A Multi-scale Spatial-temporal Attention Model for Person Re-identification in Videos

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Person re-identification (re-ID) is a challenging task in computer vision, crucial for surveillance and security.
• Traditional methods often struggle to capture complex spatial-temporal relationships within video data.

Purpose of the Study:

• To develop a novel deep neural network attention model for learning representative local regions in video sequences for person re-identification.
• To enhance the accuracy and robustness of person re-identification systems by focusing on salient spatial-temporal features.

Main Methods:

• Proposing a multi-scale spatial-temporal attention (MSTA) model to analyze video frames at various scales.
• Integrating spatial and temporal attention mechanisms to exploit the importance of local regions within the entire video context.
• Designing a hybrid training strategy combining image-to-image and video-to-video modes.

Main Results:

• The MSTA model demonstrates superior performance compared to existing state-of-the-art methods on benchmark datasets.
• The model effectively captures and utilizes discriminative local regions across different scales and time instances.
• The proposed training strategy contributes to the overall effectiveness of the attention model.

Conclusions:

• The novel MSTA model offers a significant advancement in deep learning-based person re-identification.
• Exploiting multi-scale spatial-temporal features is crucial for accurate person re-identification in videos.
• The developed model provides a more robust and effective solution for real-world surveillance applications.