panorama software,virtual tour software
Alternate Text
Joined: 2003-08-26
Send Msg:
Posts: 34
2005-09-24
#1

Perfect stitching with PW4+TCPS+20D+8mm ?

I have the following equipment : Agnos MRotator TCPShort + Eos20D + Sigma8mm, and still trying to get perfect results with PW4. And I always encounter stitching problems with the bottom on linear or tiled floor...

I'm quiet sure my nodal point is good.

Where does that come from ? The TCPS rotator ? The software ? The fish ? The camera ? The parameters in PW4 ?

Even in the PW4 new gallery, you can observe stitching problems on some samples, on the floor (restaurant QTVR for example).

Does anybody get perfect stitching results (without retouching the pano later, of course) with this equipment ? I would be glad to see an example...

Even in the PW4 new gallery, you can notice stitching problems on some samples, on the floor (restaurant QTVR for example).

In PW 3, there were more stitching parameters, and I think the result was more accurate...


Phil, French Guiana
Alternate Text
Joined: 2005-05-15
Send Msg:
Posts: 372
2005-09-24
#2

Hi Phil,

There is of course many things that can cause stitching errors even if your nodal point is perfect. But since you asked the question of how to get "perfect" results we must look at what it takes to get a "perfect" stitch. And since we are talking about "perfect" we must look into the situation with a little more detail.

In fact, it is not possible to rotate "exactly" on the axis of the "true nodal point". We are actually rotating around the perimeter of the nodal point. The nodal point is the point at which light converges inside the lens. This IS a point. But the "entrance pupil" is round and is at a vertical orientation. Take a coin and rotate it on its edge. You will notice the outside edges of the coin rotate in an arc with this arc decreasing towards the center. So you are actually rotating around both the entrance pupil and the nodal point together. Although the distance from the entrance pupil and the nodal point is very small it is enough to cause light waves to be slightly shifted as you rotate the camera.

With this inherent aberration in mind realize that stitching errors will occur but they are normally hidden due to the blending and accuracy of the software used.

What happens is any other factor that can cause a stitching error will magnify this inherent problem.

The most important of these possibilities is the level of the camera. It must be perfectly level horizontally as well as vertically.

Horizontal level is easy to obtain by using a steady tripod and added tripod leveler. It should be noted that you should never re level the camera at any point. Once you have established level on the first position of the frame set that it where it should stay.

The camera itself is mounted vertically and one should use a hot shoe bubble to level the camera. Both the spirit level on the Agnos bracket and the hot shoe should show level before you begin to shoot.

The quick release mount that attaches to the camera is secured with a screw but is held in place by a cork friction plate. It is possible to cause slight vertical rotation if you are not very careful when pressing the shutter release. When rotating for the top and bottom it is easy to grab the camera and when pulling up or down and slightly move the quick release plate.

These points are related to "mechanical" issues that are operator error. Now, I am splitting hairs here but (we are talking about perfect) it does happen especially if you are shooting a lot of scenes at once. You should check every thing after a few scenes to make sure all connections are secure. If the tripod leveler unscrews a little then you will have a slight unbalance with the rotator. So check everything.

The "scene" you are shooting will also effect stitching. You may notice that in scenes with lots of detail and contrast the stitching will come out perfect. On other scenes with less detail or with large areas of solid color or even large areas with texture like sand, grass or trees the image will have more stitching errors.

A water scene on the shore where there is all sky and a constant level of texture like sand or grass all around you will cause stitching errors more than and interior shot with lots of detail. The software relies on sharpness to calculate matching points. When there is less detail (contrast between pixels) it is difficult to accurately calculate these points automatically.

Although your lens has a huge depth of field the DOF can be moved slightly depending on the aperture selected. IF you are not using infinity focus you will not be shooting at the hyperfocal distance. If you focus on something farther away the image will still be in focus but you will be moving the DOF ever so slightly and this translates into the image being rendered onto the CCD in different levels of sharpness.

NOTE: To be even more specific. This phenomenon is related to the circle of confusion. See Daves post on what is COF and DOF anyway?

This is very, very small and would not be a factor in still photography. But IF you have one or two frames that this happens on, the sharpness between pixels makes detail areas located by the software bigger or smaller due the slight unsharpness of the pixels picked. This changes the locations of the matching points between frames. The software thinks they are in the exact position but they are in fact not. This will cause a stitching error but should only be noticeable on the edge of something sharp like a baseboard. On solid surfaces or areas with a matrix of detail you won't notice it.

We also can't rule out temperature changes either. If you are inside for a long time at 72 degree temperature and move outside to 95 degree temperature your tripod, mounts and the bracket are going to expand some. This expansion and contraction will technically effect the nodal point position. It will take a few minutes for the equipment to stabilize. If you start shooting immediately you will have different nodal points on different frames or event different nodal points on different scenes.

For the software to stitch a mathematically perfect stitch each frame must be rendered in exactly the same position on the CCD. Anything that changes this will result in some error at least at the math level.

