Investigating the Use of Autoencoders for Gait-based Person Recognition

摘要

In recent years, gait has been growing as a biometric for person recognition at a distance. However, factors such as view angles and carrying conditions often make this task challenging. This paper proposes a solution to this problem by modelling gait sequences using Gait Energy Images and then using sparse autoencoders to extract their features for recognition under different view angles. Experiments were carried out on the challenging CASIA B dataset, resulting in outstanding accuracy rates.