Hybrid Model for Gait Biometric Recognition Using Machine Learning

Abstract

Gait recognition basically focuses on the task of human identification by using identifying human beings at a gap cantered on the style they walk. The su-pervised classification methods like SVM, RVM, KNN, CNN are u sed for recogni-tion on general datasets CASIA, AVAMG and Self-generated dataset. Self-generated dataset of 20 male and 10 female is created by making a 60-sec video at one hun-dred twenty frames per sec and producing silhouettes using spatial-temporal sil-houette evaluation in which the shifting silhouettes of person's strolling outline are mapped with point and eclipse on joints and different body parts and are segmented from every picture sequence using a environment subtraction algorithm, produc-ing eigenspace. The current study proposes a simple however proficient hybrid gait recognition model in which the dimensionality of the input function space, time-differing separation signals obtained from a succession of outline pictures, is re-duced through applying principal component analysis (PCA), Adam optimizer is applied to lower-dimensional eigenspace for optimization, followed by SVM and CNN being used in parallel for classification. Extensive exp erimental results on outdoor image sequences demonstrate that the proposed algorithm has an encouraging recognition performance meeting the real-time requirements with relatively low computational cost.