Opticalmedia
Posts: 14
Location: Cape Town, South Africa
Registered: 20 Aug 2008
|
How to find control Points On cloudy sky? Using PTGui
Posted: 20 Aug 2008 at 22:31 GMT
|
|
I’m having a problem stitching an image in PTGui. I cannot locate accurate control points on images No. 4 and 5 of an 8 image sequence. In other words between the over lap region images 3 and 4, images 4 and 5 I cannot locate accurate control points. As there is only sky and clouds visible which definatly moved slightly while I shot my sequence.
In PTassembler there was a function I read about - something along the lines of interpolating images where locating control points accurately would be difficult. Is there a way to do this in PTGui?
Here is my workflow (LOW RES) as follows, where would it be most convenient for me to interpolate for my workflow? Also what do you think of my workflow? Could it be done more accurately i.e. in a different order?
1. Save images to jpeg from Adobe Camera RAW (All WB, bump up exposure half a stop)
2. Place images in PTGUI and create control points for all images using (under generate control points) (Is this PTGUIs own control point picker?)
3. Run optimizer
4. Reset all images parameters (yaw/pitch/roll) and lens correction parameters’ to zero (I have yet to calibrate my lens) and reset my HFOV using my cameras original EXIF data
5. I then delete all the worst control points leaving only the best 5 CP's (yaw, pitch, and roll, b and HFOV (because they're not calibrated) per image overlap
6. I then click on Advanced under the Optimizer heading. Enable yaw/pitch/roll for all images. Choose an anchor image (image0) and disable raw/pitch/roll. Optimize in two passes. For the first pass I disable a/b/c/HFOV and optimize to position the images. For the second pass, enable a/b/c/HFOV.
Now after this I usually get a great stitch and I'm sure that once I get my lens Calibrated properly for a/b/c - I will get an even more accurate stitch?
Only problem with this image as I already motioned are finding control points of the featureless sky.
Thanks for reading this, any help would be appreciated
Mike
|
|
alert moderator
|
|
Bradford Bohonus
Posts: 80
Location: United States
Registered: 29 Jan 2008
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 20 Aug 2008 at 22:53 GMT
updated: 20 Aug 2008 at 22:54 GMT
|
|
Opticalmedia said: Only problem with this image as I already motioned are finding control points of the featureless sky.
Thanks for reading this, any help would be appreciated
Mike
Is this for a single row pano with zenith & nadir? If you can place enough of them elsewhere you shouldn't need any in a featureless sky. If you are looking for placement of the zenith, you can position it manually without using any points. Otherwise you can try applying a template from another project using the same lens setup/shots etc.
|
|
alert moderator
|
|
John Houghton
Posts: 2317
Location: Hitchin, United Kingdom
Registered: 17 Jan 2005
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 21 Aug 2008 at 6:41 GMT
|
|
Opticalmedia said: Now after this I usually get a great stitch and I'm sure that once I get my lens Calibrated properly for a/b/c - I will get an even more accurate stitch?
If you usually get a great stitch, then it's very likely you have a previous project file to use as a template. This will set up your lens parameters in lieu of calibrated values. (But why not calibrate your lens anyway)? Assuming you are using a pano head, you can initialize the y,p,r values to their known nominal values. You should optimize the main row of the panorama with y,p,r checked for those images only, but uncheck y on the first image, and uncheck all other images in the Use Control Points Of list, and check the Link boxes above the p and r columns. This will force the optimizer to keep the pitch and roll settings the same for all the images in the row. Having done that, you can uncheck the link boxes and proceed normally, with image 0 of that row as the anchor (y,p,r unchecked).
Usually, an unlinked image will have predictable y,p,r values based on neighbouring ones, so these can be assigned manually. In the case of a zenith sky image, you probably rotated the camera upwards from the last horizontal shot, in which case its yaw value will be the same as the previous shot. The pitch will be +90. That should be good enough, given that the sky will have changed anyway so a perfect match will be impossible.
Finally, the whole panorama can be levelled using t1/t2 points and the Level Panorama option on the Edit menu of the Panorama Editor window. Or if there are no verticals or horizon for t1/t2 points, you can level it visually by dragging the whole image or by making adjustments using the Numerical Transform option. These operations will not disturb the alignment of the images with each other and keep the unlinked image in place.
