Person tracking and segmentation for human gait biometric system

摘要

Person tracking and segmentation in an unstructured environment provides an increasing demand to solve human identification problems. This paper addresses mixture of Gaussian (MoG) technique for statistically background modelling and robust human tracking method for deriving an intrinsic gait signature. The front and back leg angles are calculated from the sequence of extracted human motion silhouette frames which are being used as gait features. The training gait database is made with these extracted gait features for ten different training subjects. The principal component analysis (PCA) is applied on derived gait signatures which transforms the input features into a low dimensional feature space. The classification technique is followed by Baye's decision rule coupled with multivariate Gaussian distribution. The results are compared with k-nearest neighbour rule and minimum distance classification (MDC) techniques by accuracy and computational cost metric. The experimental verification has been performed on CASIA standard gait database. The Baye's classifier produces an encouraging classification result with minimum misclassification error rate.