can automatically detect a face in either a provided photo or video footage, then search for that same face in other video provided by networked cameras. While such facial recognition systems have been seen before, this one is able to compare the target face against others at an astounding rate of 36 million faces per second.

The Hitachi system assumes that all facial images it detects are at least 40 by 40 pixels in size, and that they are angled within 30 degrees of the camera, both vertically and horizontally.

It delivers search results immediately, in the form of a series of thumbnail images. When users click on any of those images, they are able to view recorded footage of what the person was doing, right before and after the displayed frame was shot.

Needless to say, it’s a lot more than might be required by some users – it’s intended mainly for large-scale applications such as railway stations, department stores and law enforcement agencies. The company plans to have it ready for clients by the next fiscal year.

More information is available in DigInfo’s report, below.

Source: DigInfo via Popular Science

  • Facebook
  • Twitter
  • Flipboard
  • LinkedIn
5 comments
Foxy1968
Wow. A system identifying 36 million faces per second could do the entire planet in just over 3 minutes.
I know the planet is over populated but there is nowhere that there would be traffic levels that high even if you had only one system monitoring an entire city that would only be a moderate load for this system.
I gather that once the system has I identified an individual they stay in the system for ever and their movements from camera to camera logged for future reference.
The amount of data that a system like that would create in a day would be staggering.
The people running it would need a doctorate in data storage to keep up.
Mr Stiffy
This is both very good and very horrible.....
The surveilance society to the max.
Mitch Cohen
I'm sure like every other spec, this is a maximum, with and under ideal conditions. The speed of all facial recognition depends on several factors, including size of the database being compared against, the false positive and false negative rejection ratio and the clarity of the image. This is not for use with a standard cctv video setup, The size and clarity of each face (and position) won't meet their standards.
This is, however, a great step forward for homeland security, being able to scan a passenger portal for those on a "no fly" list for example. The database and engine use an algorythm based on distance and proportion of various facial features for comparison. When a match or close result is found, it registers a "hit" and calls up the JPEG or other image file the algorythm was based on for human verification. That's how biometrics are stored and used in security systems. Otherwise the amount of storage and computing would be astronomical.
kilgatron
Privacy? Less and less. 36M now, how fast in 2020? By that time, a system's abilities will be down to seconds. Amazing. Frightening. Part of the singularity.
electric38
Coupling this info with DNA samplings and good old fashioned fingerprinting from every human and the "America's Most Wanted" list may shrink a bit faster.