Neural Network Gait Classification for On-Body Inertial Sensors

Abstract

Clinicians have determined that continuous ambulatory monitoring provides significant preventative and diagnostic benefit, especially to the aged population. In this paper we describe gait classification techniques based on data obtained using a new body area sensor network platform named TEMPO 3. The platform and its supporting infrastructure enable six-degrees-of-freedom inertial sensing, signal processing, and wireless transmission. The proposed signal processing includes data normalization to improve robustness, feature extraction optimized for classification, and wavelet pre-processing. The effectiveness of the platform is validated by implementing a binary classifier between shuffle and normal gait. Artificial neural networks and classifiers based on the Cerebellar Model Articulation Controller were tested and yielded classification accuracies (68%-98%) comparable to previous efforts that required more restrictive or intrusive apparatus. These results suggest a viable path to resource-constrained, on-body gait classification.