How to loosen tolerances in alignment to deal with minimal or poor image overlap

[sambagis]

Hi all,
I have a data set that has relatively reasonable overlap but in areas of dense forest the overlap seems to be just poor enough that I am unable to get alignment.  I have gaps in the data and have tried all the different accuracy options during alignment (highest to lowest) to get the images to get some sort of alignment but no luck.  When I view matches with those images that dont align I can see that there are many "invalid points" but they all look like good / valid points between images.  I have tried also increasing and decreasing the horizontal accuracy of the initial position.  Still no change.  I am working with imagery from a PhaseONE IXU1000 sensor that has really good camera geometry but at very high resolution where the trees have significant lean between images.  I think this is whats creating the lack of alignment.  We are unable to recollect data in this area so I am hoping someone has any ideas on how to reduce the tolerances in the initial alignment phase that will force agisoft to give me even a poor result which would be better than the gaps I have now. 

I have attached a screen grab.

Thanks all.   

[sambagis]

quick update.  A friend suggested adding manual points until Agisoft decides to align the problem images.  Im happy to work on that but I have no idea what the best workflow is for that process.  Does anyone have any insight here?

[Alexey Pasumansky]

Hello sambagis,

If you are not able to align all the cameras using only Reference pre-selection neither using High, nor Medium/Low accuracy settings, please share with us the project in PSZ format containing the matching points, providing that there are at least invalid matches between aligned and not aligned cameras.

[sambagis]

Hey Alexey.  Thanks for getting back to me.  I have shared the project in PSZ format with support via email.  I will report to the community here if we get any sort of useful resolution.  Thanks for all of your help and great support.

[Matt]

Use Photoscan Version 1.4.3 and see how you go.  Otherwise manual tie points with no coordinates can work well.

[Alexey Pasumansky]

Hello sambagis,

I have checked the project that you have provided and it seems that in the most cases the overlap is insufficient of the areas covered by the dense forest.

For stable alignment you need to have at least triple overlap in both directions (i.e. more than 50%), but the Not Aligned images in the project mostly have a few matching points detected with the neighboring cameras only.