Building Inventory Compilation for Disaster Management: Application of Remote Sensing and Statistical Modeling

摘要

This report introduces a methodology to extract spatial, geometric and engineering attributes of buildings using single high-resolution satellite images. Rational Polynomial Coefficients (RPC) are used to generate three dimensional models of buildings showing height, footprint, and shape information. Geometric information defining the sensors orientation is used in conjunction with the RPC projection model to generate digital elevation models. The report describes how the location and height of a structure are extracted by measuring the image coordinates for the corner of a building at ground level and its corresponding roof-point coordinates, and using the relationship between image-space and object-space together with the sensors orientation. The implementation of the algorithm and its application to the City of London are described. In addition, a methodology based on a multinomial logistic model is developed to infer the marginal probability distributions of the structural type and occupancy of a building. The input parameters for the statistical model are derived from the three dimensional models reconstructed from the satellite imagery. Datasets collected for southern California are used to train the models and establish inference rules to predict the engineering parameters of the buildings in the region. The predictive capability of the model is shown through the computation of the marginal probability distribution for a sample building.