Manual orientation of images using control poionts

[Igr]

Hi! Previously I had some experience with Photomodeler and started using Zephyr just recently, so some things still look quite sophisticated to me. The biggest issue I struggling with now is the following: some images from my set were nor oriented automatically and I have not succeeded with masking or increasing the parameters values. So I miss quite a big part of my model but all the not-oriented images have clearly visible points which in my opinion can be used as a control points. My idea was to subsequently add missing photos using control points, one by one. First control points are defined on automatically oriented photos, then the new ones are added as the new manually oriented photos are added.

1. So first of all I have not found in documentation detailed workflow that describes manual orientation the image using control points. I found just one video tutorial describing very simple case. As I understood this function can be called either from “Workflow\Add photo from Control Points…” or from a context menu in Cameras – Orient selected using “Control Points”. So is there any detailed documentation on this issue?
2. After performing manual image orientation using minimum five control points on some photos I receive some unpredictable results – the value of “Global Reprojection Error” becomes 400-500 pxs for some control points and some images just “fly” to absolutely wrong space. And at the same when I visually check position of control points on images - all looks good. So my question is – what are the tools\methods to track and define sources of these errors?
3. In Phtomodeler any change of position of any “control” point on image can be used for recalculation of the whole model. And definitely it influences the alignment of the photo the point was placed on. In Zephyr it looks like control points are used only to initially orient the photo. My question is – after photo was initially oriented, is it possible to improve its orientation, make it more precise (or fix errors) by adding additional control points or editing\removing bad ones?
4. After initial orientation of the image, control points are not placed on the oriented image - I have to go to the Edit for each control point and place it manually again. Combining with the lack of helper lines for potential position of points when doing initial orientation, it makes the whole procedure quite inconvenient in case you have to subsequently add several photos in a row. In Photomodeler photo can be basically oriented using only 3 points and then system gives you visual support when adding all other points. Here I have to keep all the required points in my memory for initial placement. So is it designed this way or I am doing something wrong?
5. If image is oriented manually – should it have all sells filled with data in the “Camera stats” sheet? I have not investigated it properly but some of the images have empty data for Long, Lat etc

Many thanks in advance for any comments on this issue!

[cam3d]

Hi @lgr here are my thoughts:

Firstly: We do not recommend trying to manually match images using control points unless it is your last option and you can't reshoot for better data.

- If some of your images were not oriented automatically (using General Default Settings) then the problem is probably in a lack of robust feature matches with other images in the set. Feel free to share the image set with us through support@3dflow.net (google drive or wetransfer works well for sending lots of images), and we can give advice on image acquisition for more successful processing.

- There's not a lot of documentation needed for the functionality of adding photos to the workspace, it's as simple as selecting the images which weren't successfully oriented, finding images within the set which overlap and have been successfully oriented, then running the camera orientation process on that localized subsection. It's not supposed to be a tool used frequently, it's very basic, and much better to avoid in the first instance, rather than have it a part of a standard workflow.

- The errors you are seeing are a due to the manual image orientation. Manual will almost always be noticeably off - Think about it in terms of 1000's of automatically detected features vs a handful of manually chosen features.. Best to average from the large reference data than to try and hit the target from a small pool of data.

- It is possible to improve poor camera orientation using control points and bundle adjustment, however again I must state that this is low on the priority list, better initial images are more important to focus on and will ultimately give the best results.

- I'm a bit confused by 4. & 5. - I feel like I'm missing some info here - Main point is, you should be trying to avoid manual image orientation. Can you please share a full dataset with me for feedback? Thank you!

[Igr]

Hi @lgr here are my thoughts:

@cam3d, many thanks for this response! Long period of manual postproduction is the reason why i want to find something different than Photomodeler. So that is good to hear eveything should go automatic here. ) But on the other hand i am out of ideas how to raise qualty of shooting in my case. So I would greatlly appreciate your comments on this issue. I will try to provide a data set, its about 3GB. I will also decribe my tasks and conditions.

[noramila]

In Zephyr manually orient images by adding non-oriented ones using clearly visible control points. Check for detailed workflow documentation, address high Global Reprojection errors by verifying control point placement, refine orientation by adding or editing control points even after initial placement, manually place control points on oriented images and check all data fields in camera stats are filled.