[Contextcapture] Curved model accuracy problem from nadir

Thanks. Please add this description for this and other types to documentation as I had different impression what it actually does that it will somehow fix the upside down flip of block or tilted block.

Oto

To answer your question "What does block type->Vertical views only do?" this tells the software the imagery is vertical view (Nadir) by placing a min max distance combined with the focal length, sensor, and now distance (min / max) the software now looks for pairs with only othe images that will intersect.. thus you can now choose exhaustive for pair selection and use it even with tens of thousand of images and it will finish in a shorter period of time... without this it may never finish AT when pairs "exhaustive" are selected and VERY large datasets are used.

I saw that they cheated by importing positions for UAS Master from Agisoft results and I'm aware that Pix4d also have pre-calibration data which helps in this case. For MicMac also they used radial correction calibration model F15P7 to correct "bowl effect" and very high density tie-point extraction with full  so that explains better results as much more iterations were done in MicMac than in other software. If 3D calibration were re-used in nadir project I think same accuracy could be achieved with CC also.

At least next paper will use oblique photos so they will be puzzled with much better CC performance - "Further activities will follow different directions. On one hand, the other flights described in Table 1 will be processed with attention to oblique blocks in order to investigate their influence on final accuracy"

With 3D models and DTM production there is no such issue and reviews are very positive:

Digital Photogrammetry Software Comparison for Rock Mass Characterization

Three-dimensional reconstruction of plants and a single building from UAV images

Yes I am interested in better calibration of camera but with approach that camera parameters are also optimized from GCPs because this is another mistake users makes that they export optical parameters after adjusting with GCPs and expect that it has improved camera calibration.

What does block type->Vertical views only do? Honestly hove not noticed that any of these aerotriangulation settings improve anything, only "high density" or "radial distortion" seem to work but all others are so specific and of no real use.

Hi @Oto

I briefly looked at the paper, and saw several flaws in it including as you pointed out not using most recent software versions.

However concerning "radial distortion" sometimes seen in Nadir photo sets. The primary cause of this is that Nadir only photo sets lack sufficient convergence between photos, forcing numerous parameters to be estimated, resulting in a much harder mathematical problem with the self calibration portion of the AT.

Many of the programs tested in the paper provide initial optical parameters (calibration) for cameras (IO or interior orientation). IMHO a fair comparison would have used the same optical properties (IO) on all the software. The majority of camera models in the ContextCapture camera database only have the sensor size, image size and principal point (unrefined). It is strongly recommended that users complete camera calibrations on all camera / lens configurations they are using when capturing Nadir only photos. A good calibration is also useful when working with obliques and the highest degree of accuracy is needed.

I will be posting a new document within the next week on how to calibrate a camera  > add the calibration to the library > and use it in your ContextCapture project

In the short term contact me via email and I can make the draft available to you.

Thanks

Mike Barkasi