John
|
|
alert moderator
|
|
Opticalmedia
Posts: 14
Location: Cape Town, South Africa
Registered: 20 Aug 2008
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 21 Aug 2008 at 13:58 GMT
|
|
Hi john Im newsih to Pano assembling, So I dont quite understand what you are telling me... What i have now done under image parameters is i have set the pitch and roll of images 4 and 5 (4 i set the otch and roll to image 3 and 5 i set the pitch and roll to image 6)I then maually in the panorama editoraligned the images up - the ones i could not generate control points with. What should i do now with regards to optimization?
Could you explain? I did not quite get what you were explaining.
Many Thanks
|
|
alert moderator
|
|
John Houghton
Posts: 2317
Location: Hitchin, United Kingdom
Registered: 17 Jan 2005
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 21 Aug 2008 at 15:38 GMT
|
|
Can you explain exactly how the images were shot for your panorama? i.e. is it a single row 360 degree panorama, or multi row , or a 360x180 (x around +zenith and nadir) etc. It's difficult to know how to advise in detail without knowing exactly what it is you are trying to stitch. If it's a single row, then one image would not be expected to be completely sky.
John
|
|
alert moderator
|
|
Opticalmedia
Posts: 14
Location: Cape Town, South Africa
Registered: 20 Aug 2008
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 21 Aug 2008 at 22:19 GMT
|
|
OK I shot 8 pictures in portrait position using panosaurus head calibrated to my canon 30D and canon 17-55 F2.8IS lens @17mm which is equivalent of 27mm due to my 1.6X crop factor.
My pano ran from the mountain peak across the clouds along the horizon.
This is how my image stitch up in the end. I blended to versions from one negative and created a HDR version using photomatix.
But my what i would like to know from you is should how should i have actually run my workflow?
critique of my image from you would be highly appreciated
THANKS
|
|
alert moderator
|
|
Hans Nyberg
Posts: 866
Location: Denmark
Registered: 28 Aug 2005
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 21 Aug 2008 at 23:11 GMT
|
|
The problem is that you have very little overlap.
To get enough for automatic controlpoints you need at least 15%
The best is 25-30%
That means 10 images around.
With 8 you only get 7% overlap which is very little also for the blender.
You will need to set manual controlpoints.
Use Franks calculator to find the numbers for any lens
www.frankvanderpol.nl/fov_pan_calc.htm
Hans
|
|
alert moderator
|
|
Opticalmedia
Posts: 14
Location: Cape Town, South Africa
Registered: 20 Aug 2008
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 22 Aug 2008 at 0:21 GMT
|
|
Thanks very much Hans, photo.net/photos/Michael Suttner Here is where I have placed my image. I figured out a new method to find those control points.
I first searched for control points on all my images. Image 4 and 5 (hence btw 3 and 4, 4 and 5, 5 and 6) I got no control points. Strait after this I clicked find control points on all images using auto pano and it now managed to find the control points. Strange
anyways let me know what you all think of my final stitch
And thanks for your generous good help
|
|
alert moderator
|
|
John Houghton
Posts: 2317
Location: Hitchin, United Kingdom
Registered: 17 Jan 2005
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 22 Aug 2008 at 5:42 GMT
|
|
The link to the panorama image does not work. From the description, it may not be a full 360 degree panorama.
John
|
|
alert moderator
|
|
Bob Stone
Posts: 41
Location: Rochester, NY, United States
Registered: 20 Oct 2007
|
|
John Houghton
Posts: 2317
Location: Hitchin, United Kingdom
Registered: 17 Jan 2005
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 22 Aug 2008 at 7:27 GMT
|
|
Thanks for the link. Yes, it's a very nice panorama indeed. There do seem to be suitable features for control points on all the images at the bottom at least. Manual assignment should not present a problem. Possibly a row of t3 points might be practical on the horizon, too.
John
|
|
alert moderator
|
|
Opticalmedia
Posts: 14
Location: Cape Town, South Africa
Registered: 20 Aug 2008
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 22 Aug 2008 at 9:31 GMT
|
|
Thanks for having a look. John the problem with assigning control points manualy in the middle is that the sky is featurless, the trees at the bottom are not static and the cloud also moves. But i have found a way around this but trial and error. Im not sure if it is a known method out there let me explain.
I'l generate control points on all my images using the PTGui control point generator. I set it up to generate 40 CPs per image overlap(which is exsessive I know). I then run the optimizer appropriatly. I then accpet the results and reset my lens settings and image parameters. At this point I dont have any Cps between 3/4 4/5 5/6 - as my original problem. But now all i do is run autopano on all my images and only then does it pick up CPs on the other images.
