Hideki Shimamura

A refined marker controlled watershed for building extraction from DSM and imagery

In this paper, we present a refined marker controlled watershed segmentation of the gradient of a digital surface model (DSM) to extract the buildings. To remove the effect of terrain, we first derive a digital terrain model (DTM) from the DSM by geomorphology filtering, and then generate a normalized DSM (NDSM). Since urban appearance is complex, building extraction is implemented through several stages. In the first stage, the large buildings are extracted according to the thresholds of the height and the area, to avoid being separated at watershed. Then the small buildings are segmented by watershed with markers. Finally, orthogonal imagery is used to remove the trees according to the colour principle. A case study is carried out on a DSM of a region of Kounan, Japan with a spatial resolution of 0.5 m. The experiment shows the effectiveness of this method.

Building change detection using aerial images and existing 3D data

The purpose of this research is to develop a practical system for building damage detection in dense urban areas after natural disasters such as earthquakes, typhoons, and tsunamis. A novel approach is presented that allows real-time building change detection using high resolution aerial images and existing 3D building data. The developed system contains two major processing modules: inflight processing and ground processing. Inflight processing module mainly extracts feature information from images acquired by an airborne digital camera, and compresses them in order to transfer to the ground. Ground processing module is responsible to analyze building changes by comparing the feature information to the existing 3D data. The automated change detection is performed by the following steps: firstly, edge features of all buildings in the aerial images are extracted on board of an aircraft; Then, the outlines of each building, which are extracted from the 3D building data, are projected onto the images based on the interior and exterior orientation parameters of the camera, and compared to the extracted edge features of each building in the images. The Portion Hausdorff Matching is applied to calculate the similarities between the projected outlines and the extracted edge features. Finally, collapsed buildings are detected using the calculated similarities. The maximal similarity between the edge features and the projected outlines in the direction of projection can be used to detect the change in building height. This approach is verified using an airborne digital camera UltraCamD image and corresponding 3D building data of Tokyo, Japan. The heights of some buildings in the 3D building data are manually modified in order to simulate building changes. The experimental results show that among 1320 buildings used in the test, 45 buildings are successfully detected as collapsed or removed. Among them, 9 buildings are correctly detected out of 10 collapsed buildings. Also, 2885 buildings are used in another test for detecting changes in building heights. 11 of the 27 buildings known to have height changes are successfully detected. The experimental results indicate that the proposed approach shows great potential for practical applications although still faces challenges from factors such as inaccurate 3D building data, occlusions due to high building density, and small changes in building height.

Morphological House Extraction from Digital Surface Model and Imagery of High-Density Residential Areas

Abstract High-density residential area modeling is extremely difficult because many houses reside close to or even touch each other; it usually causes the error of under segmentation. Our method solves this issue through scale detection and dome reshaping. A modified granulometry using opening-by-reconstruction instead of opening is proposed to detect the principle scales of the buildings. A morphological filtering algorithm at the detected continuous scales is developed to decompose a house into reshaped slices, which will be reconstructed as a dome. Thus, the domes are separate despite the joined houses, and are used to extract the markers of the houses. Finally, processes based on markers are developed to segment and model the houses. Among which, the multiple spectral image is used to detect the trees in the scene. Compared with other extraction methods, our technique decreases the fragment rate from 7.4 percent to 0.9 percent, the mixing rate from 14.8 percent to 0.9 percent.