Image-based modelling for object reconstruction

Abstract

Although automated object surface reconstruction via feature-based matching is commonly employed in topographic and stereo close-range photogrammetry, it has rarely been employed in conjunction with multi-image, convergent photogrammetric networks. The aim of the research undertaken for this PhD has been to develop mathematical models, algorithms, computational procedures and operational methodologies necessary to facilitate automated 3D object reconstruction in multi-image convergent photogrammetric networks.

The development of an automated 3D surface measurement and visualization strategy, suited to multi-image, convergent photogrammetric networks that can be incorporated with robust automatic network orientation within a single data processing system, is presented. This development centres upon a feature-based matching approach that incorporates image pre-processing via the Wallis Filter, application of the FAST interest operator, and utilization of the Poisson Surface Reconstruction algorithm.

The creation of 3D models using photogrammetric techniques involves the following phases: image pre-processing, camera calibration and network orientation, image scanning for point detection, surface measurement and triangulation, blunder detection and statistical filtering, mesh generation and texturing, and visualization and analysis. At present, there are no commercially available systems that allow all of these steps to be executed automatically within the same data processing environment. Instead, the phases are typically completed using multiple software systems that require the 3D data to be translated and interchanged between each of the various measurement, modelling and visualization packages. The culmination of the research project lies in the development of a fully automatic photogrammetric procedure in which all phases of the object reconstruction process can be implemented within a single software system. Notable achievements within this research include automatic high-accuracy feature extraction and subsequent point cloud generation in multi-image networks, robust network orientation, 3D object point determination and fully automatic wireframe generation and texturing, thus enabling automated surface measurement, modelling and visualization.