Temporal pattern recognition in gait activities recorded with a footprint imaging sensor system

This 2016 IEEE Sensors Journal paper uses a plastic-optical-fiber footprint floor sensor to tell different manners of walking apart from time-domain signals. Temporal features and a range of machine-learning models are compared.

“Temporal pattern recognition in gait activities recorded with a footprint imaging sensor system” is a paper by Omar Costilla-Reyes, Patricia Scully, and Krikor B. Ozanyan of the University of Manchester, published in the IEEE Sensors Journal in 2016. An earlier version had been presented at the IEEE SENSORS 2015 conference in Busan, South Korea, where it received the Best Student Paper Award in optical sensing applications. The work was supported in part by the U.K. Engineering and Physical Sciences Research Council and by CONACyT, Mexico’s national science council.

The sensor is an unobtrusive floor system built from plastic optical fiber. When a person steps on it, the fibers deform slightly and transmit less light, and the changes are used to form an image of the footprint over time. The paper asks whether the time-domain signal alone, meaning how the sensor readings rise and fall over time, is enough for machine learning to tell different ways of walking apart. Participants performed ten manners of walking chosen to change the amplitude and frequency of the temporal signals. The analysis designed several temporal features, including one called the adjacent mean, and evaluated them in linear, nonlinear, ensemble, and deep-learning models using cross-validated accuracy, in which each model is repeatedly trained on one part of the data and tested on another.

A random-forest model using the adjacent-mean feature achieved the best performance among the models compared. The authors conclude that the floor sensor can detect changes in gait through pattern recognition applied in the time domain, which supports applications ranging from healthcare to security. In the study, the time-domain signal of the sensor distinguished the manners of walking that participants performed.

The paper was the earliest manuscript of Dr. Costilla Reyes’s PhD thesis and the basis of a talk, “Pattern recognition in gait activities using a floor sensor system,” to the University of Manchester’s Data Science Club in August 2016. It is also cited in the smart-carpet poster shown at the EuroScience Open Forum 2016. In April 2017 the IEEE Sensors Council wrote to the authors that the paper was among the journal’s most-downloaded papers in the first quarter of that year.