Michal Balážia, a doctoral student at the Faculty of Informatics, uses video footage in the search for ways to recognize people by their gait. A mere two steps suffice for recognition, he says. In February he received a Swedish Innovation Prize in Civil Security for his work.
At passport control, for instance, fingerprints and facial features provide biometric data commonly used in identification. But every person has many more unique characteristics by which he or she can be identified. Gait is one of these.
Many of us recognize people we know by their gait. But can such a system have a more general application?
Every person’s manner of walking and movement is sufficiently singular to enable their identification. We are conducting research that should lead to the development of a system that will make this identification possible. In the event of a criminal act captured on camera, for instance, our system would use video footage to determine the offender’s characteristics, seek these out in other images, and on the basis of comparison of certain characteristics including gait, track down the offender and put him under observation.
How can gait be characterized?
At present our research is still at the design stage. We are still forming our mathematical concept for the parameters we need to determine by means of the study of video footage, and we’re verifying which of these capture adequately the dynamics and uniqueness of a person’s gait. The human body has many joints for us to label. We can apply algorithms to a very short recording of a person’s movement to create a kind of human skeleton and calculate the coordinates of selected joints over time. These coordinates help us draw a trajectory, from which we can deduce many other things. In my thesis I dealt specifically with signals that determine distances between two points in time, creating for these signals highly accurate comparative functions.
How can this be applied to gait?
The data available to us tells us that walking is a periodical activity. Although we do the same thing over and over again, the walking cycle comprises a mere two steps. We identified this cycle on the basis of time-distance signals at various points on the body, for instance both feet. We then tested these findings using a database of the movement trajectories of various people, which was made available to us by Carnegie Mellon University in the USA; the figures this gave us convinced us of our success. Our first task, however, was to prepare trajectories of movement for individual points of a signal. Our biggest problem was given by differences in walking speed. In order to compare walks of different speeds, we had to find a way to average out walk cycles. We are able to identify individual walk cycles and we even know with which leg the cycle begins. For purposes of comparison we need basically just one cycle – i.e. two steps – as these are typical in every person.
Is it possible to identify a person who is running rather than walking?
No. In the act of running the characteristics of your gait change, even if you put tacks in your shoes. It’s like a robber wearing a mask: you don’t recognize his face. We can identify a walk but not a run. But we assume that a person in the act of committing a crime will not wish to attract attention so will try to act naturally and move normally. And if we search for such a person on the basis of several characteristics, we increase our chances of identifying and thus finding him.
You focus on the use of characteristics of gait for purposes of security. Can they be used in other areas, too?
I’m not much concerned with the use of biometrics in other situations, but I know that gait can be analysed in sport to regulate running. It is also used in physiotherapy and in medicine for the detection of motor pathology.
How would you specify your field? What is the main focus of your work?
I started off as a mathematician. I have a Master’s degree in IT Security, having specialized in cryptography. And now my work is dedicated to biometric curves. I try to find the most effective ways of comparison in given contexts, for instance that of gait.
As a security analyst, what inspired you to join the group of experts that focuses on forms of searching?
It came about by happy coincidence. I decided to study in Brno because I believe that it is the best place for informatics in the former Czechoslovakia. As a first-year student I was excited by mathematics and became interested in mathematical informatics. I found the topic of my Bachelor’s dissertation – random vector generation – in the Laboratory of Searching and Dialogue, and it was there that I went on with a Master’s thesis focusing on human recognition. I like practical applications for mathematics.