Multilayer perceptron neural network classification for human vertical ground reaction forces

Author

Goh, K.L. ; Lim, K.H. ; Gopalai, A.A. ; Chong, Y.Z.

Author_Institution

Curtin Univ., Bentley, WA, Australia

fYear

2014

Firstpage

536

Lastpage

540

Abstract

In this paper, human motion classification using multilayered neural network is proposed to classify motion signal based on vertical ground resultant force (VGRF). VRGF readings were acquired using an instrumented treadmill. The work presented in this paper seeks to classify six activities i.e. standing to walking, walking, walking to jogging, jogging, jogging to running and running, based on the measured VGRF. The data set involved 229 healthy Asians aged between 20 and 24, yielding a total of 740 activity classes. All activities varied as a result of subjects´ desired speed. However, it was observed that the VGRF of the last five strides reaction forces was sufficient to achieve 83% classification rate for the training set and 73% for testing set. The influence of number of hidden neurons was also analyzed to obtain optimal classification performance.

Keywords

gait analysis; image motion analysis; learning (artificial intelligence); medical computing; neural nets; signal classification; VGRF measurement; VGRF-based motion signal classification; activity classification; hidden neuron; human motion classification; human vertical ground reaction forces; instrumented treadmill-acquired VGRF readings; jogging-to-running activity; multilayered neural network classification; optimal classification performance; perceptron neural network classification; standing-to-walking activity; training set classification rate; vertical ground resultant force; walking-to-jogging activity; Accelerometers; Force; Instruments; Legged locomotion; Neurons; Testing; Training;

fLanguage

English

Publisher

ieee

Conference_Titel

Biomedical Engineering and Sciences (IECBES), 2014 IEEE Conference on

Type

conf

DOI

10.1109/IECBES.2014.7047559

Filename

7047559