Morten Andersen
Posts: 60
Location: Roskilde, Denmark
Registered: 12 Jun 2006
|
Perfect auto stitch - a myth?
Posted: 6 Jul 2008 at 1:04 GMT
|
|
Hello everybody.
I am shooting a lot of panoramas for hotels at the moment, and it is comming too time consuming having to correct stitching errors in PTgui and PS. I am therefore considering to buy a new pano head - IF it will actually improve my workflow.
At the moment I am shooting with a canon 5d, sigma 8mm f3.5, Agnos TCPshort with RingT and I have to correct stitching errors almost every time.
I know that many people will most likely advice me to look at the No Paralex Point before buying a new panohead. But I doubt I can get a better NPP then I have already. Not that it is perfect, but.......... well let's just say that I have been spending a lot of time correcting the NPP, and I don't think I am able to get it any better.
So... I have been considering buying a 360precision. But..... is it as good as they claim? How easy / hard is it to find the NPP with a 360 percision? How likely is it that I won't have to correct stitching errors with their rotator?? And lastly - what rotator is the best for my setup??
Thank you all so much in advance 
Best regards
Morten Andersen
|
|
alert moderator
|
|
Ranjan Pano
Posts: 46
Location:
Registered: 20 Jun 2006
|
Re: Perfect auto stitch - a myth?
Posted: 6 Jul 2008 at 3:01 GMT
updated: 6 Jul 2008 at 3:03 GMT
|
|
Perfect auto stitch is not a myth but requires lot of understanding of the panorama technology.
To understand where the problem is we need to see the errors can you post a set of images so members here can stitch & find out the real issue.
Buying a pano head may be your priority for shooting hotel panorama but is not essential, the problem may be somewhere else which we need to figure out.
I still use a custom made head & it took a good while to calibrate & I shoot interiors too & can understand your frustration with misaligned images.
|
|
alert moderator
|
|
John Houghton
Posts: 2269
Location: Hitchin, United Kingdom
Registered: 17 Jan 2005
|
Re: Perfect auto stitch - a myth?
Posted: 6 Jul 2008 at 6:06 GMT
|
|
Morten, There's no reason why you should not get equally good results with your present setup as you would with the 360Precision head (which one?). If you want to stitch without control points using the latter, you would need to perform a calibration exercise to produce a template. However, you can do the same for your current setup (even though you need control points), which should make stitching much more reliable. As Ranjan says, a sample set of images together with your project file should enable the cause of the stitching errors to be diagnosed.
John.
|
|
alert moderator
|
|
John Hawkins
Posts: 15
Location: Las Vegas, NV, United States
Registered: 11 Dec 2006
|
Re: Perfect auto stitch - a myth?
Posted: 6 Jul 2008 at 6:49 GMT
|
|
Myth. Perfect auto stitching either from an automatic control point and alignment tool or batch stitching using a supposed 'precision' panorama head is not reliable. Unless you are doing low resolution real estate work, a precision head is a waste of money. Why is neither reliable?
Let us start with automatic alignment. Most current automatic matching tools depend on finding unique features in images. Several things will cause automatic matching tools to fail. One cause of failure is if there are moving targets in the scene. The moving target may be matched by the feature finder but if it has moved relative to the rest of the image, the result will be the feature (moving person, etc.) will be matched between images but the parts of the image that have not moved may be missed. This is especially true if the subject is of a moving target (a boat perhaps) on a featureless background (smooth water). Empty blue skies will also fail to match as they have no features. Another common failure of auto match point finders is in scenes that have repeating patterns such as is often found in mosques. Auto feature finders often mistake say the light fixture above one door with an identical light fixture above another door.
