Messages

1

Face and Body Scanning / Two-fold alignment in point cloud

« on: July 16, 2018, 05:49:10 PM »

Hi all.

First of all thank you in advance for your reading.
I'm in trouble trying to making 3D model of a set of rare abyssal fishes for the Spanish Institute of Oceanography. I made 151 photographies of the subject very accurately, using tripod and a f-value of f/18 on a 100mm macro lens, rotating around the subject with very constant lighting. When depth field was not enough and not all the subject was on focus I pointed to a certain area of the fish, shot it, moved slightly, pointed to the area that was out of focus before and shot again.
I successfully employed this technique with other models and everything went well.
What I'm obtaining now is a strange two-fold points cloud, with most of point correctly aligned and a large part of them misaligned over a symmetric and specular fold of the original cloud.
Please see the attached figure to understand what I mean.
Photo alignment with High quality, generic pre-selection and adaptive camera model fitting enabled.
Any idea of what is occurring?
Any idea of how to detect the images responsible for the misaligned part of the cloud?
Any idea of how to force photo alignment to be coherent with the other part of the cloud?

Thank you in advance.
Best regards
Simone

2

General / Re: Poor mesh with good dense cloud

« on: January 07, 2018, 11:36:38 PM »

Adding to what James points out.

It appears that the matching was based on points detected on the shark alone which means that the camera position and orientation may not be precise which in turn may cause problems with  the dense matching.
I would suggest retaking the images with a greater depth of field so that the floor is in focus as well.

You're right...
Unfortunately I will have no chance to make further photos but I will try to improve the usage of the ones I have.
Thank you very much for your kind support.

3

General / Re: Poor mesh with good dense cloud

« on: January 07, 2018, 11:35:11 PM »

In your attachments 'DenseCloud_Classes.JPG' and 'DenseCloud_RGB.JPG' the light colour points are on the top surface of the tail, and the dark points are on the underside, facing the opposite direction - away from you.

If you looked at it from below you would then see that the dark points become lighter, and the light points become dark as you view them from the other side. The dark colour indicates the orientation of the surface that the point corresponds to.

Your attachment illustrates that you do not have many points on the upper surface of the tail, and photoscan can not create a good watertight mesh of an object with only one side available in the dense cloud so it only meshes the parts with data from both sides, as shown in your attachment 'Mesh.JPG'.

So you need to check your images of the upper side of the tail to check overlap with each other and adjacent parts of the object to see why this part is not being reconstructed in the dense cloud. Likely causes are that it is out of focus in some/all images, insufficient images, or it moved during the scan.

You're right!!!
That is why the different colors!!!
Thank you!!!
I have to try to obtain more data on the opposite side (upperside) of the tail in order to obtain a "closed" mesh.
Thank you very much for your great support!

4

General / Re: Poor mesh with good dense cloud

« on: December 28, 2017, 11:51:03 PM »

One of the  images you used would help so that we could tell what this is supposed to look like but the dense cloud doesn't really look that good.
It seems to be quite noisy.

Thank you n@sk! and sorry for my late!
Find attached one of the images of the tail...
Thank you in advance

5

General / Poor mesh with good dense cloud

« on: December 19, 2017, 06:29:11 PM »

Dear all.
I'm making a 3D model of a stuffed shark and I'm struggling to obtain a complete and correct mesh. The dense cloud is generally very good and reproduces very well the animal, but in some areas the derived mesh is holed. In particular the tail is great in the dense cloud but it is later not correctly reconstructed in the mesh. What I noticed is that the dense cloud class shows pixels with not exactly the same colors (they are mainly gray and some of them are black) and I noticed that the black ones are not used for the mesh and let the holes. I did not use dense cloud classification.
What are those black pixels in the dense cloud?
Why the mesh does not use them?
Thank you in advance.
Best regards
Simone

6

General / Re: Zenital plus oblique photos for real estate photogrammetry

« on: July 15, 2017, 11:27:48 AM »

As you can see from the previous image, no points matching in more than one image is found in the case of the oblique photos (white circles).
And in the case of the few oblique images in which some points matching has been achieved, the result is somewhat shifted with respect to the part of the mesh obtained by the zenital images.



The slightly coarser (because obtained by the zenital images) wall closer to the observer is the correct one, coherent with the roof position although obviously more fragmented and less extended because of the limitation of the zenital point of view. The secondary wall, behind the first one, shows more resolution and is more complete but is somewhat shifted inside the house and does not match with the rest of the structure.

So, resuming....

Two kind of problems here:

- The lack of points matching among (most of) oblique images. ¿Why? The images are taken manually attempting to provide the correct overlapping (and indeed a great portion of each image is clearly visible in the next one), but the software seems to not be able to match them. In some case the images are shot not in order...I mean I shot a photo and then I came back to take another part of the wall, ad then I came back forward to shoot the next part of it. Is it of concern?
Also some of the images are taken horizontally (zero elevation). Could it affect the result? I don't think so because unmatched points are found either among horizontal and oblique images. Any idea?

- The strange shift of the point cloud not matching the structure obtained by the zenital images. I'm really not able to guess the origin of this issue.

By the way the flight has been made with a DJI Phantom 4 drone.

I hope to have been clear enough.
Thank you in advance for your help.

Simone

7

General / Zenital plus oblique photos for real estate photogrammetry

« on: July 15, 2017, 10:44:43 AM »

Dear all

First of all, a big hello to everybody and thank you for having created such a cool community of such a powerful software.
I'm really happy with Photoscan, using it as the main tool for photogrammetry of real estate.

Unfortunately there is a magnific villa here in Costa del Sol, in Andalusia, which is causing me several headhaces.
It is a detached villa with a quite large and very articulated roof and a wide garden with a pool.
All concerning the zenital flight worked like a charm.
A lot of details and perfect geometric coherence in the mesh. No problem at all.

One "funny" feature of that house is that the roof oversteps the perimeter of the internal house and create a sort of canopy all around the structure. While it is really confortable to take a cup of tea shielded from Spanish hot sun, it is a nightmare for a correct 3D reconstruction of the whole house and its vertical walls. These walls are indeed "hidden" by the canopy and are only partially seen by the drone flying over the house.

In order to cope with this limitation I included in the flight a new set of oblique photos all around the house, taken manually while I was travelling along the perimeter of the villa. In the following images you can see the zenital set of images and the belt of oblique perimetral ones.





You can also see how only a portion of those images are displayed as blue squares, while few of them, especially the ones on the opposite side of the house, are depicted as circles. As far as I understand it means that these photos have not been correctly georeferenced/matched/used to extract point clouds.

Since the oblique images include a piece of sky I masked them to take into account only the portion from the border of the roof downwards.



More in the next message