Title :
An improved road and building detector on VHR images
Author_Institution :
Inst. fur Inf. VI, Tech. Univ. Munchen, Garching, Germany
Abstract :
A method is proposed for building and road detection on VHR multispectral aerial images of dense urban areas. In order to exploit all available information both spatial and spectral features of segmented areas are classified, using a 3-class SVM. Geometrical object features improve the classification accuracy in the difficult case where many building roofs are grey like the roads. In order to exploit more deeply spatial information, a road network regularization based on straight segment detection is suggested.
Keywords :
image classification; image resolution; object detection; roads; support vector machines; SVM; VHR multispectral aerial image; building detector; dense urban area; geometrical object feature; road detector; road network regularization; spatial feature; spectral feature; straight segment detection; Accuracy; Buildings; Hyperspectral imaging; Image segmentation; Roads; Support vector machines; Multiclass support vector machine; classification map regularization; data merging; mean shift; very high spatial resolution image;
Conference_Titel :
Geoscience and Remote Sensing Symposium (IGARSS), 2011 IEEE International
Conference_Location :
Vancouver, BC
Print_ISBN :
978-1-4577-1003-2
DOI :
10.1109/IGARSS.2011.6049176
