Overview of deep learning based pedestrian attribute recognition and re-identification.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Pedestrian attribute recognition (PAR) and re-identification (ReID) are crucial for intelligent surveillance and smart city applications.
• Deep learning has significantly advanced these fields, necessitating a comprehensive review of current methodologies.

Purpose of the Study:

• To review deep learning-based ReID methods.
• To analyze the relationship and differences between PAR and ReID.
• To provide insights into attribute-assisted ReID.

Main Methods:

• Summarizing key concepts in attribute-assisted ReID.
• Comparing datasets and algorithmic challenges in PAR and ReID.
• Introducing representative deep learning ReID methods, detailing network structures, loss functions, and training strategies.

Main Results:

• Analysis of cutting-edge ReID research, including solutions for cloth-changing, domain adaptation, occlusion, and resolution variations.
• Evaluation of state-of-the-art (SOTA) ReID methods' performance and characteristics.

Conclusions:

• Attribute-assisted ReID shows significant effectiveness.
• The review provides a foundation for future research in ReID and PAR for enhanced intelligent surveillance systems.