Gait-based human gender classification using 5/3 lifting based wavelet filters and principal component analysis

Researches about gait recognition systems have begun to spread with the increase of the amount of video data. Human gender can be estimated by using machine learning methods from gait data. In the present study, a human gender classification system is designed by using CASIA - B gait database and OU-ISIR Gait Database Large Dataset. The silhouettes were extracted from the gait videos, the features were extracted using 5/3 lifting scheme, the feature vectors were then classified using C4.5 decision tree classifier, the genders were obtained, and the system performance was evaluated. Results showed that by using the proposed method, human gender were classified with an accuracy of 97.98% on CASIA - B gait databases and 97.5% recognition rate on OU-ISIR Walk Database large Dataset. This study demonstrates that using gait data followed by proposed feature extraction methods, human gender can be successfully estimated.