Number of Images for a Spherical Panorama

# Thread: Number of Images for a Spherical Panorama

1. ## Number of Images for a Spherical Panorama

I would be interested to hear (read) the opinions and experiences of fellow Spherical Panorama Photographers on the subject of the number of images for a panorama, if they are prepared to share it.

I have read through Nick’s topic “How to choose a detent interval (shooting interval) updated” and followed the link to Frank’s calculator:
www.frankvanderpol.nl/fov_pan_calc.htm
and noticed that the default overlap value is 25%.

I have also followed Nick’s and John’s links to:
www.vrwave.com
and other links including:
http://www.cambridgeincolour.com/tut...rojections.htm
which has a 20% overlap as a starting point, and Aslan’s site:
http://photos.yves.over-blog.com/article-34928981.html
where the default is 25%, but with 30% recommended, and have found values as low as 12.5% for the overlap.

Nick's Tables, e.g.
http://www.pbase.com/digital_initiat...60734/original
have an overlap range of 15% to 50%, which suggests to me that there is room for discussion.

I appreciate that the overlap value is a starting point as the overlap increases as we move away from the “equator” of the images.

The reason for my interest in this topic is that when I started taking panoramas back in the early 1970s I drew on my experience with photogrammetry where we used a 60% overlap when taking aerial photographs with the big 230mm x 230mm format cameras. This gave a 10% overlap between every other photograph and we would place our Ground Control (ground truth) points in this area. This lead to my using approximately a 50% overlap for my panoramas and I used the technique of taking my first photograph to the extreme left of the panorama, noting the subject at the centre and rotating the camera clockwise until this subject was against the left edge of the frame, taking the next photograph and repeating this process until I reached the right hand end, giving an overlap between each photograph of 45% to 50%..

Having this large overlap made the putting together of the panorama a lot easier as I used the same technique as making mosaics from aerial photographs by tearing the prints along a suitable feature to “feather” the edge. Provided that you did not look too closely, this gave great results.

Then I saw a spherical panorama on the Internet and wanted to achieve this myself. At that time surfing the net did not produce a lot of information and about the only thing that seemed to come from this research was the need for a fisheye lens, which I now know not to be true, but I eventually saved my pennies (cents) and purchased one and started experimenting.

I deliberately chose a “difficult” subject on the basis that it is not always possible to revisit the site if things were not right the first time and since then I have become involved with panoramas for introducing colours into HDS (High Definition Surveying) scans where the camera must be centred at the same point as the scanner so once the tripod has been moved on it is impossible to retake any photographs.

The results of my experiments, and previous experience with panoramas, has tended to produce much better results with overlaps of 45% to 50%, so I tend to take 8 horizontal images plus 2 nadir images with a Sigma 8mm or Nikon 10.5mm lens on a camera with an APS-C sensor. I found that using PTGui, 4 horizontal images required a lot of Control Point editing to get a sort of respectable solution and 6 horizontal images gave a better solution, but still needed Control point editing, whereas the 8 horizontal images processed sweetly. I also found that too many images (e.g. 12 horizontal images) created a problem as Control Points then appeared where they should not, and the only way to get a solution was to split the 12 images into 2 sets of 6.

I have also experimented with other lenses, such as the Sigma 10-20mm, and found that I get much better results using the bigger overlaps.

When it comes to the fisheye lenses it is clear that much less lens distortion has to be dealt with by using more images because of the way the no parallax point, nodal point, entrance pupil, centre of perspective, apparent pupil, or whatever you wish to call it, moves towards the front of the lens as the angle of view increases and that if 8 horizontal images are taken with a lens such as the Sigma 8mm or Nikon 10.5mm that this point coincides well with the gold ring around the centre of the overlap.

However, I find that when I look at the suggested number of images for a Spherical Panorama with a particular lens or camera that they are always much less than what I would use. For me, more is less (i.e. more shots in the field mean much less time on the computer), but I am wondering what others think.

Best regards, Hugh.
2. ## Re: Number of Images for a Spherical Panorama

Hugh,

cause of being "beamed to the nowhere" , maintenance of the server to offer us better service, I just answer you today. - Felt like something as loosing the mobile these days? -

I personally tend to take more pictures with a greater overlap cause I do not like to "loose a scene". Cause I do not came back to this place, very special light conditions, the very special moment of the scene. So I always calculate my settings with minimum 30% overlap, re checking the number of pictures I got in the calculator in %overlap.

