Crest
AI-powered aerial flood detection and monitoring — computer-vision pipelines flag inundation patterns from drone and satellite frames.
Problem
Flood response teams need fast situational awareness across watersheds where ground reports arrive late. Crest trains vision models to highlight standing water and expanding inundation from aerial imagery so planners can prioritize evacuations and relief routes.
What it covers
- Aerial frame ingestion with geospatial metadata
- Segmentation and detection heads tuned for flood water classes
- Temporal differencing to catch expanding inundation
- Python evaluation harness with precision/recall reporting
def score_frame(frame: np.ndarray, model: FloodDetector) -> FloodMask:
tiles = tile_image(frame, size=512, overlap=64)
masks = [model.predict(tile) for tile in tiles]
return stitch_masks(masks, frame.shape)