CONICET | Buscador de Institutos y Recursos Humanos

(En proceso de edición)Active lower limb assistive devices are intended to contribute to human movementactivities such as locomotion. While some devices are capable to function among avariety of terrains (i.e. level surfaces, inclines, or stairs), wide instrumentation sets(combining mechanical and bio-signal data) have been usually applied to minimizeclassification errors for locomotion mode recognition, even if this potentially bias usercomfort. Hence, a selection of parameters for classification with limited sensor datacould represent a milestone on the development of comfortable lower-limb assistivedevices. The objective of this study was to find the best combination of parameters forthe offline classification of five locomotion modes based on data from discrete areas ona pressure insole. Three virtual force-sensing resistor sensors were located under heeland forefoot regions of six subjects without discernible gait abnormalities to study thepotential of kinetic signals for locomotion-mode recognition. Parametric tests wereused to compare the effect of two types of data, four time-windows and thecombinations within five data features on the classification error of linear discriminantanalysis classifiers for steady-state gait. Statistical comparisons were made usingclassification errors measured through five-fold cross-validation. Results showed thatthe best overall classifier had an average classification error of 7.85+-4.76%. Ourresults were in close resemblance with related studies which relied on larger sensorsets as their instrumentation strategies. This method presents an insight about the useof a reduced number of sensors for accurate locomotion mode classification on thedevelopment of comfortable lower limb assistive devices.