Person Reidentification via Multi-Feature Fusion With Adaptive Graph Learning

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Person reidentification (Re-ID) is crucial for surveillance but often relies on supervised methods requiring extensive labeled data.
• Supervised learning approaches for Re-ID face scalability challenges in real-world scenarios with abundant unlabeled data.

Purpose of the Study:

• To develop an unsupervised person reidentification model that overcomes the limitations of supervised learning.
• To enhance the scalability and applicability of Re-ID systems in practical surveillance networks.

Main Methods:

• Proposed a novel multi-feature fusion with adaptive graph learning model for unsupervised Re-ID.
• Incorporated multi-feature dictionary learning and adaptive multi-feature graph learning into a unified framework.
• Utilized an alternating optimization algorithm with proven convergence for model training.

Main Results:

• The proposed model achieves superior performance on four benchmark datasets.
• Demonstrated the effectiveness of integrating multi-feature learning and adaptive graph learning for accurate Re-ID.
• The learned dictionaries are discriminative, and the graph structure learning is accurate.

Conclusions:

• The developed unsupervised Re-ID model offers a scalable and effective solution for pedestrian identification in surveillance.
• The multi-feature fusion and adaptive graph learning approach significantly improves Re-ID accuracy without labeled data.
• This method provides a promising direction for advancing unsupervised learning in computer vision applications.