Following this I delete all the worst CP's per image over lap leaving only the best ones which happen to be incredibly accurate.
This works for me. With reagrds to the t3 points i have not looked into that yet. Does it straighten the horizon? surely i can just set that straight in the Pano editor
Thanks to all
|
|
alert moderator
|
|
John Houghton
Posts: 2317
Location: Hitchin, United Kingdom
Registered: 17 Jan 2005
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 22 Aug 2008 at 11:20 GMT
|
|
There are various ways of optimizing your images. You can make good use of information you know about the shoot.
All the images were taken at the same angles of pitch and roll. You simply rotated the pano head round to change the yaw. You can tell the optimizer that any solution it comes up with must have the pitch and roll the same for all the images (using the link options already described). In addition, you need a minimum of only one control point assigned between each pair of images. That is sufficient to generate a good alignment of the images, provided that you also have good lens parameter values established. These can be automatically evaluated using several control points between some of the images in this run, but could alternatively be evaluated more accurately in a separate calibration run.
When you have a good clear view of the horizon, a set of t3 line control points will serve to align the horizon in each image along a single straight line - like hanging sheets on a washing line. In effect, they then only have the freedom to slide sideways along the line. Again, just one control point between each pair is needed to peg them together so that they cannot move sideways with respect to each other. A pair of t2 points at either end of the line will get it horizontal. Lens parameters as above.
I'm not suggesting that you do actually only use one control point per overlap, but when necessary, you can make do with only one. Slight misalignments will probably not be visible after blending if the points are a just bit iffy.
It's all about marshalling your resources and using them intelligently. Of course, there are other solutions, as you yourself have found.
John
|
|
alert moderator
|
|
Opticalmedia
Posts: 14
Location: Cape Town, South Africa
Registered: 20 Aug 2008
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 22 Aug 2008 at 14:04 GMT
|
|
John, with regards to the pitch and roll of my images as I composed them on the Panosaurus is that as i rotated the head the spirit level bubble moved slieghtly out of place as it is so sensitive to movement and because i dont think the base of my Pano head is 100% smooth. It even mentions this in the manual that came with my pano head.I therefore allow the optimizer to calculate the correct yaw and pitch values based on the control points I have selected. Would you say that this is correct? How many Pano heads out there can actually rotate 360 with a sleight change in pitch and/or Roll? Therefore im assuming most people optimise for these values 90% of the time.
For Calibrating my lens accurately could you share with me a link you feel shows the most accurate way ? Are there already values out there for the canon 17-55 f2.8 IS lens that i can simply fill in or is each lens differnt in this regard and therefore each lens needs calibrating? Does the body( canon 30D)attached to the lens affect the calibration of my lens?
I really appreciate all this help,
Mike
|
|
alert moderator
|
|
John Houghton
Posts: 2317
Location: Hitchin, United Kingdom
Registered: 17 Jan 2005
|
Re: How to find control Points On cloudy sky? Using PTGui
Posted: 22 Aug 2008 at 22:28 GMT
|
|
Yes of course people generally optimize pitch and roll individually rather than linked, but they can be linked temporarily in order to straighten a panorama. You need control points in order to optimize the parameters individually, and it's when you have a severely limited number of control points that linking pitch and roll can be useful. Even with your Panosaurus, you may well find that pitch and roll are reasonably constant as you pan round. Temporarily linking them will have an averaging effect anyway.
Calibrating the lens has been discussed many times before: e.g. in thread www.panoguide.com/forums/qna/4340/
Usually this is in connection with full 360x180 spherical panoramas. You can shoot just a 360 degree single row for the calibration if that's how you intend to use the camera. It will yield an accurate horizontal angle of view parameter for the lens, which a partial panorama (<360) might not do. Contrive a 50% overlap for some of the images and 25% for others.
Lens parameters will vary for different sensor sizes, and the shift parameters d and e will certainly be different for different cameras. They don't seem to be very constant even for the same lens and camera owing perhaps to tolerances in the fit of the lens in the mount and maybe other factors. I always optimize them for fisheyes and often also for rectilinear lenses. Be alert for silly values being generated by the optimzer in unusual circumstances. Reset them to 0 if that happens. The zoom setting will also affect the lens parameters.
John
|
|
alert moderator
|
|
back to threads list |
this thread is closed
Prev
