Forum

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 - sarko

Pages: [1]
1
General / Height Field vs. Arbitrary
« on: May 21, 2015, 12:21:49 AM »
Hello,

I have posted on this topic before, but am hoping someone perhaps has figured it out.  I use a lot of historic imagery and occasionally have problems generating height fields.   I have attached three images showing a dense point cloud, arbitrary mesh, and height field mesh.   As you can see, the dense cloud looks good, as does the arbitrary mesh.  The height field mesh looks horrible, however.  This may not be the best example since it is in a highly glaciated region and many areas are void due to saturation in the input imagery, however, I have seen this issue in other non-glaciated regions as well.  A couple things I have tried:

1.  Used gradual selection to try to eliminate lower quality points in sparse cloud -- no impact
2.  Cropped the edges of the dense cloud thinking that this was an edge effect.  I even cropped to only an interior region as far away from edges as I could. -- no impact
3.  Tried various options for the number of points calculated and maximum tie points retained -- no impact

It seems as though this is happening more often, but it may just be the data I'm using of late.

I have posted a tar ball of the project file and raw data scans here:

https://akarise.asf.alaska.edu/photoscan/BlackRapids.tar.gz

The screen caps were from my MacBook Pro, but I see the same problem on my primary processing system, which is Windows.  I am using the latest stable release of Photoscan Professional. 

Any suggestions would be greatly appreciated. 

2
General / RAM Locking during Export Orthophoto
« on: April 23, 2015, 07:21:59 PM »
Hello,


I have used Photoscan for quite a while now, but recently have run into an issue that seems new.  I work primarily with historic aerial photography where individual images have dimensions of around 10000x10000 pixels.  In a typical project I will use anywhere from 25 to 200 images of this size. 

Lately, when I export my orthomosaic, my machine RAM locks and it takes forever to export.  I export in tiles that are generally either 4096x4096 or 8192x8192.  Either way, they are smaller than a single image that I'm importing into the project.  The strange thing is that when I export I will see that Photoscan is blending anywhere from 10 to 20 images per export tile.  Given the overlap that I have in my imagery, realistically I can't see why this number would ever exceed 9.  I have about 60% overlap and about 10% sidelap.  Most points only exist in 2 input images.  I am not using the color correction option. 

My machine only has 16GB of RAM, so I know most people will just say to upgrade my machine.  However, I have used this machine with Photoscan for over 2 years now and this issue has only popped up in the past couple months.  I'm getting a new machine tomorrow, but I still think this is an issue that may impact many people. 

Thanks,
Scott

3
General / Valid invalid matches in historic photography
« on: April 03, 2015, 02:38:51 AM »
Hello,


I have been a Photoscan user for a number of years now and have started seeing something new in some of my projects.  I work primarily with historic aerial photography anywhere from a few years to over 70 years old.  Recently I have been working with some historic B&W imagery from the 1950s.  I have about 500 scenes scanned by the USGS (approximately 10,000x10,000 pixels).  I masked the images and aligned the photographs.  I used 40,000 points and set the tie point limit to 0 to get as many matches as possible. 

I have a number of images that created a good number of matches that were incorrectly marked invalid.  In this case, they were not included in the model.  In an example image I have over 10,000 match points generated between its adjacent images and they are all marked invalid when it is very clear from the "View Matches" window that they are indeed valid. 

I can't get a screen cap right now, but will try to do so later.  Just thought I would see if others are having this problem as well. 

Thanks,
Scott

4
General / Distorted Geometry Result
« on: January 08, 2013, 11:35:18 PM »
Hello,


I have run into a strange problem with Photoscan recently.  I generally use a lot of historic data and seldom have any issues.  However, recently I've started to see this issue more often.   These attached images show the point cloud and the distorted geometry that resulted.  In general, the point cloud looks very good, but when I calculate the geometry, I get distortions along the upper and lower edges that ruin the accuracy of the model. 

I am performing the alignment with 80,000 points on high.  The geometry was computed using height field on low/medium with 1000000 faces.  The point selection and matching seems to look fine and the point cloud generally looks very good.  The issue occurs with both PS 0.8.5 and 0.9.0. 

Would appreciate hearing if anyone has seen something like this before. Will upload the second image in next post. 

5
Python and Java API / Increasing number of point
« on: September 06, 2012, 02:50:19 AM »
So, I've had a fun day of learning about the PhotoScan API.  I have a couple questions that I'm hoping someone can answer. 

1.  Is there any planned date for release of the ability to alignPhotos and filter by mask?  That is a pretty important feature and a time saver. 

