Skip to main content
Talk

Pattern Recognition in Gait Activities Using a Floor Sensor System

Dr. Omar Costilla Reyes presented a floor-sensor gait-analysis project at the University of Manchester’s Science Club #3 in August 2016. The session was hosted by the Data Science Club.

Dr. Costilla Reyes beside his poster on smart-carpet gait analysis at the EuroScience Open Forum 2016 in Manchester, July 2016. Related photograph from What Can Machine Learning Infer about the Way You Walk? Credit: Image courtesy of Dr. Omar Costilla Reyes; poster by Omar Costilla Reyes, Patricia Scully, and Krikor Ozanyan.

A talk at Science Club #3

In August 2016, Dr. Omar Costilla Reyes presented “Pattern recognition in gait activities using a floor sensor system” at Science Club #3, a Data Science Club session hosted in the Department of Computer Science at the University of Manchester. He was then a doctoral researcher in the university’s School of Electrical and Electronic Engineering, and the talk gave students and researchers interested in data science an account of how walking can be measured and analyzed with a floor sensor.

How a floor sensor records walking

Gait is the way a person walks, and it changes with the task being performed, such as walking quickly, slowly, or while carrying a weight. The Manchester floor sensor, called iMagiMat, works through plastic optical fibers: footstep pressure deforms the fibers, which changes the intensity of light traveling through them, so each step becomes a signal that can be recorded over time. Because the sensor is in the floor, nobody has to wear a device, which makes repeated, unobtrusive measurement possible. Pattern-recognition methods then compare the recordings and ask which characteristics of the signal separate one way of walking from another.

Research behind the talk

The presentation drew on “Temporal pattern recognition in gait activities recorded with a footprint imaging sensor system,” published in IEEE Sensors Journal in 2016. In that paper, participants walked in several different manners, and a range of machine-learning models were compared on features taken from the time-domain signal alone. A random-forest model, which combines the votes of many decision trees, performed best, with high cross-validated accuracy, showing that the time-domain signal of a single floor sensor carries enough information to tell the ways of walking apart. The paper was later listed among the journal’s most downloaded papers. Dr. Costilla Reyes also demonstrated the smart carpet to the public at the university’s Open Labs, asking what machine learning can infer about the way you walk, and a photograph of his ESOF 2016 poster on the same question accompanies this page.

Topics