JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Research on Person Re-Identification through Local and Global Attention Mechanisms and Combination Poolings

  • 0State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing 100876, China.
Sensors (basel, Switzerland) +

|

September 14, 2024

|

September 14, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

This study introduces MGNACP, a novel network for person re-identification. By integrating attention mechanisms and combination pooling, it significantly enhances accuracy in identifying individuals across different views.

Area Of Science

  • Computer Vision
  • Artificial Intelligence
  • Machine Learning

Background

  • Person re-identification (re-ID) is crucial for surveillance and security.
  • Existing methods often struggle to effectively combine global and local features.
  • Attention mechanisms and advanced pooling strategies offer potential for improvement.

Purpose Of The Study

  • To develop an advanced person re-identification network, MGNACP.
  • To enhance feature representation by incorporating attention mechanisms and combination pooling.
  • To improve the accuracy and robustness of person re-identification systems.

Main Methods

  • The proposed MGNACP network builds upon the Multiple Granularity Network (MGN).
  • Attention mechanisms (channel attention) are integrated into global and local feature branches (forming MGNA).
  • Combination pooling, blending max and average pooling, replaces single pooling methods in each branch.

Main Results

  • MGNACP achieved competitive results on the Market-1501 dataset, with mAP of 88.82% and top-1 accuracy of 95.46%.
  • Experiments validated the effectiveness of attention mechanisms in learning discriminative features.
  • Optimized combination pooling parameters preserved advantages of max and average pooling while mitigating disadvantages.

Conclusions

  • MGNACP demonstrates superior performance as a person re-identification network.
  • Attention mechanisms and combination pooling are key contributors to improved re-identification accuracy.
  • The proposed approach offers a promising direction for advancing person re-identification technology.

Keywords: