A Grassmannian Approach to Address View Change Problem in Gait Recognition.

Area of Science:

• Biometrics
• Computer Vision
• Machine Learning

Background:

• Gait recognition is a valuable biometric when conventional methods are not feasible.
• Human locomotion is complex, affected by kinematics and external factors, posing challenges for gait recognition.
• View variation significantly degrades gait recognition performance.

Purpose of the Study:

• To develop an effective method for gait recognition that addresses the challenge of view variation.
• To enable accurate gait matching between query and reference sets despite differences in viewing angles.
• To create a practical gait recognition system that does not require recording angle or walking direction information.

Main Methods:

• Generating virtual views to standardize query and reference sets.
• Employing a multiview matrix representation combined with a randomized kernel extreme learning machine.
• Utilizing Grassmann manifold treatment for an end-to-end solution to view changes.

Main Results:

• The proposed method effectively compensates for view differences in gait recognition.
• The approach achieved superior performance compared to several state-of-the-art methods on benchmark datasets.
• Demonstrated successful multiview recognition in previously unconsidered scenarios.

Conclusions:

• The developed virtual view generation technique offers a robust solution for view-invariant gait recognition.
• This method significantly improves the practical applicability of gait recognition systems.
• The approach outperforms existing methods, highlighting its potential for enhanced biometric security.