Person Re-Identification with Attribute-Guided, Robust-to-Low-Resolution Drone Footage Considering Fog/Edge Computing

Area of Science:

• Computer Vision
• Artificial Intelligence
• Surveillance Technology

Background:

• Person re-identification (ReID) is vital for aerial surveillance using drones.
• Low-resolution drone footage significantly degrades ReID performance.

Purpose of the Study:

• To investigate the impact of low resolution on drone-based ReID.
• To develop a method for improving ReID accuracy under low-resolution conditions.

Main Methods:

• Utilized a synthetic dataset with controlled variations in drone altitude and distance.
• Employed an eXplainable Artificial Intelligence (XAI) framework to analyze resolution effects.
• Proposed a feature filtering method to retain robust attributes for low-resolution ReID.

Main Results:

• Achieved a 6.59% improvement in ReID accuracy at a 16% resolution scale on the Market1501 dataset.
• Demonstrated the effectiveness of filtering non-robust attributes in low-resolution scenarios.

Conclusions:

• The proposed method significantly enhances person re-identification accuracy in low-resolution drone imagery.
• The approach is effective for drone-based aerial surveillance, particularly within Fog/Edge Computing paradigms.