TWO-VIEW STRUCTURE FROM MOTION FOR BUILDING MODEL RECONSTRUCTION FROM VHR SATELLITE IMAGERY

3D information can be of a great use for different applications. The generation of point cloud (PCL) for building reconstruction from very high resolution (VHR) satellite imagery can be achieved in a variety of ways. Structure from Motion (SfM) generates 3D PCL from a set of 2D images based on the sensor information, the scene captured and the motion of the sensor. This study utilizes multiple space-borne images captured by Pelaides-1B VHR satellite as tri-stereo pairs. The key of this study is the generation of sparse PCL from VHR satellite imagery with two-view SfM by utilizing moving window operation. The generation of PCL from VHR satellite imagery can be accomplished in four major steps. Firstly, feature detection by utilizing multiple feature detector operators (MFDO) withing a moving window operation. Secondly, masking out features with Homography matrix and removing detected outliers. Thirdly, matching the remaining features. Finally, pose estimation and point triangulation to generate sparse PCL. This study emphasizes the importance of detecting robust features by applying MFDO as well as the number and quality of features. The applied two-view SfM algorithm is resource aware and an efficient solution for 3D reconstruction. Further, the incorporation of a moving window operation increases the number of effectives features whereas the masking enhances the quality of the SfM operation. The resulting sparse point clouds show great alignment and the target building is recognizable. The efficiency and resource awareness of our study lead to promising results for subsequent sophisticated applications.