This paper presents a novel two-view Structure-from-Motion (SfM) algorithm with the application of multiple Feature Detector Operators (FDO). The key of this study is the implementation of multiple FDOs into a two-view SfM algorithm. The two-view SfM algorithm workflow can be divided into three general steps: feature detection and matching, pose estimation and point cloud (PCL) generation. The experimental results, the quantitative analyses and a comparison with existing algorithms demonstrate that the implementation of multiple FDOs can effectively improve the performance of a two-view SfM algorithm. Firstly, in the Oxford test dataset, the RMSE reaches on average 0.11 m (UBC), 0.36 m (bikes), 0.52 m (trees) and 0.37 m (Leuven). This proves that illumination changes, blurring and JPEG compression can be handled satisfactorily. Secondly, in the EPFL dataset, the number of features lost in the processes is 21% with a total PCL of 27,673 pt, and this is only minimally higher than ORB (20.91%) with a PCL of 10,266 pt. Finally, the verification process with a real-world unmanned aerial vehicle (UAV) shows that the point cloud is denser around the edges, the corners and the target, and the process speed is much faster than existing algorithms. Overall, the framework proposed in this study has been proven a viable alternative to a classical procedure, in terms of performance, efficiency and simplicity.

Abstract. The automatic establishment of image relationship between oblique images can be a challenging task. Feature based image matching (FBM) establishes this relationship by detecting and matching corresponding feature points. A robust matching is beneficial for many tasks including reconstruction, mapping and localization. The need of automatic processing of remotely sensed data, like very highresolution (VHR) satellite imagery, increased over the years. Furthermore, green vegetation and water are changing the physical properties with respect to the amount of light emitted, season and pollution. In addition, with human interaction, the change in appearance increases. The Normalized Differential Vegetation Index (NDVI) is a well-known and studied index in order to detect healthy green vegetation. The Normalized Differential Water Index (NDWI) can help identify water areas in an image. They can be used to preliminarily segment images into different categories for later process. This study proposes a novel framework to explore the potential of feature based stereo matching for very high-resolution satellite imagery with segmentation. The proposed framework will perform first, image segmentation with NDVI and NDWI on stereo VHR satellite image pairs. Then, classification by threshold to detect healthy green vegetation, water and image frame. Features within these three classes are masked out and kept from being processed during the feature matching step. The idea is that, features in these classes are easy to cause miss matching because they are more prone to be affected by different image and environmental conditions in the stereo image pairs. As a result, the amount of miss matches can be reduced on average by 7.5%. Furthermore, the segmentation decreases the total amount of detected features by 13.71%, so that the processing time decreases. This study has successfully proven that segmentation can lead to improved stereo matching. In future, segmentation driven can be utilized by AI matching processes as well as traditionally photogrammetric or computer vision tasks.