Application of ultra-wide band radar for classification of human activities

The authors investigate the feasibility of classifying different human activities using ultra-wide band (UWB) radar. Eight human subjects performing eight different activities are measured using a UWB radar. The eight activities include walking, running, rotating, punching, jumping, transitioning between standing and sitting, crawling and standing still. The dimension of the UWB returns is reduced using principal component analysis (PCA). The time-varying UWB signatures are characterised within a time window through observing the variation of the PCA coefficients. A support vector machine (SVM) is used to classify the activities based on the signatures. A multi-class classification is implemented using a one-versus-one method. Optimal parameters for the SVM are found through a 4-fold cross-validation. The resulting classification accuracy is found to be more than 85%. The potential of classifying human activities with different ground planes and with cluttered environments is also investigated. To extract more information regarding the target motion, human walking style classification with the developed method is also discussed.