Human gait recognition using orthogonal least square as feature selection

Abstract

<span>This study investigates the potential gait features that are related to human recognition using orthogonal least square (OLS). Firstly, video of 30 subjects walking in oblique view was recorded using Kinect. Next, all 20 skeleton joints in 3D space were extracted and further selected using OLS. Additionally, SVM with linear, polynomial and radial basis function (RBF) kernel was used to classify the selected features. As consequences, OLS was proven to be able to identify the significant features using all three kernels of SVM since all recognition accuracy attained is higher as compared to the original gait features. Results attained showed that the highest recognition accuracy was 90.67% using 48 skeleton joint points for SVM with linear as kernel, followed by 46 skeleton joint points for SVM with RBF kernel namely 88.33% and accuracy of 86.33% for 38 skeleton joint points using polynomial kernel.</span>