Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Topics - patcarbon

Pages: [1] 2
General / V1.1 lens distortion equation
« on: June 02, 2016, 06:21:00 PM »

I'm having a debate with a reviewer on an upcoming Phantom 3 photogrammetry paper.  Could someone please clarify the exact form of the distortion correction used in the old V1.1?  If we assume no tangential effects and neglect K4, which of these 2 forms would be correct

(1) K1*r^2 + K2*r^4 + K3*r^6
(2) r*(K1*r^2 + K2*r^4 + K3*r^6)

Obviously the key diffference is that r factor at the start of (2) and whether or not Ks ultimately work in even or odd powers of r


Bug Reports / Tiled model preprocessing stalled
« on: March 13, 2016, 05:57:45 PM »
I recently upgraded to 1.23 and the tiled model isn't processing.  I'm only processing a small model with 126 cams and a dense cloud built on lowest.  But it stalls on the preprocessing has and is hardly using the CPU.  DEM export was easier in 1.0!

General / Optimisation from camera stations
« on: February 26, 2015, 07:34:44 PM »
Is there any point in optimising from camera stations alone?  From experimentations with UAV imagery I get the impression that yes but I'm not sure about the maths.  How can the X0Y0Z0 position of cameras model radial distorsion (and others) of a tie point?  What exactly does PS do when you hit the optimise button with only camera stations?


Python and Java API / Acessing calibration parameters
« on: June 13, 2014, 04:10:25 PM »
I have a basic .psz file and I'd like to create slightly modified versions of this project where only the camera calibration parameters change.

In this forum I've found leads on how to script and load calibration xml files.  But can a script directly access one of the calibration parameters (e.g. fx,fy,k1,k2, etc..)? 


P.S. I'm a python ultra-newbie

Feature Requests / Manual update: markers
« on: October 24, 2012, 11:47:31 AM »
The manual seems to have nothing on 'pinning' and 'unpinning' markers.  So the difference between green and blue markers is ambiguous.  Please update and see my post in the general section.  Anybody in the earth sciences or surveying who will use dGPS to control a set of photos needs to know this.


General / Green vs Blue markers and their impact on geocorrection
« on: October 24, 2012, 11:41:11 AM »
I've seen mention of blue markers in the forum, but what about green ones?  I know that you can pin a marker, then it goes green, you can unpin it, then it goes blue and you can remove it to go grey.  None of these things seem defined in the manual!

So what is the difference between a blue and green marker?  How, exaclty, are the different markers (grey, blue and green) used in the initial 7-parameter transform and then in the optimise stage.  what I mean here is will bleu and green markers be treated differently.  Are there weights associated?

I have 25 markers to place in 2000 images.  In some cases, the images are a bit blurry and I'd rather not use the marker.  Should I leave it grey?


General / Best way to optimise a multi-chunk model?
« on: October 22, 2012, 05:17:53 PM »

I've not got a 6 chunk model with a total of 1973 images.  I have about 400 low accuracy (10m) GPS camera positions (2 chunks) and about 20 high accuracy (5cm) ground control points. I've processed all the chunks (align: high, build geom: med) and I've aligned them all (point based, high).  Everything looks pretty good so far.

Now I want to optimise this model.  Since I only have 20 GCPs, there are only 2-3 per chunk.  Should I merge all the chunks before I optimise?   I have 16 gig RAM so according to the Photoscan wiki, I can re-build the geometry on medium just like I already did.


Bug Reports / Error reporting in 'Align chunks' batch job
« on: October 22, 2012, 11:52:29 AM »

Very minor bug:

I tried to align chunks as a batch job over the weekend.  When I arrived this morning, the batch display said that it had done the job with no failures.  However, nothing had happened.  There was a bug because I had changed a folder name by mistake after a chunk had been processes.  So photoscan could not find the image in the designated path.   So somehow, the fact that it did not find the images and aborted the alignment was not recorded as a failed alignment.  Nothing major but needs to be fixed.


General / Aligning chunks in steps?
« on: October 16, 2012, 04:22:08 PM »
I'm processing a 2000 image project.  The surface is quite self-similar and aligning the entire project in 1 chunk was a disaster (see below). So I'm having to use several chunks.  I'm now on 7 chunks (which correspond to 7 UAV flights) and I'd like to end up with 1-2 chunks before I optimise.

Is it possible to use markers to performa an initial chunk alignment and then have a point-cloud match to reffine this alignement?  Does the 'points' align option just delete any previous alignment?


General / Build Geometry subsampling
« on: October 16, 2012, 04:14:17 PM »

I can't find this in the wiki or older posts.  What level of subsampling occurs at Ultra high, high medium and low geometry builds?

I think Ultra uses the raw image resolution and high uses a 2X2 pixel average.  but what about medium and low?  These things should go in the wiki.  It's quite usefull to know if you have an idea of the raw image resolution and you have an idea of the final average point cloud/DEM resolution you need.

General / The mysterious 'bowl effect'
« on: July 27, 2012, 11:57:29 AM »
I've seen a few references to this bowl effect on pages here.  i've also seen what looks like the bowl effect in data.  In fact, this was in images taken with the UAV produced by the Swedish re-seller of Photoscan, SmartPlanes.

What causes this?  Since the effect coveres a whole scene, in my case with over 200 images, I have always assumed that it can't be lens distortion since the bowl fits a 2nd order surface.  When is this distorsion introduced?  Alignment? Build Geometry? Georeferencing?  Surely it's in one of the first 2 since, if I understand right, georeferencing just uses a rigid 7-parameter transform so that means the 'bowl has to be generated earlier'.  Correct?

And what's the maths behind the optimisation?  It it relying on a photogrammetry-style epipolar constraints approach?


General / Number of images vs final number of vertices?
« on: July 25, 2012, 09:12:47 PM »

What is known about the relationship between the number of photos going into Pscan (or any other SfM routine) and the final number of vertices in the raw point cloud?  Can anybody point my to some literature in computer vision/SfM research?


Feature Requests / Autosave after aligning chunks
« on: July 06, 2012, 11:06:22 AM »
A very little thing:  the align chunks function, which can take a while for big multi-chunk projects, should at least have a checkbox option to save after completion.  Or alternatively, add this function to the batch process dialog where the option to save does exist.


General / Exporting vertices
« on: July 04, 2012, 04:26:05 PM »

Is there an easy way to export the vertices of a calculated model?  The way I understand the exports, if you 'export model' you get the mesh, not the point cloud (or at least the point cloud comparison software I'm using only sees the mesh).  But then if you export points and re-select the desired quality, it seems to reprocess the model.

So what I want is to export a dense point cloud for data which has already been aligned and built.

What's the quickest way?


General / Improving chunk alignment
« on: November 17, 2011, 03:53:56 PM »
For my type of Images, I've found that the automated chunk alignment can give bad results.  There are many similar features, and the system I'm looking at even has some fractal characteristics so cross-scale similarities exist.

So I was wondering if the following is possible:
1-  Manually align the chunks with 3 markers.
2- Optimise this alignment, hopefully to sub-pixel accuracy, with an automated match run.

So I'm hoping that with 2 approximately aligned chunks, the pattern match algorithm will limit it's search area via some sort of epipolar constraints and thus reduce false positive matches.  Can this be done in 8.4?


Pages: [1] 2