MS2P: A True Multi-View Satellite Stereo Pipeline without Rectification of Push Broom Images

In this study, we propose a new and true multi-view stereo pipeline (MS2P) to reconstruct surface model from push broom satellite images. The proposed method utilizes all unrectified multi-view stereo image at the same time, thus it is a true multi-view stereo (MVS) method. Using the rational polynomial coefficient (RPC) of the images, the ground-to-image geometry between all 3D cost volumes and multi-view images are defined. Then EnSoft3D, a true multi-view stereo algorithm, is used to iteratively compute and refine all height maps. In addition, VoteValue in the original EnSoft3D is adjusted to facilitate satellite images for more accurate height map generation. Experimental results show that an MVS method in the computer vision community can be used to reconstruct height map and surface model from multi-view unrectified satellite images. Based on experimental results using DFC19 dataset, the MSE and RMSE are about 0.572m and 2.09m in average.