AA-RGTCN: reciprocal global temporal convolution network with adaptive alignment for video-based person...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Person re-identification (Re-ID) is crucial for retrieving individuals across different camera views.
• Video-based Re-ID leverages both spatial and temporal features, but existing methods suffer from redundant spatial descriptions and insufficient temporal modeling.
• Current approaches often ignore frame misalignment and rely on fixed-length sequences, limiting temporal representation.

Purpose of the Study:

• To develop a novel network architecture for video-based person re-identification.
• To address the challenges of frame misalignment and insufficient temporal feature modeling in existing Re-ID methods.
• To improve the accuracy and robustness of person re-identification in video surveillance.

Main Methods:

• Proposed the Adaptive Alignment-Reciprocal Global Temporal Convolution Network (AA-RGTCN).
• Introduced an Adaptive Alignment block to dynamically adjust frame positions for optimal temporal modeling.
• Developed a Reciprocal Global Temporal Convolution Network to capture robust temporal features across varying time intervals and directions.

Main Results:

• Achieved 85.9% mAP and 91.0% Rank-1 accuracy on the MARS dataset.
• Obtained 90.6% Rank-1 accuracy on the iLIDS-VID dataset.
• Reached 96.6% Rank-1 accuracy on the PRID-2011 dataset, outperforming state-of-the-art methods.

Conclusions:

• The AA-RGTCN effectively handles frame misalignment and models discriminative temporal representations.
• The proposed method demonstrates superior performance in video-based person re-identification compared to existing approaches.
• The Reciprocal Global Temporal Convolution Network contributes to robust temporal feature extraction for Re-ID tasks.