Kernel Partial Least Squares Based Hierarchical Building Change Detection Using High Resolution Aerial Images and Lidar Data

Abstract

Map databases usually suffer from obsolete scene details due to frequently occurring changes, therefore automatic change detection has become vital. Recently, researchers have explored change detection by combining high resolution images with airborne lidar data to overcome the disadvantages of using images alone. However, multiple correlations between different features are usually ignored and false alarms will further depress the value of final detection result. In this paper, we propose an hierarchical framework for building change detection by fusing high resolution aerial images with airborne lidar data that provides elevation information. The kernel partial least squares (KPLS) method is introduced for dealing with feature correlations, and dimension reduction and pixel level change detection are conducted simultaneously in a single learning process. To address the relatively high false alarm rate, an object based post processing technique is proposed to further eliminate those pseudo candidates. All spectral, structural and contextual information are combined together in this step. Experimental results demonstrate the capability of our proposed method for building change detection.