Abstract
Gender classification plays an important role in many applications such as surveillance systems and medical applications. Most of approaches for gender classification are based on features of human face, voice and gait. Among these approaches, gait-based approaches are becoming more and more popular since the way that they collect human gait information is non-contact and non-invasive. In this paper, we propose a novel method to classify human gender using their gaits. For the gait-base gender classification, we collect silhouettes of human walking pattern from Microsoft Kinect sensor, and extract two main gait features, i.e., Gait Energy Image (GEI) and Denoised Energy Image (DEI) from a sequence of an entire cycle of walking silhouette images. GEI is an appearance-based gait representation while DEI is used to remove the noises from GEI. For the gait features, we use a low dimensional feature vector to represent the gait features. The extracted feature dataset are divided into two parts, i.e., training and testing datasets. The training data set are used for training a support vector Machine (SVM) classifier while the testing dataset are used for the evaluation. Figure 1 shows overall procedure of proposed gait-based gender classification system using Microsoft Kinect sensor. Despite of the limitation of the dataset, i.e., different races and thickness of clothes which weaken the distinct differences between males and females, the average accuracy of the proposed approach reaches up to 87% under 10-folds validation. According to the experimental results, we know that GEI is an applicable feature for human gait representation. Figure 1 General design of the system.