Identifying Abnormal Gait in Older People during Multiple-Tasks Assessment with Audio-Visual Cues

Abstract

This research presents a feasibility to adopt a decision support system framework as a rehabilitation and assessment tool for supporting the physiotherapist in identifying the abnormal gaits of older people. The walking movement was captured by the Microsoft Kinect cameras in order to collect the human motion during 4-meters clinical walk test. 28 older adults participated in this research and perform their gait in front of the affordable cameras. To distinguish an abnormal gait with balance impairment from those of healthy older adults, two machine learning algorithms; ANN and SVM, were selected to classify the data. Experimental results show that SVM achieves the best performance of classification with 82.14% of accuracy, in single-task and double-task conditions, when compared with the standard clinical results. However, SVM cannot achieve an acceptable performance when classifying triple-task condition, achieving only 71.42% of accuracy. As a comparison, ANN delivers only 75.00% of accuracy, which is inferior to SVM. This study show that SVM can be considered as a rehabilitation measuring tool for assisting the physiotherapist in assessing the gait of older people.