Speed invariant gait recognition-The enhanced mutual subspace method.

Area of Science:

• Biometrics and Pattern Recognition
• Computer Vision and Image Analysis
• Machine Learning for Behavioral Analysis

Background:

• Gait recognition, a dynamic biometric, faces challenges with variations in walking speed.
• Traditional Mutual Subspace Method (MSM), effective for static biometrics like face recognition, requires adaptation for dynamic behavioral biometrics.
• Accuracy loss in MSM's Principal Component Analysis (PCA) step due to covariance matrix calculation hinders performance.

Purpose of the Study:

• To introduce an enhanced Mutual Subspace Method (eMSM) for robust gait recognition.
• To improve the accuracy and discrimination capability of gait recognition systems under varying walking speeds.
• To adapt MSM, a method successful in static biometrics, for dynamic behavioral biometrics.

Main Methods:

• Developed an enhanced Mutual Subspace Method (eMSM) by adapting MSM for gait recognition.
• Employed a 2D PCA-based mutual subspace to mitigate accuracy loss in covariance matrix calculation.
• Integrated image rotation across multiple angles and feature fusion using a boosting method to enhance discrimination.

Main Results:

• The eMSM methodology demonstrated superior performance compared to state-of-the-art methods on gait databases with variable walking speeds (CASIA-C, OU-ISIR).
• The combination of 2D PCA, rotation, and boosting effectively extracted richer gait features.
• Achieved enhanced robustness against variations in walking speed, a key challenge in gait recognition.

Conclusions:

• The proposed eMSM methodology offers a significant advancement in gait recognition accuracy and robustness.
• The study validates the effectiveness of combining 2D PCA, rotation, and boosting for dynamic biometric analysis.
• Future research can explore other combinations of operations to further optimize MSM for behavioral biometrics.