Multi-level features fusion and selection for human gait recognition: an optimized framework of Bayesian model and binomial distribution

Abstract

A biometric classification system is utilized to judge the features of human expression by recognizing distinct parameters. Human Gait Recognition (HGR) is a current research area which is mostly used for various security applications such as video surveillance etc. HGR is also utilized in medical imaging for the investigation of several diseases such as Parkinson disease which is identified by gait features. Still, various challenges occur in this domain that affects system accuracies such as shoe type, change in angle, load carriage and change in walking speed. In this research, a new approach for HGR is proposed which is based on Quartile Deviation of Normal Distribution (QDoND) for human extraction and Bayesian model along with Binomial Distribution for features fusion and best features selection. Initially, in the pre-processing step, the most excellent channel is selected and its motion flow is estimated. The motion regions are extracted by QDoND that are later utilized for shape and texture feature extraction. Afterward, the extracted features are fused by a Bayesian model based on their similarity index. Finally, BDs based best features are selected and recognition is performed on the basis of best features using multi-class support vector machine. Four publicly and famous datasets are utilized for the evaluation of proposed system such as AVA multi-view gait (AVAMVG), CASIA A, CASIA B and CASIA C having an accuracy rate of 100%, 98.8%, 87.7%, and 91.6% respectively. The results reveal that the proposed method outperforms in contrast to existing methods.