Lens Distortion Correction
Read Kineticorp‘s SAE research papers about lens distortion:
(Toby Terpstra) Whenever we’re talking about photogrammetry, accuracy is always a consideration for us, regardless of the project. We want to know that the solutions—the photogrammetric solutions are accurate such that we can take measurements out of those solutions and know that those measurements are accurate. With lens distortion being present in a photograph, the camera matching photogrammetry or other kinds of photogrammetry can often be difficult or really impossible to get good solutions out of. So, it’s important to always remove those distortions whenever possible.
At Kineticorp, our first step is to look at the EXIF data on the photograph. Now, the EXIF data is the meta-data contained within the image file itself. So, within that image there’s header information or meta-data where you can see what camera was used, what the settings were within that camera, the field of view, that kind of thing.
(Sean Miller) Most of the lens distortion removal that we do is using software that already has the profiles for lens distortion that’s caused by the curvature of the lens. And there’s a big, long equation that you can calculate out the actual distortion and use that to remove it and there’s programs that can do that automatically.
(Toby Terpstra) If you’re stuck with an image that doesn’t have EXIF data, which happens to us all the time, we get images that are provided that have been saved out as a PDF, which strips out typically all of the EXIF data and then you don’t have the information about the camera. When you’re stuck with those kinds of images and you’re trying to perform photogrammetry on those, it can be really problematic if not possible.
We get photographs that we knew had lens distortion in them, visible lens distortion, but we had no EXIF data or no great methodology for being able to remove that distortion. So, out of the results of that frustration we began studying ways to remove that distortion without the EXIF data. The first one is what we refer to as the straight-line method, where we look at the photograph and determine, “are there any lines within this photograph that should be straight.” Edges of buildings, telephone poles, streetlights.
The second method is the point cloud method. We use a scanner–a 3D scanner like a FARO scanner or a Leica scanner, anything would work, to document the scene and really encompass it with a point cloud of data. That data is accurate, when thinking of a FARO scanner right now, plus-or-minus two millimeters. So, you have an accurate representation of that 3D environment that you can then take and compare specific points in the photograph to or against and know that, “look, here’s the true world measurements, here’s the measurements in the photograph, which is not accurate,” and then in comparing those you’re able to remove the distortion based on that. You should consider lens distortion. It’s not that it’s going to effect every camera match or every photogrammetry solution, but if you’re not considering it, you can potentially achieve that much error in your solutions or have that much error in your solutions and that can be problematic.