Gender Recognition Based on Gait Using Multi-View Fusion

ABSTRACT

Considering the problem of gait based gender recognition when gait information can be acquired from multiple views, this paper presents a detailed analysis on how to combine different views and proposes a fusion method derived from Bayesian theory. For feature extraction, a spatio-temporal gait representation is adopted and improved to reduce data redundancies. Then the class separability of each view angle is analyzed by using such features and the gender recognition rate is also computed under every single view. Next, three kinds of fusion scheme are designed to combine these different view angles for a comparison. Experiments are implemented on CASIA Gait Database (Dataset B) and the results demonstrate that the proposed fusion method achieves the superior recognition performance of 97.5% in large datasets.

Keywords: