Sensing and Data Analysis for Assessing Human Balance Ability

Abstract

This paper presents the development of a novel balance sensor capable of assessing human balance ability. This sensor can effectively detect the tread force distribution information under human feet with high resolution and frame rate while they are trying to balance themselves. The principle and performance of the sensor are discussed in the paper. We have used this sensor to capture tread force distribution variation videos from 6 categories of different human balancing actions. A novel 3D convolutional neural network is developed to classify these 6 actions. The high accuracy of video classifications have been achieved using 3D CNN. In addition, our preliminary experimental results demonstrates that applying deep learning method, this sensor is able to recognize different human motion patterns. Ongoing work is to exploit its potential use in neural disease diagnosis.