Automatic Matching and Three-Dimensional Reconstruction of Free-Form Linear Features from Stereo Images
Photogrammetric Engineering and Remote Sensing
Also, surfaces constitute an important layer of any GIS database. Abstract With the growing availability of high-resolution digital cameras, Automatic matching of free-form linear features in overlapping the need for automatic and reliable surface reconstruction from large-scale imagery over urban areas still remains to be a imagery is becoming urgent. Automatic surface reconstruction problem in both the photogrammetric and computer vision from large-scale imagery over urban areas remains an
... olved communities. Although there is a variety of algorithms that problem in spite of the many efforts in the photogrammetric have been developed to solve this problem, reliable results are and computer vision fields. The complexity of the input imagnot always guaranteed. Differences in illumination conditions, ery and the ill-posed characteristics of that problem are among relief displacement, and occlusions are some of the factors the major obstacles encountered by researchers. The traditional that make solving the matching problem more challenging. photogrammetric solution for surface reconstruction has three The deficiency of available techniques stems from the fact that basic steps. First, conjugate (matching) entities within overlapthey do not consider the perspective geometry of the imaging ping images are determined. The second step involves the system in the matching process. Moreover, it is usually determination of the relative orientation parameters (ROP) relatassumed that conjugate entities are almost exact copies of ing the two images of a stereo pair. Finally, matched entities are each other (this is rarely the case). The need for a reliable projected into the stereo model using the derived ROP in step 2. algorithm that can handle large-scale imagery over urban Solving the correspondence problem is the most difficult step areas is growing, especially with the increasing availability within the surface reconstruction process. Strategies described of high-resolution aerial imagery. In this research, a new in the photogrammetric and computer vision literature for approach for automatic matching and three-dimensional finding conjugate entities within overlapping images include reconstruction of free-form linear features from stereo images area-based matching, feature-based matching, and relational is proposed. The suggested strategy is based on The Modified matching (Ackermann, 1984; Grimson, 1985; Rosenholm, Iterated Hough Transform (MIHT) for Robust Parameter 1987; Schenk, 1999). Estimation. MIHT relies on the mathematical relationship Automatic matching of distinct points is common in the between conjugate entities (the coplanarity condition when photogrammetric community. It usually starts by searching for dealing with a stereo pair). As a result, it overcomes problems interesting points that satisfy distinctness, stability, invariance, arising from relief displacements and/or occlusions. Moreover, uniqueness, and interpretability criteria. An extensive body of MIHT simultaneously solves for the relative orientation research has dealt with extracting interest points from digital parameters as well as the correspondence between conjugate imagery (Moravec, 1977; Barnard and Thomson, 1980 ; Mikhail entities in a stereo pair. Matching ambiguities along and across et al., 1984; Paderas et al., 1984; Fö rstner, 1986 ; Fö rstner and conjugate epipolar lines are observed in the correspondence Gulsh, 1987; Hannah, 1988; Haralick and Shapiro, 1993) . Roux output from MIHT. Ambiguities along the epipolar lines stem and McKeown (1994) used epipolar geometry, and height and from the nature of the coplanarity model, which establishes orientation constraints to ensure the robustness of matched the correspondence between conjugate epipolar lines rather corner points. However, matched point features do not provide than conjugate points. On the other hand, ambiguities across semantic information that can be useful for the following the epipolar lines are expected due to the discrete cell size of image interpretation and surface reconstruction activities. the implemented accumulator array in MIHT. In this work, the High-level features (e.g., linear features) are more useful for ambiguities are resolved through several pruning techniques subsequent processes such as object recognition and surface such as correlation coefficient, edge connectivity, height reconstruction because they contain more information. There information, and epipolar geometry. Conjugate entities with has been significant research in the area of matching and threesub-pixel accuracy are finally used to reconstruct accurate and dimensional reconstruction of straight lines. In this type of reliable three-dimensional image-space discontinuities. work, preprocessing of extracted edges is essential for seg-Various experiments using large-scale imagery over urban menting and identifying straight lines prior to the matching areas proved the reliability and robustness of the proposed process. Several methods have been proposed to aggregate method. local edge information into more globally defined lines and to Surface models are important for various operations such as orthophoto generation, city modeling, and object recognition.