As for 'precision' panorama heads, they can work for low resolution panoramas but will not for anything else. Take for example a Nikon D300 with the Nikor 10.5mm fisheye lens on it. The lens has an ~87 degree field of view in the small edge direction. The nikon D300 creates images 4288 x 2848 pixels which equates to 33 pixels per degree (2848 pixels divided by 87 degrees) or 0.03 degrees per pixel (87 degrees divided by 2848 pixels). You won't find any claim of accuracy on 360Precision's site (at least I haven't) but I doubt their heads are repeatable to 0.03 degrees. However, if you shoot low resolution panoramas, the precision needed for reproducible results and batch stitching is lower. In addition, modern blending routines (e.g. smartblend) are able to mask small errors by varying the seam locations so the seam is placed where the images match but that has nothing to do with a panorama head's precision.
So what are the keys to making a perfect stitch? I think you hit on what is your likely problem. Your 'no parallax point' is probably not correctly set. Note that fisheye lenses do not have 'no parallax points'. What you need to find in the case of fisheye images are the adjustments so the images align where the seam is going to be placed. Once you do find the proper camera mounting location, mark it on your panorama head so you can remount the camera in the same location on your next shoot. Rectilinear lenses do have 'no parallax points'.
A sturdy panorama head is a big help as is a sturdy tripod and a solid footing. The various 'precision' heads are generally sturdy but you can also find sturdy panorama heads for far less cost. Note that you don't even need a panorama head if the subject matter is distant.
Good luck.
|
|
alert moderator
|
|
irieman
Posts: 140
Location: East Sussex, United Kingdom
Registered: 8 Jul 2006
|
Re: Perfect auto stitch - a myth?
Posted: 6 Jul 2008 at 7:29 GMT
|
|
Yes a 360 Precision is as good as the makers claim. If you use the Absolute ( actually the earliest version which didn't have a name) like I do you don't have to set the NPP as the arm is calibrated for your body/lens combination.
Yes, you will have to make an accurate lens calibration and then an accurate template - and once you have done that you can batch stitch with accuracy.
I use a 5D with a 15mm for my hi rez panos and particularly when I am shooting for .hdr (I mean true HDR for lighting probes in C4D - not what is erroneously called HDR) - I think you will know what I mean. That in itself is enough to justify buying the 360 P.
Of course 360 P products are expensive but then so is your time - I bought mine very shortly after they were announced and have never regretted it for one moment.
Usual disclaimer - I don't have any connection with the company other than as a satisfied customer.
|
|
alert moderator
|
|
mhc1
Posts: 188
Location: Hamburg, Germany
Registered: 29 Dec 2005
|
Re: Perfect auto stitch - a myth?
Posted: 6 Jul 2008 at 19:36 GMT
updated: 6 Jul 2008 at 19:39 GMT
|
|
Morton,
I changed to a ring - would be the Atomic or Agnos I guess (mine is "handmade" by a friend)- and increased accuracy a lot.
I love using a monopod now: just a fast exposure/focus setting and then: only the spirit level in view turning & shooting 4 images.
I did some 30-40 panos last week, some belly-belt poles and some cliff-hangers too.
Nearly all - even the belly-belts- are 95% perfect with very little retouche. Some even didn´t need a Zenith retouche. The Nadir I covered with a logo - fine.
Creating CPs and optimizing - takes only seconds.
I have to redo my lens parameter though, John Houghtons set is much better then mine  .
But this I´ll do when I have a apropriate object.
So: a very economic workflow finally.
I had much more work before with a NodalNinja- but the fault was not the NN, it was surely me.
So I can´t see, that a new or other NPP- adaptor would increase your workflow. It´s as good as it can be.
Ciao
Mike
www.360de.de
|
|
alert moderator
|
|
DennisS
Posts: 51
Location: Los Anglels, United States
Registered: 1 Sep 2007
|
Re: Perfect auto stitch - a myth?
Posted: 6 Jul 2008 at 20:30 GMT
|
|
Perfect stitching is not a myth. Your basic setup needs to be correct before you start taking pictures.
Post some pictures so we can all have a crack at stitching your samples. A picture is worth a thousand words.
|
|
alert moderator
|
|
Mark Schuster
Posts: 714
Location: Welwyn Garden City, United Kingdom
Registered: 25 Jan 2006
|
Re: Perfect auto stitch - a myth?
