Gait-Based Multi-View Person Identification with Convolutional Neural Networks

Due to the uniqueness of each person's stride, gait recognition is a promising method of identification and authentication. Machine vision systems have been developed to capture the individuality of a person's gait, however the variability introduced by elements like walking pace, viewpoint, clothing, and the presence or absence of accessories makes gait detection difficult and susceptible to improvement. One interesting application of gait-based person identification is in holographic telepresence systems for enhanced user identification where gait-based identification could provide an additional layer of identification and help distinguish between multiple participants. In this work we propose a system for person identification based on multiple views of human, walking in straight line. The decision of classifiers performing in each view are fused to deliver final decision about the person ID. The proposed approach is tested on two dataset. The first is a popular dataset for gait recognition and the recognition rate achieved is 93.27 %. The second dataset is recorded by the authors and the recognition rate achieved on this dataset is 95.83 %.