Temporal pattern recognition for gait analysis applications using an "intelligent carpet" system

We report on the demonstration of a novel floor sensor system for gait analysis in the time domain. The ability of the system to detect changes in gait was evaluated using pattern recognition techniques. The selected machine learning models successfully classified 10 different walking manners performed on the floor sensor system. Their range was defined in terms of the amplitude, frequency and type of the temporal signal. Between three and five consecutive footsteps were captured per gait experiment. For the data analysis five machine learning time series features were engineered for assessment of 12 machine learning models. The tested machine learning models includes linear, non-linear and ensemble methods. The top F-score performance obtained was 88% using a finely tuned Random Forest model. We conclude that pattern recognition in gait activities monitored by the floor sensor system is suitable for gait analysis applications, ranging from biometrics to healthcare.