Combined use of photogrammetric and computer vision techniques for fully automated and accurate 3D modeling of terrestrial objects

Nowadays commercial software able to automatically create an accurate 3D model from any sequence of terrestrial images is not available. This paper presents a methodology which is capable of processing markerless block of terrestrial digital images to perform a twofold task: (i) determine the exterior orientation parameters by using a progressive robust feature-based matching followed by a Least Squares Matching refining and a bundle adjustment; (ii) extract a dense point-clouds by using a multi-image matching based on diverse image primitives. The final result is an accurate surface model with characteristics similar to those achievable with range-based sensors. In the whole processing workflow the natural texture of the object is used, thus images and calibration parameters are the only inputs. The method exploits Computer Vision and Photogrammetric techniques and combines their advantages in order to automate the process. At the same time it ensures a precise and reliable solution. To verify the accuracy of the developed methodology, several comparisons with manual measurements, total station data and 3D laser scanner were also carried out.