Feature Aggregation With Reinforcement Learning for Video-Based Person Re-Identification.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Video-based person re-identification (re-id) is crucial for matching individuals across different camera views.
• Current methods often use simple averaging or recurrent neural networks for feature aggregation, which can be suboptimal.

Purpose of the Study:

• To develop an intelligent feature aggregation method for video-based person re-identification.
• To improve re-id accuracy by selectively aggregating frame features.

Main Methods:

• A reinforcement learning agent was trained to act as a decision-making process.
• The agent identifies and discards irrelevant frames during feature aggregation.
• This selective aggregation generates a more robust track feature.

Main Results:

• The proposed method significantly boosts person re-identification accuracy.
• Experimental results on benchmark datasets demonstrate clear improvements over state-of-the-art models.
• The approach effectively avoids noisy information by retaining valuable frames.

Conclusions:

• Reinforcement learning offers an effective strategy for intelligent feature aggregation in video person re-id.
• The developed method enhances the performance of existing state-of-the-art re-id models.
• Selective frame aggregation is key to improving robustness and accuracy in person re-identification.