IMAGE-BASED TOPOGRAPHY RECONSTRUCTION BY MINIMIZING THE REPROJECTION-ERROR OVER A SURFACE

Abstract. Image based three dimensional modelling of small scale scenes sees an increased demand in recent years due to algorithmic advances, and growing awareness of users to the value of scene documentation in three dimensions. Nonetheless, most of the reported image-based reconstruction methods are either point driven, relying on matching large set of automatically sampled points (e.g., SIFT, SURF), or integral approaches. The latter are commonly based on generating an entire map of matching. For many such stereo-correspondence methods, however, images which are acquired from a narrow baseline and similar orientations are usually needed, in order to ensure satisfying results. We present in this paper a reconstruction approach, assuming both visibility and Lambertian reflectance, which is intended to be less affected by image difference. The proposed method performs a surface evolution which is implemented by an iterative scheme, applying a gradient flow which minimizes a reprojection-error cost function. The algorithm is applied on scenes of complex topography and varied terrain. Our results show applicability of the method, along with its satisfactory performance for some difficult imagery configurations.