151. Walking pattern analysis and SVM classification based on simulated gaits.
- Author
-
Mao Y, Saito M, Kanno T, Wei D, and Muroi H
- Subjects
- Computer Simulation, Diagnosis, Computer-Assisted methods, Humans, Reproducibility of Results, Sensitivity and Specificity, Artificial Intelligence, Gait physiology, Image Interpretation, Computer-Assisted methods, Leg physiology, Models, Biological, Pattern Recognition, Automated methods, Walking physiology, Whole Body Imaging methods
- Abstract
Three classes of walking patterns, normal, caution and danger, were simulated by tying elastic bands to joints of lower body. In order to distinguish one class from another, four local motions suggested by doctors were investigated stepwise, and differences between levels were evaluated using t-tests. The human adaptability in the tests was also evaluated. We improved average classification accuracy to 84.50% using multiclass support vector machine classifier and concluded that human adaptability is a factor that can cause obvious bias in contiguous data collections.
- Published
- 2008
- Full Text
- View/download PDF