View Transformation Model Incorporating Quality Measures for Cross-View Gait Recognition

Area of Science:

• Computer Vision
• Biometrics
• Pattern Recognition

Background:

• Cross-view gait recognition faces accuracy degradation due to differing observation angles.
• Existing view transformation models (VTMs) can produce biased dissimilarity scores, limiting recognition performance.

Purpose of the Study:

• To enhance cross-view gait recognition accuracy by addressing biased dissimilarity scores.
• To propose a VTM integrated with a score normalization framework using quality measures.

Main Methods:

• Developed a VTM that encodes a joint subspace of multi-view gait features.
• Incorporated quality measures to quantify score bias.
• Calculated posterior probabilities using quality measures and dissimilarity scores for recognition.

Main Results:

• The proposed VTM with score normalization and quality measures demonstrated improved accuracy.
• Accuracy enhancements were observed across various cross-view settings in gait recognition.
• Evaluated on large-scale over-ground and treadmill gait datasets.

Conclusions:

• Integrating quality measures and score normalization into VTMs is effective for improving cross-view gait recognition.
• The proposed method offers a robust solution for handling view variations in gait analysis.
• This approach contributes to more reliable person authentication using gait biometrics.