A bank robber who has covered his fingers and face may think he is safe and nobody can identify him, but it’s not so easy to escape.
He can be easily identified by the way he walks out of the bank, thanks to automated gait analysis.
The technology could soon form the basis for a new generation of security systems, New Scientist reported.
In a leading technique, known as the gait energy image, .computer vision techniques use video images of a person to create a blurred silhouette that is characteristic of their gait..
A human operator links this gait “signature” to a person's identity, allowing the system to automatically spot that person when they are next caught on film.
This technique uses just a blank silhouette, but Martin Hofmann and colleagues at the Technical University of Munich in Germany have developed a version that also extracts information from the person's image, such as the shadows on their clothing, which leads to a more detailed signature.
Hofmann also used Microsoft's gaming sensor Kinect to measure depth, allowing him to better separate the target from the background.
The result is a system that is better at tackling tasks that cause problems for the standard version of the technique, such as recognising a person carrying a briefcase.
In tests using videos of several hundred people the system achieved a recognition rate of almost 80 per cent, outperforming 13 other gait analysis methods, including the one using gait energy images.
Finding a way to identify a person captured at different camera angles is another problem that has troubled researchers, Daigo Muramatsu and colleagues at Osaka University in Japan are now working on a solution.
They filmed 20 people on a treadmill using 24 cameras ranged around them and used this data to write software that can model the appearance of a person's gait when viewed from different angles. In preliminary tests, the system led to lower identification error rates at almost all angles, results Muramatsu describes as “promising”
According to New Scientist, Muramatsu and Hofmann will present their work this week at the BTAS biometrics conference in Washington DC.
These and other developments suggest that automated gait analysis might be ready for commercial use in the near future.
Not only this researchers at the National Physical Laboratory in Teddington, UK, have developed a demonstration system that can track people as they move through the laboratory building by their gait alone.
One can also tell if a phone has been stolen by a change in the walking pattern of the person carrying it.
An Android app developed by Marios Savvides and colleagues at Carnegie Mellon University in Pittsburgh, Pennsylvania, uses data from the accelerometer and gyroscope that come as standard on modern smartphones to record the movements that a phone makes as its owner walks.
In a study to be presented this week at the BTAS biometrics conference in Washington DC, Savvides shows the app can identify a particular gait with over 95% accuracy.
The technology could one day be used to shut a device down if it registers a gait that does not match that of its owner.