Posted: 6 Jul 2008 at 22:36 GMT
updated: 6 Jul 2008 at 22:46 GMT
|
|
Morton,
You have had a lot of experts reply but none have asked how many shots around you do. However many it is, why not try increasing it by one. This will increase overlap considerably and make things easier for your stitching programmes. Am I being simplistic? Maybe, but isn't it worth a try before you lay out more cash?
Oh, and by the way, have you tried Autopano Pro? I get away with really shoddy alignment, especially of the nadir which I shoot handheld, and as often as not get a perfect (well near perfect) stitch. Here's one I did yesterday and I didn't even have to remove my boots and belly in an alpha channel. (But that only works sometimes)
panoradiant.co.uk/reza/karmann-1.html
Mark
PS There was a small stitching error, a break in the black extension lead snaking across the floor. Quickly corrected in PS Cut,Paste,Transform. If you look hard enough you will find a remnant of the broken cable.
PPS And I've just looked up, and lo and behold the corrugated, semi-transparent, skylight has errors. Had there been more contrasty detail I suspect this wouldn't have happened.
|
|
alert moderator
|
|
Morten Andersen
Posts: 60
Location: Roskilde, Denmark
Registered: 12 Jun 2006
|
Re: Perfect auto stitch - a myth?
Posted: 7 Jul 2008 at 3:38 GMT
|
|
Hello everybody
Tank you all for the response.
I am still on "vacation" but I will be back in front of my work station monday or tuesday - and then I will of course gladly upload some images for testing.
@Ireman
Calibrating the lense and making a template - how difficult is that??
Best regards
Morten
|
|
alert moderator
|
|
mediavets
Posts: 420
Location: Isleham, Cambs., United Kingdom
Registered: 8 Feb 2008
|
Re: Perfect auto stitch - a myth?
Posted: 7 Jul 2008 at 9:51 GMT
|
|
Mark Schuster said: Here's one I did yesterday and I didn't even have to remove my boots and belly in an alpha channel. (But that only works sometimes)
panoradiant.co.uk/reza/karmann-1.html
Mark
Love that Karmann Ghia - is it yours?
Andrew
|
|
alert moderator
|
|
John Houghton
Posts: 2269
Location: Hitchin, United Kingdom
Registered: 17 Jan 2005
|
Re: Perfect auto stitch - a myth?
Posted: 7 Jul 2008 at 11:26 GMT
|
|
Morten Andersen said: Calibrating the lense and making a template - how difficult is that??
This has been discussed many times before. See www.panoguide.com/forums/qna/4340/ .
John
|
|
alert moderator
|
|
irieman
Posts: 140
Location: East Sussex, United Kingdom
Registered: 8 Jul 2006
|
Re: Perfect auto stitch - a myth?
Posted: 7 Jul 2008 at 13:34 GMT
|
|
Here is what I regard as the definitive guide to lens calibration from Kevin Kratske of Kekus Digital as posted on the PTMac forum some time ago. Obviously this relates to using PTMac but the process would be similar if using PTGUI.
I've emailed you the 360 P guide to template construction.
Are these easy to do? - depends on how you define easy - both require some patience.
Calibrating your lens
Accurately determining your lens distortion parameters (a, b and c) is a key factor in making panoramas that need little retouching after stitching. Calibrating your lens is easy and will save you a lot of time when creating new panoramas. After you calibrate your lens, you need only set two or three control points per image pair to obtain a good panorama. You can also use your lens parameters in the LensFix and Correct plug-ins. In fact, this calibration method is the best and easiest way to determine the settings for LensFix and Correct.
While these instructions are long, they contains a lot of background information on how PanoTools works which should help you in creating your own tricks. Actual lens calibration is only a little more time consuming than creating a typical panorama. This tutorial is divided into the following sections:
Basic concepts - A discussion of the lens parameters we will be determining.
