Condition-Aware Comparison Scheme for Gait Recognition.

Area of Science:

• Computer Vision
• Biometrics
• Machine Learning

Background:

• Gait recognition is crucial for human identification but is challenged by variations in camera viewpoints and clothing.
• Different body parts provide varying recognition cues under diverse conditions.

Purpose of the Study:

• To develop a condition-aware gait recognition system that enhances similarity measurement.
• To improve robustness against confounding factors in gait data.

Main Methods:

• A novel 'Instructor' module for condition-aware similarity comparison.
• A geometry-guided data augmentation technique ('Dresser') for handling dressing variations.
• Modeling temporal local gait information from coarse to fine granularity.

Main Results:

• The proposed method achieves state-of-the-art performance on the CASIA-B and OULP benchmarks.
• Significant improvements were observed in cross-condition gait recognition scenarios.

Conclusions:

• The condition-aware approach effectively addresses challenges in gait recognition.
• The proposed methods enhance gait representation and similarity measurement for improved accuracy.