Trying to follow several forums, the conclusion of discussions is "not too much overlap", like 40 - 50%, "this will cause difficulties during stitching".

As a first comment of my experience: bigger overlap is better to save "the second of a moment".

I will do my next shoots with "my normal settings" and go for another overlap like 40 - 50%, I will be eager to get to know what happens while stitching, more manual CP's, less, more or less errors. I will try to check the difference in time and more "ghosts" in the resulting panorama cause of taking more time to succeed in the shot versa time to spend in the "backstage" on the computer. I will let you see the results.

Thx for sharing your great experience about this theme with the forum.

I adore to have some time free these days to spend my time with questions about these themes.

Regards

Heinz
3. ## Re: Number of Images for a Spherical Panorama

Originally Posted by Hugh
However, I find that when I look at the suggested number of images for a Spherical Panorama with a particular lens or camera that they are always much less than what I would use. For me, more is less (i.e. more shots in the field mean much less time on the computer), but I am wondering what others think.
Four shots around with the Sigma 8mm is certainly enough to get a good stitch. If you use a template with PTGui (containing calibrated lens parameters), then the natural fisheye distortion can be accurately corrected without the necessity of having many control points well spread out. There are other considerations that might make more shots desirable on occasion, but mostly 4 around is fine. 8 would be overkill.

John
4. ## Re: Number of Images for a Spherical Panorama

Thank you Heinz and John for your comments.

It is interesting that with only two comments that the debate is wide open.
I am pleased to see that Heinz has a similar opinion to mine.

Originally Posted by hindenhaag
Trying to follow several forums, the conclusion of discussions is "not too much overlap", like 40 - 50%, "this will cause difficulties during stitching".
I find that with the larger overlap (e.g. 8 shots @ 45° with a Nikon 10.5mm and APS-C sensor) that the stitching in PTGui using "Align images.." gives me a "very good" with tiny residuals (Average control point distance: 1.1), but using only 4 images doesn't even compute because the overlap is so very small.
12 images is too many as well and confuses PTGui.
6 images and I have to manually edit the Control Points and spend a lot of extra time making the panorama.

To me it makes sense to use the "centre" of the images, especially having read Joost's comment on the PTGui forum:
http://groups.google.com/group/ptgui...9a3276a3f8d8e#
Does PTGui pick sharpest pixel within overlap?
Yes, the PTGui blender prefers to use the center parts of the images, it will place the seam as close to the center as possible.
If you increase blend priority, more of the image gets used.
Joost

Originally Posted by John Houghton
use a template with PTGui (containing calibrated lens parameters), then the natural fisheye distortion can be accurately corrected without the necessity of having many control points
I have taken your comments on board John and will experiment with lens calibration (Perhaps I will get a chance to photograph the stars when I visit Nigeria next month), but unfortunately the opportunity for calibrating the lens is not always possible.
Last week I had the privilege of using a Nikon D300 + Nikon 10.5mm, but it was hired for just the day and that was a very full day where we managed to completely fill am 8Gb card because we were bracketing to get images to experiment with HDR.
Also, I sometimes support HDS (High Definition Survey) users and the lack of overlap, especially in conditions where the detail is non-existent or reparative (e.g. pipe-work and perforated ducting) 6 images is not enough and al I receive are the original photographs with no access to the camera and lens.

Taking Spherical Panoramas for HDS has two important considerations:
1) The centre of rotation of the Panorama Head has to coincide with the centre of the Scanner to tenths of a millimetre, which is not a problem using a survey tripod and Tribrach , but does mean that once the tripod and Tribrach have been moved that there is no possibility to retake the photographs;
2) The location of the Scanner is the first consideration. When taking Spherical Panoramas for their own sake we can locate the camera at the perfect spot for the Panorams, but when working with HDS the Panorama has to be taken at the best location for the Scanner, which can be right in the corner of a room (e.g. to "see" through a door to link the scans) or where there is little information for PTGui use for its image matching.
With today's generation of Scanners it is possible to make 20 scans in a few hours so the time for the photography has to be minimal and right first time.

A final consideration is the precision of the Spherical Panorama, which is easily checked when the colours are superimposed on the 3D points.
In inaccurate Spherical Panorama is not detectable to the human eye, but is a problem for an HDS Point Cloud.

Thanks again for your comments, and I would still be interested to receive other users experience or opinions.

Best regards, Hugh.
+ Reply to Thread

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts