Using aerial images for urban planning Meenal Pore
Results
Results The method detects too many small rooftops. This is due to repairs on tin roofs which are registered as separate roofs due to the edges created by having a new piece of roofing next to the older rusted rooftops The area of rooftops compared to groundtruth data from Open Street Map building polygons
Methodology Google Map images (Quickbird) Data sources LiDAR, Digital Surface Model (DSM) Google Map images (Quickbird) Drone imagery Building extraction Building classification (rooftop type and size) Other features(slum, vehicles, road, river, vegetation) Wealth index Approximation Image segmentation (Drone, Satellite, LiDAR) Image feature extraction Image classification
Scaling feature extraction from aerial images Ground truth data from manual tracing of images and ground surveys (Open Street Map) Drone data Used to extract rooftops over a larger area WARD LEVEL Roof map from drone data Generate map of rooftops for a larger area than can be manually mapped Satellite data Used to extract rooftops from low resolution data CITY LEVEL Roof map from satellite data Used to extract rooftops from low resolution data COUNTRY LEVEL