What causes "Can't build dense cloud"?

[photohokie]

I recently started exploring the trial version of Photoscan Pro and am really liking it so far, but I'm having an issue I can't get to the bottom of.

Some of my projects won't let me build a dense point cloud.  I looked through the manual, help file, and internet and could be missing it, but I don't see anywhere that addresses what causes this error.  My current project has 91 photos and around 30,000 points in the sparse cloud, but I've experienced it with other photo sets too.  Does anybody know how I can resolve the error and create a dense cloud?

Thanks!

[aggieair]

Have you tried zooming way out to see if there are any random points floating in space?  Highlight them with the square, circle or draw tool, and delete them with the delete key.  That was one known reason for not building the point cloud.

[aggieair]

I've also had this error too even after updating the projection system, re-optimizing (like when I add control points), save and then it doesn't build then either.  Sometimes if I close out and then open the project back up it will build suddenly (not sure why?).  Other times the computer will crash on me!

[Marcel]

30000 points in the Sparse Cloud seems really low for a project with 90 photos. (I can't check now,  but I think our projects with this number if photos have at least 500k points in the Sparse Cloud).

What kind of camera are you using? Could you show us a source photo at full resolution?  Also, what settings did you use for the Alignment step?

[James]

I think another reason this can happen is if in each 'pair' of images one of the cameras is in the field of view of the other, as pairs like this are filtered out before building dense cloud. Probably because it does something odd when trying to rectify an image where the epipolar points are within image bounds. In other words you moved the camera forwards and backwards more than left and right. But that's pretty much guesswork in all respects!

If it is that, then the good news is i think they are working on a way around it for the next update.

[Marcel]

... Probably because it does something odd when trying to rectify an image where the epipolar points are within image bounds. In other words you moved the camera forwards and backwards more than left and right. But that's pretty much guesswork in all respects!

If it is that, then the good news is i think they are working on a way around it for the next update.

That sounds interesting. Does that mean it no longer crashes, or will the results also be better?

It would be interesting for me because often my tripod is not big enough to get a high viewpoint. I tried positioning the camera further away (to give it at least some positional change) and angling it upward to get the high parts, but this only leads to noise in the mesh at these parts.

I got a huge tripod now, but it would still be nice if I could be a bit more flexible in camera positioning.

[James]

I'm not sure exactly what it will mean. I guess it just means it will be able to use these image pairs to extract information where currently it is unable.

I found the post by Alexey that I was thinking of when I posted all that here:

Добрый день,

Не могли бы Вы прислать проект, желательно с исходными кадрами?

Что касается невозможности построения плотного облака, то это может быть связано не только с проблемами калибровки (значения Fx, Fy, Cx, Cy выглядят явно странными), но и с тем, что съёмка велась таким образом, что камеры "видят" положения практически всех остальных камер. Такие пары отфильтровываются на этапе построения карт глубины. Обычно это характерно для случаев, когда камера двигалась в  направлении своей зоны видимости.

Пожалуйста, сообщите ещё версию ОС.

and here

Спасибо за предоставленные данные.

Воспроизвести проблему с неправильным отображением значений с плавающей точкой пока не удаётся - в обоих проектах значения адекватные, так что пока ещё неясно, почему в Вашем случае отображается только первая ненулевая значащая цифра.

Что касается самих фотографий, то можно было бы направить камеру чуть вбок, чтобы не захватывать положение съёмки следующего кадра.
Однако, в следующей версии мы планируем улучшить обработку кадров, сделанных подобным образом. Ниже результат обработки Вашего проекта с использованием улучшений из следующей версии:

I don't speak Russian but Google translate seemed to do a decent job of it: http://goo.gl/P0Nomk

[Alexey Pasumansky]

Hello photohokie,

The most likely reasons of "Can't build dense cloud" problem are the following:

- insuffient number of matching points inside the bounding box between every pair of images,
- presence of camera shooting position in the field of view of other cameras,
- camera calibration issues.

Maybe you can provide a screenshot of the sparse cloud and camera positions?

[photohokie]

Thank you all for your replies!  I tried the first suggestion, and sure enough, there were extraneous points a good bit away from the subject I was trying to model.  I thought by resizing the region, I was excluding everything further away in the model, but perhaps not! 

After deleting all the extraneous points, it works like a charm.  Thank you!