TOWARDS A COMMON TERMINOLOGY

3. TOWARDS A COMMON TERMINOLOGY

The ‘type’ of photogrammetry discussed in this paper has been referred to under various names, including Multi-Image Photogrammetry, Close-Range Photogrammetry or Structure- from-Motion Photogrammetry. A discussion between the author, Kotaro Yamafune, Massimiliano Ditta, Massimiliano Secci, Bruno Parés, Kevin Edwards and Rodrigo De Oliviera Torres – all of whom work in this specific domain – has highlighted the importance of using a common designation. Names such as Close-Range or Multi-Image Photogrammetry are ambiguous, simply because earlier photogrammetry approaches were also capable of processing multiple images, or images taken at close range. Structure-from-Motion Photogrammetry makes reference to a specific method commonly used to automatically generate a 3D point cloud from images, but the name leaves no room for other, often complementary approaches. As such, for the reasons discussed below, we believe the term ‘Computer Vision Photogrammetry’ best describes the type of photogrammetry referred to in this article. Traditionally, the history of photogrammetry can be divided into several phases, based on the prevalent processing procedures used at different points in time. A commonly used classification is the one proposed by Gottfried Konecny, who differentiates between Plane Table Photogrammetry theodolite survey aided by pictures, Analogue Stereo Photogrammetry stereo pairs plotted using an analogue plotter, Analytical Photogrammetry analogue pictures plotted using an analytical plotter, i.e. based on mathematics rather than on mechanics and Digital Photogrammetry digital rather than analogue pictures, plotted on a computer rather than on a plotter, using the same principles as Analytical Photogrammetry Konecny, 2010. Each of these subsequent approaches to photogrammetry developed within the traditional field of photogrammetry itself – a field which was devoted almost exclusively to aerial mapping. In the 1960s a much younger, completely different field of study emerged, namely computer vision. At the time, researchers were optimistic about the prospect of creating intelligent machines; robots that could mimic human behaviour, that were capable of understanding and interacting with the world around them. Within this broader agenda of creating artificial intelligence, computer vision emerged as the field concerned with solving the ‘visual input problem’: developing methods which would allow machines to see and eventually perceive their surroundings. Researchers reasoned that if machines were going to understand their surroundings, they would first need to have an understanding of the 3D structure of those surroundings. Consequently, for several decades now the main objective of computer vision research has been ‘to develop mathematical techniques for recovering the three-dimensional shape and appearance of objects in imagery’ Szeliski, 2010. As such, all the techniques used in modern photogrammetry software – including automated feature detection, camera auto- calibration, feature-based alignment, Structure-from-Motion, etc. – were developed not in the traditional field of photogrammetry, but in the field of computer vision. It is thanks to computer vision algorithms that applications such as PhotoScan can automatically detect points in overlapping pictures, in order to automatically calibrate our cameras, in order to automatically align images, in order to finally automatically generate a detailed 3D model. This unprecedented degree of automation provided by computer vision techniques is what truly sets the current generation of photogrammetry software apart from past approaches, and that is why we believe ‘Computer Vision Photogrammetry’ is the name which best describes this specific type of photogrammetry.

4. CASE STUDY