Classification Of Pathologies Using A Vision Based Feature Extraction
Published 2017 · Computer Science
A lot of studies linking gait to different pathologies exists. However, few have addressed the automatic classification of such pathologies through computer vision. In this paper, a method to classify different gait pathologies is proposed. Using a smartphone camera, a sagittal view of the subject’s gait is recorded. This record is processed by a computer vision algorithm that extract different gait parameters. These parameters are then used to perform a classification between 5 types of gait: normal, diplegic, hemiplegic, neuropathic and parkinsonian. Using a standard smartphone camera allows to simplify the data capturing step making this method suitable for Ambient Assisted Living. The experiments performed show an accuracy rate of 74% with a hierarchical classifier using Support Vector Machine combining Gait Energy Images and legs angle time series. The accuracy is improved to an 80% by applying data augmentation techniques during test, i.e., obtaining one sample per gait cycle and then combining the results to provide a more robust classification of the entire record.