Skip to main content
Talk

Spatio-Temporal Gait Analysis: A deep learning approach

Dr. Omar Costilla Reyes presented “Spatio-Temporal Gait Analysis: A deep learning approach” at Innovation Match MX in May 2017. The event was the Second International Forum of Mexican Talent, in Mexico City.

Title slide and room during Dr. Costilla Reyes’s talk at Innovation Match MX, Mexico City, May 2017. Credit: Image courtesy of Dr. Omar Costilla Reyes.

Presenting at a forum for Mexican talent

Innovation Match MX, the Second International Forum of Mexican Talent, ran from May 31 to June 2, 2017, in Mexico City. Mexico’s Secretariat of Economy described it as a forum that brings Mexican talent from around the world together with Mexican companies and institutions, and CienciaMX reported its aim of strengthening links between innovative companies and Mexican researchers abroad, with funding bodies also taking part. Dr. Omar Costilla Reyes, then a doctoral researcher at the University of Manchester, presented his work there in May 2017.

Gait as a signal in space and time

His presentation, “Spatio-Temporal Gait Analysis: A deep learning approach,” brought machine learning and healthcare together for an audience of researchers, companies, students, and public institutions. Gait is the way a person walks. A floor sensor records it as pressure that changes both across the surface of the floor and from moment to moment, so each walk becomes a spatio-temporal signal. Deep learning, which uses layered neural networks that learn their own features from raw measurements, can model those signals without hand-designed gait measures. Gait also changes with age and with cognitive demand, which is why it is studied in research on movement and health.

Related work

The talk drew on research later published in “Deep neural networks for learning spatio-temporal features from tomography sensors” (2017), whose authors report very high accuracy in classifying a range of walking activities from raw floor-sensor data, and on his University of Manchester thesis on gait analysis for healthcare and security. Learning from the floor sensor’s full spatial and temporal signal, rather than from a few summary measures, is what gives the method its strength. It was his second appearance at the forum series, following an invited talk on machine learning in healthcare at the first International Forum of Mexican Talent in 2016.

Topics