A Feature-Based Matching Approach to Automated Object Reconstruction in Multi-Image Close-Range Photogrammetry

SUMMARY Automated object surface reconstruction via feature-based matching is commonly employed in topographic and stereo close-range photogrammetry. However, it has rarely been utilised in conjunction with the convergent photogrammetric network geometry that is typically adopted for engineering surveys. This paper describes a new approach to automated 3D object reconstruction from highly convergent, multi-image networks in which the very geometric diversity and redundancy that presents difficulties in area-based matching is a distinct advantage in the feature-based matching approach. Following an initial, automatic exterior orientation of the network, the FAST interest operator is applied to Wallis-filtered images to extract interest points, and a matching of interest points is carried out to yield a dense and accurate 3D point cloud. This is subsequently converted to a triangulated mesh via a Poisson Surface Reconstruction technique, and textured. The paper commences with a brief overview of three important components of the proposed object reconstruction approach: the FAST interest operator, the Wallis filter and the Poisson Surface Reconstruction. Experimental results are then discussed, via an engineering measurement application in which surface modelling is performed to support deformation analysis. The results highlight the practicability, robustness and accuracy of the proposed automated multi-image measurement approach incorporating feature-based matching.