Photographing the calibration scene - Describes the type of scene we want to photograph for lens calibration and other scene requirements.
Setting up the calibration project file and selecting control points - Special settings useful for calibrating lenses.
Optimizer strategy - How to use the PanoTools optimizer to determine your lens parameters.
Using a calibrated lens in PTMac - Shows how a calibrated lens reduces the work needed to create future panoramas.
Basic concepts
PanoTools contains tools that correct for most imperfections in cameras and lenses. This "How to" will try to explain how best to take advantage of those correction tools. In calibrating your lens, we will determine the following settings:
fov - the horizontal field of view of your lens
lens parameters "a", "b" and "c" - these parameters correct for distortion present in all lenses. All three parameters operate on the entire image but concentrate their effects on different regions of the image. The effect of the "a" parameter is most pronounced on the outer region of the image, the "b" parameter effect is relatively evenly distributed throughout the image and the "c" parameter concentrates it's effect near the center of the image. Of the three parameters, the "b" parameter is the variable that corrects most common lens distortion and should be the only lens parameter you optimize if you do not have a lot of image overlap. Positive lens parameters contract the image (cause a pincushion effect). Conversely, negative lens parameters expand the image (cause a barrel effect). Combining negative and positive parameters results corrects for "wavy line" distortions.
Image shift parameters "d" and "e" - Image shift parameters adjust for image sensor offset. Due to variations in the manufacturing process, few image sensors are perfectly aligned with the axis of the lens.
Image shear parameters "f" and "g" - these parameters account for scanning shear and do not apply to digital images. If you are shooting with a digital camera, leave these at zero. If you are using scanned film, only optimize "f" or "g" but not both.
Photographing the calibration scene
Ideally the scene you use for calibrating your lens should be an exterior scene having a good number of vertical lines and window corners or other distinctive point features. Part or all of the scene should be photographed with the images having a 60 percent or more overlap. You only need shoot a single row of images.
Here are the reasons for the choice of scenes. An outdoor scene is preferred to minimize parallax errors resulting from your camera and lens not being rotated around it's nodal point. The vertical lines in the scene aid in determining the lens parameters and are used to set the roll and pitch of the anchor image. Only a single row of images is necessary and is desirable to minimize rotational errors especially if you don't have a spherical panorama head. Finally, the window corners and distinctive features will be used to set control points.
Setting up the calibration project file and selecting control points
Set up the project file as usual with the following settings:
Lens Settings tab: Set the lens parameters a, b and c to 0.00001. Set image shift and shear settings to zero. Set the horizontal field of view as appropriate.
Panorama settings: Choose equirectangular or cylindrical and 360 degrees as the field of view. Set all other settings as desired.
Crop: As needed.
Image parameters: Set the initial guesses for yaw, pitch and roll as appropriate.
Control Points: We only need to set a minimum of control points to link most of the photos. However, in two or three "control point rich" areas, pick control points with the following goals:
a. Set five or more control point pairs spread over the entire overlap area.
b. Since your images overlap by more than 50 percent, the same point is available on three images. Pick the same control point location on each of the three images. For example, images 5, 6 and 7 will have some image area that is the same on all three images. Pick a point that is the same on all three images and set control point pairs on images 5 and 6, 6 and 7 and 5 and 7 that are the same location.
c. Set two or more vertical lines approximately 90 degrees yaw from each other. These vertical lines provide the information necessary for the optimizer to determine the roll and pitch of the anchor image so that the roll and pitch of all images is determined by the optimizer. Only the yaw of one image (the anchor or control image) is fixed. When setting vertical lines, set the control point type as "horizontal" optimization.
d. You can also set some vertical lines in the images to help determine the lens distortion parameters. For these lines, you want to select a point near the top of the line and one midway down the image where the line is bowed the greatest amount. If you choose two points at the top and bottom of the image where the horizontal position of the line is the same, the optimizer sees nothing wrong with the line as a vertical line can be drawn between the points.