2.  When I run my python script from from the Tools->Run Script menu it does not seem to be using the preferences that PhotoScan has set.  In my case, I would like to detect more than 40000 points per image, but that is the default and all I can get.  My preferences are set to 150000, but they seem to be ignored.  Is there a way to correct this? 

I guess that is about it for now.  I like the API for now and thank you for providing it. 



6
Python and Java API / Question on setting user calibration
« on: September 06, 2012, 12:37:16 AM »
I am taking my first stab at using the python API in Photoscan.   I have been able to create a document, load photos into a chunk and align/create geometry.  However, I want to update the camera calibration information and have been unable to get it to work.  Relevant snippet of code is below:


Code: [Select]
for i in range(0,numFiles):
    p = PhotoScan.Photo()
    testFile = data[i].rstrip()
    p.open(testFile)
    cenX = p.width/2
    cenY = p.height/2
    p.user_calib = ps.Calibration(cx=cenX,cy=cenY,fx=12180.0,fy=12180.0)
    newChunk.photos.add(p)
   

I'm trying to use p.user_calib to set the calibration, but it sets all values to zero when I do what's above. Any thoughts???


Thank you. 

7
General / What Vertical Accuracy Should I get?
« on: May 26, 2012, 12:08:26 AM »
Hello,

I have been a Photoscan user for some time, but typically use older, course resolution imagery.  I have recently received a set of about 300 high-resolution images collected from a helicopter.  They were collected with a Nikon D3 camera and generally are very high quality.  I have run through my typical workflow, which is:

1.  Align photos on high
2.  Run a low geometry model for quick verification
3.  Insert ground control.  Ground control for this project were collected by a professional survey crew and should be accurate to a few centimeters. 
4.  Adjust ground control marker in images to ensure markers are properly placed
5.  Optimize model
6.  Generate Medium heightfield model
7.  Output Orthoimage and DEM (output resolution is 1.1cm on the ortho and 7cm on the DEM)

I have used 6 gcps spread throughout the image area and have about 10 reserved as check points.  When I perform my accuracy analysis on the 10 check points, I get a horizontal CE95 of about 4.4cm.  That is very good and certainly good enough for my application.  However, when I run the analysis on the DEM, I get a LE95 of about 19cm.  This is not as good as I would like and worse than I was expecting.  Given other people's experience with high resolution imagery, is this a reasonable value for elevation error? 

Also, would it help to run the model as high or ultrahigh?  I typically don't see much advantage to doing this with the historic imagery, but perhaps it is more useful with high-resolution imagery. 

Thanks,
Scott

8
General / Strange Artifacts in Models
« on: September 14, 2011, 06:34:21 AM »
Hello,

I have been trying to resolve a strange problem that I am having.  I have used Photoscan Pro successfully on many sets of aerial photography, but recently have been having a problem.  I have a set of older aerial photos (1949-era) that I am trying to generate a DSM from.  I import and align the images and all looks fine.  They generate a good number of points and the alignment seems to work properly.  When I try to build the model, however, I get the following result,


If I try to change parameters (high vs. medium alignment or low vs. medium on model build) I can sometimes get a better result, but often it has this problem.  Then some runs will just work and it will look good.  It is a very strange issue and I'm wondering if others have seen it.  Perhaps I am setting up my project wrong, but I have run many other data sets without this problem. 

Thanks,
Scott

9
General / Flattening Water Bodies
« on: June 14, 2011, 01:03:29 AM »
Hello,


I work with Photoscan with aerial data and have run into a small problem.  As is often the case, water bodies tend to show a lot of distortion due to problems correlating on these surfaces.  Is there a way in Photoscan to flatten these bodies in the model?  I would like to define an area and either give it a height value or just have Photoscan flatten from one side to the other.  Any thoughts?

Thanks,
Scott

10
General / Photoscan with Old Aerial Photos
« on: June 10, 2011, 04:10:47 AM »
Hello,


I am using Photoscan on a couple different research projects and we seem to have hit a wall that I'm hoping someone can help with.  We are using aerial photography that is about 30 years old for which we do not know that camera calibration coefficients.  We are generating orthomosaics using 4 or 5 images at a time.  Our general workflow is as follows:

1.  Image are scanned from film at 1200 dpi (approximately 1.2m ground resolution)
2.  Crop all images to remove fiducials and black border.  Images are cropped to the same dimension so that PhotoScan interprets them to be from the same camera. 
3.  Load images into Photoscan and align images at highest resolution
4.  Generate geometry using Medium resolution height field.  200000 faces. 
5.  Apply ground control matching known points in the imagery.  Our ground control is collected using a survey-grade GPS and is generally accurate to better than 20cm. 
6.  Export orthomosaics as geotiff

I believe this workflow is appropriate.  Not knowing what the camera calibration is we allow Photoscan to calculate the K1, K2, K3 coefficients.  The values it calculates are reasonable for this type of imagery, so that seems ok. 

