Recognizing Gaits Across Views Through Correlated Motion Co-Clustering

Area of Science:

• Biometrics
• Computer Vision
• Pattern Recognition

Background:

• Human gait serves as a unique biometric for remote identification.
• Viewpoint changes significantly challenge gait recognition by altering visual features.
• Different gait components are variably affected by changes in viewing angle.

Purpose of the Study:

• To develop a novel method for robust gait recognition under varying viewpoints.
• To precisely describe relationships between gaits from different views by accounting for correlation variance.
• To improve the accuracy of remote human identification using gait biometrics.

Main Methods:

• A novel motion co-clustering technique is employed to group related gait segments from different views.
• Canonical Correlation Analysis (CCA) maximizes linear correlations within each co-clustered group.
• Gait information is projected across views using trained CCA subspaces for similarity measurement.

Main Results:

• The proposed method precisely describes inter-view gait relationships through multiple co-clustering groups.
• Canonical Correlation Analysis enhances the linear correlation between gait features across views.
• Experiments on standard gait databases demonstrate superior performance compared to existing methods.

Conclusions:

• The novel motion co-clustering and CCA approach effectively handles view variations in gait recognition.
• This method offers a more precise way to measure gait similarity across different viewpoints.
• The findings suggest a significant advancement in remote human identification using gait biometrics.