A gait recognition method using L1-PCA and LDA

• Published in: 2009 International Conference on Machine Learning and Cybernetics Vol. 6; pp. 3198 - 3203
• Main Authors: Han Su, Zhi-Wu Liao, Guo-Yue Chen
• Format: Conference Proceeding
• Language: English; Japanese
• Published: IEEE 01.07.2009
• Subjects: Biometric; Biometrics; Computer vision; Cybernetics; Feature extraction; gait recognition; Humans; L1-PCA; LDA; Legged locomotion; Linear discriminant analysis; Machine learning; Principal component analysis; Scattering; width analysis
• ISBN: 9781424437023; 1424437024
• ISSN: 2160-133X
• DOI: 10.1109/ICMLC.2009.5212776

Gait is one of biometric technologies which can be identified at a distance or at low resolution. This paper proposes a gait recognition method using PCA based on L1-norm maximization and LDA. The gait pattern is described by the periodic sequence width images, which contain the static and dynamic gait feature. L1-PCA is adopted to represent these features and LDA is used to analyze and classify the features. L1-PCA tries to find projections through maximizing L1-norm and LDA tries to find the projective direction which minimize the within-class scatter of examples and maximize between-class scatter. L1-PCA and LDA can keep gait feature and reduce the dimension of the feature. The performance of our approach was tested on the gait databases. The result of experiment proves that our method is effective for the recognition of gait sequence which is lower image resolution and noisy data.