A Systematic Approach to Urban Building Extraction from High Resolution Remote Sensing Images

Abstract

Remotely Sensed space based High-resolution data inputs are widely used to extract urban building information. This information consists of the footprint of the building represented by the building boundaries and its geographical position on the ground. These maps are one such important information used by decision-makers for applications such as digital taxation, Land Information System, utility and transportation planning, traffic, Environmental Impact Analysis, Disaster Management, Crime Analysis etc. newline newlineTraditionally, the images from High-Resolution (HR) satellites and their derived products, such as DEM (Digital Elevation Model) and NDVI (Normalized Differential Vegetation Index), are used to achieve this goal, with limited success. Subsequently, many researchers achieved superior results by adopting specialized algorithms/methods, using the Airborne LiDAR (Light Detection and Ranging) data collected from aircraft platforms in conjunction with satellite images; due to the inherent advantage of LiDAR data having high fidelity offering the highest elevation accuracy with closely spaced/ dense grid points of elevation. However, the limitations of Airborne LiDAR data such as longer acquisition time lines and cost; associated data registration problems due to the usage of multiple data sets and the requirement of complex processing software pose constraints on its adoption to cover larger areas in a quicker time frame. Hence, through this research, a systematic method/algorithm is developed with a single dataset (using less resources) confining to HR satellite data only (covering larger areas) but performing on par with using high-fidelity LiDAR data. This algorithm is developed by incorporating the improvements to the already existing building detection methods used by previous researchers by (a) using a curvelet-based pan sharpening method (b) generating synthesized per pixel DSM (equivalent to LiDAR DEM) using a semi-global matching method and robust DTM height point filter using newline newline