Image Orientation by Fitting Line Segments to Edge Pixels
The linear features are easily to be identified in the natural environment, such as roads, walls, or building edges, which gives them more potential than the point features. Usin g linear features for recovering image orientation is based on the constraint that, a 3D line as any vector connecting the perspective center with an arbitrary point on the 3D line must lie in the plane defined by the perspective center and the correspondi ng 2D lines (Zalmanson, 2000). If the line equations of a 3D line is known from existing maps, building models, or GIS data, then measuring any two points on the corresponding 2D line will provide a set of constraint equations, which is similar to the collinear equations formed by control points. Although the measuring points do not have to be identical on overlapped images, it still requires manual identification and measurement.
In this research, the interior orientation is derived from the camera calibration before the photogrammetric task. Our major concern is to recovering the exterior orientation by fitting projected line segments to extracted edge pixels without any manual measurements. Many linear features in the environment, such as a section of road or a piece of wall, could provide the 3D control line segments. The existing Constructive Solid Geometry (CSG) building model is a better source, for several line segments can be composed of their vertices simultaneously. These line segments are then projected onto the image with approximate exterior orientation for the fitting purpose. On the other hand, the edge pixels are extracted automatically by using edge- detecting algorithms. The image orientation now can be determined by varying the exterior orientation parameters so that the extracted pixels are optimally fit to the projected line segments. Least-squares Model-image Fitting (LSMIF) is applied to achieve optimal
fitting, the details will be described in the next chapter. Figure 1 illustrates the process.
Figure 1. The process of recovering image orientation by fitting line segments to edge pixels.
3. Least-Squaresmodel-Image Fitting
All of the edges of the CSG building model are projected onto the photo by collinear equations with approximate exterior orientation parameters for the fitting purpose. There are two ways to get approximate exterior orientation parameters: (1) given from other sources, for example, the Inertial Measuring Unit (IMU) and Global Positioning Sys tem (GPS); (2) calculated by manually adjusting the projected model to fit its image approximately. After the projection, all of the edges are evaluated by self- occlusion test to determine if it is visible. Take the box-like building model for example, the re are maximally 9 edges visible on a photo and others are occluded by the model itself. Even the edge passed the self-occlusion test, it is still possible occluded by other objects, like trees, building, or shadow. Unfortunately, the situation is difficul t to be detected automatically due to the lack of an efficient image interpretation algorithm, therefore, it still requires manually interaction to remove the
fully occluded line segments.
