Cross View and Cross Walking Gait Recognition Using a Convolutional Neural Network

In this paper, we propose a gait recognition method using a convolutional neural network (CNN). A CNN architecture is designed and trained to learn an efficient representation with which walking patterns i.e., gait can be disentangled from the visual appearance of the subjects caused by covariate factors such as variation in view angles, clothing and carrying conditions. Since dynamic areas contain the most informative part of the human gait and are insensitive to changes in various covariate conditions, we feed the gait entropy images as input to CNN model to capture mostly the motion information. The learned gait features from CNN are then fed into a K-NN classifier to identify individuals based on their unique gait patterns. Experiments are carried out for cross-view and cross-walking gait recognition using the CASIA-B dataset. Our experimental results demonstrate the effectiveness of the proposed method.