Pose-based Deep Gait Recognition.

Human gait or the walking manner is a biometric feature that allows to identify a person when other biometric features such as face or iris are not visible. In this paper we present a new pose-based convolutional neural network model for gait recognition. Unlike many methods considering the full-height silhouettes of a moving person, we consider motion of points in the areas around the human joints. To extract the motion information we estimate the optical flow between current and subsequent frames. We propose the deep convolutional model which computes pose-based gait descriptors. We compare different network architectures and aggregation methods. Besides, we experiment with different sets of body parts and learn which of them are the most important for gait recognition. In addition, we investigate the generalization ability of the algorithms transferring them from one dataset to another. The results of the experiments show that our approach outperforms the state-of-the-art methods.