ASM
The End of Photogrammetry
Photogrammetry may well be nearing the end of the road as a profitable field of academic study. At least, that was one possibility to emerge from a presentation to ISPRS last week.
Wolfgang Forner, from Universitat Bonn in Germany, said that most photogrammetrists face extreme challenges from computer and machine vision specialists. Many such people have discovered geography as a profitable field for their applications, he added.
This is not so much a take over as a contest over new territory. Forner says the basic problems of photogrammetry have been solved, or will be within the next few years. That is, we can now take a string of images and orient, rectify and derive a full 3D model from them. All the complexities of geometric and instrument correction are well within our grasp.
The key issue, at least for those academics looking for the step after next, is semantic. That is, it's not whether we can produce a 3D model, but can we make a machine interpret the image? Is this blob a house, an office block or a rock in a field? If it is a house, how big is it? Where is the entrance? How many storeys does it have?
These are important and valuable questions, and photogrammetrists have not even begun to answer them.
Up till now, such questions have been the exclusive domain of computer vision people. They take an essentially flat image and try to derive 3D information based on an understanding of the shapes in the image.
So, for instance, the state of the art in computer vision might be the ability to recognise windows in a wall, or find specific architectural features in an image, and thereby infer a building. This is done with the aim of, for instance, finding similar images in many different images.
The addition of photogrammetry 3D information will make this process more robust. But the practical question is whether photogrammetrists will be in any position to do so.
Forner says photogrammetrists typically work in groups that are too small and isolated from one another. It is difficult for them to compete with computer vision or machine learning specialists, who are traditionally cashed up and work in much larger groups.
So the way forward for the profession might be to forge alliances with computer vision specialists, or maybe to try to break down the barriers that exist between photogrammetrists today.
The bottom line is that photogrammetric questions have never been as important as they are today, or worth as much money. That might be food for thought if Bill Gates comes calling.
