Towards Vulnerability Mapping on High Resolution Aerial Images: Roof Detection, GIS, and Machine Learning Techniques

Determining disaster-critical areas are essential when it comes to disaster risk reduction and response. One tool that can assist in this effort is vulnerability mapping. This study explores roof detection and segmentation using machine learning and geospatial information systems techniques in which target areas are characterized by urban density and spatial arrangement. This information can then be used to contextualize and quantify a certain area's vulnerability to natural and man-made disasters. In contrast to previous efforts of geospatial and mapping analysis which involves using satellite images, higher resolution aerial images of Philippine areas provided by the Aerial Imaging Laboratory of the Ateneo Innovation Center is used for this study. Results show that the proposed system can extract various information such as population estimation, urban density, and ultimately, disaster vulnerability by detecting roofs from aerial images. The system includes a disaster simulation capability to determine which specific regions are in highest risk and which of the population in the area are directly affected. Lastly, the proposed system is able to handle batch processing of aerial images that allow for fast response times and wider coverage for assisting in disaster response and rescue efforts.