Kinect-Based Frontal View Gait Recognition Using Support Vector Machine

This paper investigated the most suitable multi-class support vector machine (SVM) coding design in recognising human gait based on frontal view that include one-versus-all (OVA), one-versus-one (OVO), error correcting output codes (ECOC), ordinal, sparse random and dense random algorithms. Firstly, walking gait of 30 subjects is captured using Kinect sensor. Next, all 20 skeleton joints within the full gait cycle are extracted as input features. Further, the gait features acted as inputs to the SVM classifier, specifically using linear kernel with various coding design algorithms are evaluated and tested in determining the most optimum results in recognition of human gait based on frontal view. Result proven that one-versus-all (OVA) attained the highest accuracy, specifically 96%.