UNCALIBRATED CAMERA IMAGES IN DELINEATING PARCEL BOUNDARIES

Abstract

The reconstruction of 3D points from two view camera images has found several applications in various fields of engineering, science and architecture today. Many of these fields are involved in the generation of 3D models using diverse technology in 3D reconstruction. In geomatics, estimating 2D or 3D locations can be made through direct or indirect measurement. Direct measurements involve laser technology (e.g. total stations), satellite positioning and terrestrial laser scanning. Indirect methods, on the other hand, involve remote sensing, aerial and terrestrial photogrammetry. In delineating parcel boundaries, direct measurements with total stations and global navigation satellite systems (GNSS) are used in field operations. Although these have been quite stable with satisfactory results, the cost of operations is still relatively higher compared to indirect measurements, especially in larger areas of coverage. This study investigated the feasibility of using close-range photogrammetry thru uncalibrated image pairs for fast acquisition of target locations. The general algorithm will derive the fundamental matrix, generate the camera matrices, triangulate the projective 3D positions and finally transform these points to their metric ground 3D positions using Ground Control Points (GCP). The extracted northing (y-coordinates) and easting (x-coordinates) of the parcel’s corners were compared to coordinates derived from total stations. These showed an error vector of up to 10 cm. Results also revealed that the required accuracy for tertiary survey can be achieved around 30 percent distance from the image center. Beyond this, the positional error increases, making this method unsuitable for boundary delineation. The RMSE of this set of points was 0.0758 m or 7.58 cm.

Keywords: Close Range Photogrammetry, Computer Vision, 3D Reconstruction, Direct Linear Transformation