Authors: Shabnam Jabari and Yun Zhang
Detection of buildings in urban and suburban areas using very high resolution satellite (VHR) images is a challenging task due to different illuminations on different sides of the same building roof. This causes different brightness values on a single roof. Therefore, in the segmentation step of an object-based classification, different sides of a roof will be assigned to different segments due to their intensity variation. To solve this problem, the majority of the studies merge the building roof segments together based on elevation information. However, because of the uncertainty of the borders in elevation layers as well as misregistration between the spectral and elevation layers, building boundaries usually cannot be detected precisely. In this study a novel method which utilizes IHS color transform is used to overcome the problem of intensity variation to segment each building roof as one segment. Then, the elevation information from LiDAR data is used to differentiate roof segments from other segments.
The proposed method is tested on the QuickBird satellite imagery of Fredericton, Canada. The achieved results are quite promising with an overall accuracy of more than 90% for building detection. Considering the off-nadir situation of imagery and consequently miss-registration between the elevation data and the image, the produced unified color segments are of great benefit for precise building boundary detection.
You are reviewing the
Download.
Your machine is locate at 3.143.5.133.