Tolerance is what we are talking about. What is the error factor of the software? There is one of course but this error is within the acceptable range of "perfect". Anything that increase the inherent error factor to bring it out of the acceptable range will result is some amount of stitching error. Whether you can find it or not is another question. As the errors get larger it gets to the point where you can see it.

I really think that most stitching errors result from leveling problems and details in the image. The sharper the image is the better the stitch will be. Aperture plays a big role in this. With the sigma at 8mm you should be shooting in the F7 F8 range and controlling exposure with shutter speed.

General Lee.

 

 


Alternate Text
Joined: 2003-08-26
Send Msg:
Posts: 34
2005-09-25
#3

WOW General !

What for a lot of detailed explanations ! Thanks for your answer, but even if the temperature may move the nodal point (!), I think that that the human factor and leveling adjustements are the keys of the problem.

When I mean a prefect stitching, I mean the kind of stitching you can get with the Precision Pro, (what they say they get, its very hard for me to believ that they shoot and then stitch and nothing more)

You can see one example on the following link : http://www.360precision.com/360/popup/dsp_fullscreen_bo_1.cfm

(there are other samples on the home page)

I never got a so "perfect" stitching without correction with my agnos TCPShort and PW4...

Do you think this quality can be obtained, once again without any correction, with the TCPShort + Sigma8mm + 20D + PW4 ?

Another question, they say you can downloaf the sets of original images, (http://www.360precision.com/360/downloads/batch_test/bo_node_01.zip for example), what I did, and then try to stitch by yourself. I tried in PW4, it was a... disaster ! and the images are very dark... Did anybody success stitching these photos in PW4 ? Maybe these photos are not "fullframe" ?

Phil

 


Phil, French Guiana
Alternate Text
Joined: 2005-05-15
Send Msg:
Posts: 372
2005-09-25
#4

Hi Phil,

Yes, I write long posts LOL. But I enjoy the topics on this fourm and find it interesting to do the research. I'm just a geek I guess.

I downloaded the test shots and did a stitch. The test images are horrible. They are not exposed properly and the scene consists of all solid walls with no detial anywhere in the room. This is the worst case scenerio for stitching. Calculating matching points requires definition or sharpnes in detail among pixels. When the scene is like this there is nothing to for the software to calculate. If you look at where the matching points are calculated you will see they are always at the edge of sometning dark and light. If they are not the software is guessing. They really need to get a better set of test shots.

I looked at the full screen example. LOLOL. It is interesting they picked that room. On the ceiling there are clearly stitchng errors at the tile edges. The reason you can't see any other errors is because the walls and floor are a solid tone of color without any detail. They did not shoot a bottom either. Just covered that with a cap. They also did some serious color correction to the final image.

The scene stitched ok with only noticable problems on the ceiling. I suspect the same problem would have occured if the floor had been shot.

But I had to manually pick the matching points And I think this is related to type of scene. I could have spent 15 or 20  min manually setting matching points at the pixel level and probably gotton a perfect stitch.

I began to think the images were shot out of order. You have to shoot clockwise then top then bottom. If you do it counter clock wise it wont work.

You asked if you can get "perfect" stitching with the Agnos and the answer is a big YES.

The Agnos is precisly engineered and quite strong enough to suppor camera and lens. As I have said many times before it is not hard to locate the nodal point. It is a matter of uderstanding a few basic principles. I had never done it before when I got my Agnos TCP short. Yet on the very first try I got it. Many complain about difficulty in finding it because they see errors in the stitching and immediatly blame the nodal point. There are many other factors that effect stitching errors and most of them are operator error not the nodal point.

The nice thing about the Agnos is that you CAN adjust the nodal point if necessary and even use different camera setups with the same setup. No so with the P360. Yes you can get camera upgrade kits. Blah. One rotator is all you need.

Below is a pano I took just 3 days ago.
Nikon D70s - Nikkor 10.5mm - AgnosTCPshort
3 bracket set - F5.6- 1/4,1/2,1" - Photomatrix Combine H&S details Auto
Panoweaver 4.0 auto stitch with no manual point setting.
http://northwestflorida.com/360/test/sales_desk_4000.html

I did repair two small errors on the floor but that was related to my bumping the tripd on the rotation to the floor. Othere wise the image is untouched. I have looked and looked at this image to find any errors on the horizon and up top but I can't. There may be some but only you and I are going to find them not the consumer or client. So I guess I would have to ask is this perfect "enough"?

This is the 360P test shot I stitched.
http://northwestflorida.com/360/test/360_pre.html

The test images are only 260k. Not big enough to get full screen quality. Even the errors on the ceiling are not that bad at the givin display size. I would  give this a "good" rating on the stitching.

I know everyone desperatly wants a setup this is a "wham bam thank you mam" deal. Shoot everything in 30 minutes, go home and stitch 10 panos is 30  minutes and get perfect results.