This paper introduces THC-VPR, a framework that leverages the temporal consistency of flight altitude to enhance recognition accuracy in multi-altitude aerial scenarios. It includes a height-aware temporal smoothing module, a learnable altitude transition model, and a hierarchical matching strategy.
Key findings
THC-VPR improves recognition accuracy in multi-altitude aerial scenarios.
The framework introduces a height-aware temporal smoothing module.
A learnable altitude transition model captures physical constraints of aerial platform dynamics.
A hierarchical matching strategy combines frame-level, sequence-level, and height-level cues.
Limitations & open questions
The paper is a research proposal and does not include experimental results.
The effectiveness of THC-VPR is yet to be validated on real-world UAV datasets.