The proposed technique aims at the automatic location and measurement of Ground Control Points (GCPs) in aerial images using vertical terrestrial chips and considering large search spaces.
The availability of data derived by integrated Global Navigation Satellite System (GNSS) and Inertial Navigation System (INS) systems have been used for the direct sensor orientation (DSO) in photogrammetric processes. The correction of systematic errors and reliability assessment of direct georeferencing (DG) data still require GCPs in the integrated processes of determination of Exterior Orientation Parameters (EOPs).
These data enable the restriction of the image search space for a GCP when it is projected to the aerial images by collinearity equations. However, in some scenarios, the DG data are not available or are of low quality, such as in light systems carried by Unmanned Aerial Vehicles (UAV).
In this case, the projection of object coordinates to the image space can transfer them too far from their correct positions, and the search space can be large, which makes difficult to apply image matching techniques successfully. To solve this problem, the proposed approach uses a combination of matching techniques to restrict the search space and to locate a GCP in the aerial images.
A feature-based matching based on Scale-Invariant Feature Transform (SIFT) is used to find homologue points, as described by Lowe (2004). This enables to restrict the search space in image by extracting the most distinct features and by matching them. Next, area-based matching is applied to determine the match point position of GCPs.
Several types of ground entities have already been used as control information. Some examples can be found in Malmström (1986), Tommaselli and Tozzi (1996), Heipke (1997), Habib and Morgan (2003), Schenk (2004), Fraser et al. (2001), Marcato Junior and Tommaselli (2013), among others.
The technique presented in this paper uses vertical terrestrial images of GPCs to be matched with aerial images improving the integrated image orientation process. Using image matching techniques, image coordinates of GCPs are automatically determined with sub-pixel precision.
Experiments were performed with real data considering an indirect sensor orientation. The results obtained by bundle block adjustment were assessed on GCPs and check points. As a result, the discrepancies obtained using the proposed approach demonstrated smaller values in comparison with interactive measurements.
2. IMAGE MATCHING
SIFT is a technique for image processing originally conceived by Lowe (1999). It is widely employed to detect and to extract high distinctive features based on local gradients. Such features are fairly invariant to changes in illumination, image noise, rotation, scale, and 3D camera viewpoint. Descriptor vectors with 128 entries are used to store histograms of local image gradient orientations, and the matching is established by comparing...