Deep learning and sensor fusion methods for studying gait changes under cognitive load in males and females

Abstract

Human gait is the manner of walking in people. It is influenced by weight, age, health condition or the interaction with the surrounding environment. In this work, we study gait changes under cognitive load in healthy males and females, using machine learning methods. A deep learning model with multi-processing pipelining and back propagation techniques, is proposed for cognitive load gait analysis. The IMAGiMAT floor system enabling sensor fusion from plastic optical fiber (POF) elements, is utilized to record gait raw data on spatiotemporal ground reaction force (GRF). A deep parallel Convolutional Neural Network (CNN) is engineered for POF sensors fusion, and gait GRF classification. The Layer-Wise Relevance Propagation (LRP), is applied to reveal which gait events are relevant towards informing the parallel CNN prediction. The CNN differentiates between males and females with 95% weighted average precision, and cognitive load gait classification with 93% weighted average precision. These findings present a new hypothesis, whereas larger dataset holds promise for human activity analysis.