In photogrammetry, the development phase of image matching is moving into a new era, namely the dense matching. At this phase, there are bottlenecks for blunder detection and quality evaluation of dense matching results, including (1) no output on photo coordinates of matching points, (2) a huge number of matching points, and (3) unstable bundle block adjustment caused by too close matching points. This paper proposes four easy and applicable methods to overcome those bottlenecks. They are visual check, relative orientation (RO) using a huge number of tie points, bundle block adjustment, and comparing with check points. Four matching algorithms are tested, including SIFT-based Multi-image Matching (SMM), Structure from Motion (SfM), DAISY and Semi-Global Matching (SGM). The results show that 50.72% and 47.00% of wrong matching points appear at roof ridge lines and break lines. Only 0.05% of those wrong matching points are located in homogeneous color area. This might be for reference for development and application of image dense matching. Test results of RO method demonstrate that SMM and SGM have the wrong matching rate of 2.82% and 2.36%, respectively. The evaluation results of bundle block adjustment method show that the matching accuracy and wrong matching rate of SMM are 0.23pixel and 3.97%, respectively. Both RO and bundle block adjustment methods have the computation speed of 14,984 points/second and 292 points/second, respectively. By comparing 12 check points, the object points measured by SGM have the absolute elevation differences with maximum 0.935GSD, minimum 0.006GSD, average 0.315GSD and RMSD 0.238GSD, where 1GSD=0.168m.

In recent years, there is an extensive use of computer vision and stereo vision techniques in various fields. Among them, stereo matching algorithm aims to reconstruct 3D models by calculating disparities of conjugate points in epipolar image pair. Based on the detailedness it would involve, the stereo matching can be divided into the following categories: Local algorithm, Global algorithm and Semi-Global Matching (SGM). When it comes to performing pixel-wise matching, the stereo matching technique can be utilized to generate dense point clouds of the interested scene. This study employs a SGM based stereo matching algorithm with a good trade-off between runtime and accuracy to generate dense point clouds. Cross-based Matching, a local algorithm is integrated into SGM to ease the high sensitivity of parameters chosen in penalty function. In addition, the ranges of parameters needed in carrying out the matching operation have been suggested to support practical as well as quality disparity estimation, and thus a satisfactory 3-D scene reconstruction.

Image-based semi-global matching (SGM) utilizes epipolar images as a geometrical constrain in image matching. However, the epipolar geometric of frame camera and push-broom satellite images are not the same. Therefore, the traditional image-based SGM cannot be applied to push-broom satellite images directly. This study proposed an object-based SGM (OSGM) to overcame the problem of non-linear epipolar line for push-broom satellite image. The aim of this study is to generate the digital surface model (DSM) for image orthoretification using proposed OSGM. This study also compare the DSM generated from feature-based matching and OSGM. The feature-based matching only performs the image matching on particular feature points. Then, the 3D coordinates of feature points are interpolated into DSM. The experimental images are Formosat-2 in-track stereo images, the number of 3D points generated from feature-based matching is 15% of OSGM. We use these two DSMs to produce orthoimages, the standard deviations of disparity between orthoimages are 3.07pixels and 1.94pixels, respectively. The DSM generated from OSGM show higher relatively accurate than the one from feature-based matching.

The Unmanned Aerial Vehicle (UAV) has been developed mature. It is both immediacy and convenience on the field of images shooting. The three-dimension information that close to real can be constructed via the images which taken from UAV. The utilizing on geospatial of UAV is construct complete three-dimension model from shooting a building surround and produce orthophoto from shooting topography vertical. There are good algorithms both on the field of set up relative relationship between ground and images and calculating camera parameters because of the well developing of relative technique; only on the field of the procedure of calculating dense point cloud has not been developed a suitable method. In order to investigate characteristics and applicability of different dense matching, two methods from different theoretical basis and both widely used are chosen in this study, including Semi-Global Matching (SGM) and Patch-based Multi-view Stereo (PMVS) from Global Method. We use these two methods to analyze the images taking from shooting a historical sites surround and campus vertical. The result can be a reference for construct three-dimension point cloud from the images taking from UAV.

Due to the rapidly urbanization development, to monitor the change of city environment is more and more important for urban land resource management. Different to traditional vertical aerial imagery (VAI), the oblique aerial imagery (OAI) is more stereoscopic for manually recognition, and has more information practically on building façade. Various studies has investigated the potential of its applications, such as building seismic damage assessment, building objects extraction and classification and so on. In this study, both vertical aerial imagery (VAI) and oblique aerial imagery (OAI) are collected from airborne and UAV platform in two different periods for building change detection purpose. Through photogrammetry techniques including orientation reconstruction and dense image matching, the colored high density 3D point clouds of two period data are generated from both vertical and oblique images. We apply object-based image analysis (OBIA) to classify the images into several classes of the cyber-city. Then, two period of point clouds with information of classes which is extracted from classified images are generated by back-projection and Hidden Point Removal opereator. To detect the change of building, two point clouds are normalized to local voxel coordinate system. Voxel-based change detection is performed to detect the building change in 3D space and calculate the volume.