CCTV Camera Identify Criminals by Way they walk
Criminals could soon be identified simply by the way they walk if a new CCTV system from Japanese researchers is adopted. Reuters

Police could soon be able to identify individual criminals in crowds of 1,000 people simply by the way they walk.

Japanese scientists have designed the first computer program for criminal investigators that can identify people captured on CCTV security camera footage with 99% accuracy by measuring how they walk, together with other physical characteristics.

A sequence of images from one camera that shows an individual in movement is fed into the specially designed piece of software, and once the person's walking style, including hand movements and stride - collectively known as gait recognition - has been identified, the system can see whether the footage from other CCTV cameras offers up a match.

The team of scientists from the Institute of Scientific and Industrial Research at Osaka University, led by Professor Yasushi Yagi, used the computer program to successfully single out individuals in videos that featured 1,000 people.

Gait recognition

When only walking measurements were used to analyse the videos, the computer program achieved a 90% identification rate, but once facial recognition and other physical statistics like height were added, the program was able to successfully identify individuals 99% of the time.

Criminal ID system using Gait Recognition to spot suspects
Osaka University scientists have invented the first criminal ID system using Gait Recognition to spot suspects Osaka University

Gait recognition is a fairly new addition to biometrics, the technology that measures physical characteristics related to the body, such as facial recognition, fingerprints, iris recognition, and even DNA.

Movies and particularly serialised crime investigation dramas on TV are often ridiculed for showcasing criminal ID systems whereby the police only need to click a few buttons and a low-resolution CCTV image is magically enhanced, clearly showing the face of the suspect, however this could soon become a reality.

Daigo Muramatsu, Haruyuki Iwama, Yasushi Makihara and Yasushi Yagi state in a paper: "An advantage of gait recognition is its ability to ascertain identity from a distance and hence it works well even if the image resolution is significantly low (eg just 30 pixels in height)."

A seminar has been planned to educate Japanese police on the new system "in the near future", but there is no timeline given for a wider release for the system.

UK courts

The scientists believe that their system, which is able to analyse multiple perspectives from a single camera, has potential to be applied to "real situations", since criminal investigators are not always able to obtain multiple views of the suspect from multiple cameras, which is what some of the other research into gait recognition concentrates on.

Automatic gait recognition footage from CCTV cameras was first admitted as evidence in UK courts to convict a burglar in 2008.