Search

The paper presents results and analysis of a methodology for automatic building extraction that integrates remote sensing sources and readily available geographic information system data (tax assessor parcel data). The methodology consists of a series of image processing and spatial analysis techniques and incorporates an initial simplification, parcel-based image partitioning procedure. The results showed great promise for such image partitioning in reducing the complexity of the building extraction procedure by localizing its search while maintaining the integrity of the structures (85 percent of buildings reside completely within parcels and the majority of the rest consists of very small portions outside the parcel boundaries). Herein, we discuss the results as they apply to three distinct building types (commercial, residential, and high-rise). The paper presents successful building extraction scenarios as well as challenges, difficulties and drawbacks of the process and identifies the underlying factors that impact the success of footprint extraction for each type of building. The results show that 83 percent of commercial buildings are successfully extracted with average footprint area discrepancy less than 10 percent, while challenges in high-rise and residential scenes are examined and further investigated. [ABSTRACT FROM AUTHOR]

Earthquakes cause massive loss of property and lives, and mitigating their potential effects requires accurate modeling and simulation of their impacts. Earthquake building damage modeling and risk assessment applications require accurate accounts of inventories at risk and their attributes such as structure type, usage, size, number of stories, shape, year built, value, etc. This paper describes the development of algorithms for automatically extracting and recognizing 2-D building shape information using integrated aerial imagery processing and Geographic Information Systems data. We use vector parcel geometries and their attributes to simplify the building extraction task by limiting the processing geography. Extraction is significantly improved by innovatively weighting the histograms. Extracted buildings are cleaned, simplified, and run through 2-D shape recognition routines that classify the footprint. We discuss reasons for successes and failures in both extraction and recognition. [ABSTRACT FROM AUTHOR]