Optimizer strategy
Optimization is best done in stages. Think of the results of the equation you are optimizing as a mountain range. If you take small steps, you may think one of the first mountains you climb is the tallest peak and stop there. The idea is to first optimize the parameters that have the greatest impact on the result (i.e. get close to the tallest mountain), then optimize the more delicate parameters to fine tune the results. Optimize in stages as follows:
a. Optimize the roll and pitch of all images, the yaw of all but one image.
b. Optimize as in "a" above plus fov. Go back to the control point tab, press table and check for any control points that are not properly located. Be careful with the results for the vertical line control points as they include the vertical difference between the control points. Delete or reset any erroneous control points.
c. Optimize as in "b" above but also include the "b" lens parameter in the optimization.
d. Optimize as in "c" above adding lens parameters "a" and "c" to the optimizer list.
e. Optimize as in "d" above adding optimization for horizontal and vertical offset.
Check your results on the Lens Settings tab. Are they reasonable? In general, "a" and "c" lens parameters should be small (0.00x) or less. The "b" parameter is larger (in the 0.0x range or possibly larger). Horizontal and vertical image shift values are given in pixels and will depend on the resolution of your image and camera model. Coolpix cameras may have 50 pixel offsets where the sensor placement of professional grade cameras should be more centered.
|
|
alert moderator
|
|
Mark Schuster
Posts: 714
Location: Welwyn Garden City, United Kingdom
Registered: 25 Jan 2006
|
Re: Perfect auto stitch - a myth?
Posted: 7 Jul 2008 at 22:18 GMT
|
|
Mediavets said
Love that Karmann Ghia - is it yours?
Not exactly, Andrew. Bought it for my son soon after his seventeenth birthday when he past his driving test. He turned 40 a couple of weeks ago, so he (we) have had it a long time. The T-reg number plate is no indication of its age as that was when it was first registered in the UK. I think it's a 1972 model.
But my point in posting this panorama, which took a couple of minutes to set up and a lot less to shoot, was to demonstrate the amazing way APP together with Smartblend assembles 7 not very carefully aligned shots including the handheld nadir. Not the perfect auto stitch Morton asks about, but with a bit more care about lens alignment it probably could be. And APP/Smartblend together with Pano2VR don't-half make life easy.
Mark
|
|
alert moderator
|
|
mediavets
Posts: 420
Location: Isleham, Cambs., United Kingdom
Registered: 8 Feb 2008
|
Re: Perfect auto stitch - a myth?
Posted: 8 Jul 2008 at 9:40 GMT
|
|
Mark Schuster said: But my point in posting this panorama, which took a couple of minutes to set up and a lot less to shoot, was to demonstrate the amazing way APP together with Smartblend assembles 7 not very carefully aligned shots including the handheld nadir. Not the perfect auto stitch Morton asks about, but with a bit more care about lens alignment it probably could be. And APP/Smartblend together with Pano2VR don't-half make life easy.
Mark
Yes, I'm an APP/Smartblend and Pano2VR user too. I'm all for an easy life - as Shirley Conran once said (in an entirely different context)'Life's too short to stuff a mushroom'.
Andrew
|
|
alert moderator
|
|
Morten Andersen
Posts: 60
Location: Roskilde, Denmark
Registered: 12 Jun 2006
|
Re: Perfect auto stitch - a myth?
Posted: 9 Jul 2008 at 15:50 GMT
updated: 9 Jul 2008 at 15:52 GMT
|
|
Hi Guys
Thank you for all of your help so far.
Some of you have requested a set of images for testing. I have therefore uploaded a file contaning 5 images. 4 at every 90degree and one up.
You can find the files in a zipped folder at www.wideeyes.dk/temp/test-stitch.zip
I sometimes leave out the "up" shot when stitching. It is often the ceiling that is difficult getting to stich correctly. Even when I ad control points.
If anybody can tell me anytjhing helpfull from these testing images - or if you need additional images for testing, please let me know
Best regards
Morten
|
|
alert moderator
|
|
Prev
