Human gait gender classification based on fusing spatio-temporal and wavelet statistical features

Gait recognition is one of the biometric recognition systems that do not require observed subject's attention and assistance. This paper proposes gender classification based on human gait. Gender is an important demographic attribute of people that can play a significant role in automatic gait recognition, the perception of gender determines social interactions. Humans are very accurate at recognizing gender from face, voice or the manner in which an individual walks. In our proposed technique we focus on using three different types of features; Spatio-Temporal Model, Leg Motion Detection, and Statistical Wavelet Model. These features have different characteristics to be used in gender recognition system based on gait recognition. For testing the performance of our method we used CASIA B gait database this paper proposes a way of testing the performance by selecting randomly equal subset of males and females then run the experiment repeatedly many times to cover the entire subjects in the database. This testing approach makes the achieved result more reliable compared with the existing approaches. Two different classification methods used in our proposal; K-Nearest Neighbors and Support Vector Machine. Our experimental results, of 96.47% classification rate in average, show that our approach is providing more trustworthy accuracy compared with the existent approaches.