Efficient On-Board Processing of Oblique UAV Video for Rapid Flood Extent Mapping
The Temporal Token Reuse (TTR) framework accelerates video segmentation from oblique UAV footage by exploiting spatiotemporal redundancy. It treats image patches as tokens and uses a lightweight similarity metric to identify static areas, reusing precomputed features instead of recomputing everything. This results in a 30% reduction in inference latency on edge hardware with minimal loss in accuracy, enabling real-time flood mapping.