When we analyse the resulting mosaic (approximately 250 sq km) we find that many areas have significant deviation (> 10m horizontally).  The areas where we apply ground control are significantly better, but as we move away from those areas it seems the model is breaking down.  We have used varying numbers of ground control, but going from 3 to 4, 5, or more doesn't seem to improve the result, so we have been sticking to 3 to try to limit the warping that Photoscan is doing. 

We are hoping to generate orthomosaics that meet the U.S. NMAS 1:24000 standard, which is 12.2m at CE90.  At the current time, we cannot do this . I would appreciate any thoughts people may have that might help us improve our product. 

Thanks,
Scott

11
Feature Requests / Total RMSE in New Version
« on: May 21, 2011, 12:50:59 AM »
Hi Dmitry,


I downloaded the most recent release of PhotoScanPro and am using the new ground control interface.  Seems to work well, but there is no longer a total RMSE for the ground control that you have input.  At least I was unable to find it.  It's easy enough to calculate, but is just a hassle if you have large numbers of ground control.  If that could be added back in a later version I know I would appreciate it very much.  Thanks for the great product. 


Scott Arko

12
General / Using Oblique Satellite Imagery
« on: April 06, 2011, 03:44:49 AM »
I have successfully used Photoscan to work with a number of aerial images and I am trying to move to satellite imagery.  I have a set of oblique satellite images that I would like to use to generate a DEM.  I have one image that is collected at nadir, one at -35 degrees and one at +35 degrees oblique.  They were collected in the same swath, so are nearly simultaneous.  

I have tried to run them through photoscan, but they won't even align.  Am I just missing something fundamental and these images will not work?  


Thanks,
Scott

13
General / Possible Bug
« on: March 24, 2011, 06:03:33 PM »
Hi Dmitry,


I wanted to let you know of a possible issue I have noticed.  The project I am currently working on involves about 100 relatively small, 6MP, aerial camera images.  I am running the most current version of Photoscan on a new Core i7 Macbook Pro with OS 10.6.6 and 8GB of RAM.  Image alignment goes fine, but the problem is in building the geometry. If I try to change the volume and tell Photoscan to use the selection in geometry building it will crash my MBP.  If I run the same geometry, but do not tell it to use the selection, but rather run with the Auto setting it runs fine and completes with no problems.  The crash point seems to vary and it crashes on both medium and high resolution settings.  I didn't try low.  

Not sure what the issue is, but it seems to not be the data as when I run with Auto it works fine.  Just something I thought you might want to be aware of.  Does Photoscan generate logs when it crashes?  I would be happy to send them to you if they exist.  If I am just doing something wrong and this is my fault, please let me know that also.  


Thanks,
Scott  

14
General / Volume Select during Geometry Build
« on: March 23, 2011, 12:18:41 AM »
Hello,


This is hopefully an easy question.  After I align my images and go to build the geometry I often want to use the option to change the volume selection.  When I go to select volume I get the box and can drag corners around, etc just fine.  However, the volume selection box and my area of interest are often not aligned properly.  There seems to be no way to adjust the angle of the volume selection, only the corner point positions.  Am I missing something?  Can you rotate the volume selection as well as move the corners?  

Thanks

15
General / Adding Precision Markers
« on: March 20, 2011, 10:18:39 PM »
Hello,


I have been using Photoscan for a couple weeks now and have been very happy with the results thus far.  One problem I have is with the positioning of markers.  I am ingesting aerial imagery and trying to apply ground control to them.  My workflow is as follows:

1.  Ingest images  -- No problems here
2.  Compute geometry -- no problem here
3.  Build Texture -- no problem here

When I build my texture I generally choose a single image to use as the orthoimage.  The texture looks nice and at this point I try to apply my ground control.  I have precision GPS measurements for points in my image, but what I find is that with the way Photoscan allows you to navigate the image (i.e. the trackball function) that I cannot zoom into specific areas closely enough to put down my marker in what I feel is an accurate manner.  

Am I doing something wrong?  Is it possible to apply the markers earlier in the workflow, but I don't really see how?  Any help you can provide would be great.  


Thanks,
Scott

Pages: [1]