Two-stage method improves gait recognition accuracy across views
Two-Stage Multi-View Gait Recognition with a Re-Embedding Network
Computer Vision and Pattern Recognition
Summary
Recognizing people by the way they walk is hard because cameras see them from different angles, which can confuse systems. The authors propose a two-step approach: first, they use a simple network to capture gait information from each camera view, then they use a transformer to combine these views flexibly. This helps the system learn better and recognize people more accurately, even when they carry bags or wear coats. Their method works well on large public datasets with many subjects and different conditions.
What this means in practice
- •For security system developers: Design gait-based biometric systems that accurately identify people from multiple camera angles and under clothing variations for enhanced security checks.$Commercial implications: Enables commercial biometric security products that remain reliable despite viewpoint changes and disguises, improving user authentication.
- •For surveillance system engineers: Integrate flexible multi-view gait feature fusion to improve person identification in public safety monitoring across varied camera views.
Authors
Long Hoang Le, Trung Thanh Ngo
Abstract
Gait recognition always remains challenging due to severe overfitting and the rigid view constraints common in single-stage approaches. We propose a two-stage framework, termed Translate-First-Then-Reason (TFTR), to address these issues. In the first stage, a shallow Siamese convolutional network with triplet loss maps Gait Energy Images (GEIs) into a 128-dimensional view-specific embedding space. In the second stage, these per-view embeddings are treated as tokens and processed by a 12-layer Transformer encoder, which re-projects them into a new space with improved cosine separability. This design enables flexible fusion of an arbitrary number of views at inference, overcoming the fixed-input limitations of prior methods. Trained on the OU-MVLP dataset (6,000 subjects) and evaluated on unseen CASIA-B across normal, bag-carrying, and coat-wearing conditions, our pipeline achieves 96.91\% single-view and 99.49\% three-view accuracy on OU-MVLP, and attains 100\% accuracy on CASIA-B with three views.