Expression and automatic recognition of exhaustion in natural walking

Besides their function, human body movements express ones personality, intention and emotions, and give cues about a person's condition. This work focuses on the expression of exhaustion during natural walking. The gait of 14 participants was recorded using 3d optical tracking. Physical exhaustion was induced by performing full-body exercises at a rowing ergometer. A student's t-test analysis of predefined parameters like ankle stroke, range of motion (ROM) of human joints and center of gravity (COG), revealed that, first, there exist significant changes between normal and exhausted gait patterns and, secondly, the expression of exhaustion differs strongly among subjects. The same data sets were analyzed with techniques from machine learning to investigate if automatic recognition of an exhausted gait is possible. Principle Component Analysis (PCA) and Fourier Transformation were applied to the data set for feature extraction. Linear Discriminant Analysis (LDA), Naive Bayes, K-Nearest Neighbor Clustering (KNN) and Support Vector Machine (SVM) were compared for classification. Classification of exhaustion was achieved with various classifiers, but recognition of an unknown gait is challenging. Without features standardized to normal gait, recognition above chance was accomplished only with K-Nearest Neighbor Clustering.