An Investigation of Different Lenses to create 360° Panoramas
The conventional wisdom is to use fisheye lenses to create 360° Panoramas, but these are expensive items and the cost can discourage people from experimenting with this exciting form of photography. I was asked if a DSLR with a "kit" lens could be used to create 360° Panoramas and if the resulting Cube Images were OK for import into the Cyclone software for use with HDS (High Definition Surveying) so I decided to investigate the matter and publish the results as the information could be useful to those wishing to create 360° Panoramas, whether or not they are for use with HDS.
For the investigation I used a Nikon D60, which has an APS-C sensor, with the following lenses: Sigma 8mm and Nikon 10.5mm fisheye and Nikon 18-200mm VR set at 18mm. I also used a Panasonic FX100 compact camera at the wide angle setting.
The photographs were taken with the cameras mounted on a Panosaurus panoramic head. The Panosaurus is a friction head so the "pointing" is by lining up a mark with the built in protractor manually. A Nodal Ninja or Manfrotto panoramic head with click stops would have provided more accurate "pointing", but as is evident from the results, this was not a problem.
The images were processed using PTGui to create the Equirectangular images and Pano2VR used to produce the Cube Images for import into Leica Cyclone. The Cube Images were 4096 x 4096 to provide maximum resolution and orientated so that the direction was the same for each set of images.
The Scan World was created using a Leica HDS6000. All the points (100%) were imported into Cyclone to provide the maximum resolution, which together with the 4096 x 4096 Cube Images provided the best possibility to detect any discrepancies. The Cube Images were "fitted" in Cyclone using the minimum three points, with one on each of the Front, Left and Back images. The same points were used in each set and the points chosen were very clear in both the Cube Images and the Scan World.
The subject was a room, which was chosen because it contained a substantial concrete pillar ensuring that the HDS scan and 360° Panoramas would be coincident, even if more photographs were taken sometime later. The room proved to be a good subject because there were large areas with little detail on the walls and the tiled ceiling had a "regular pattern". Symmetry in the subject can cause difficulties. An example is the Marble Saloon at Stowe House in Buckinghamshire, which confused the image matching algorithms in PTGui to add Control Points to images on the opposite side of the room. This was solved by running Align Images... in PTGui, then editing out all the unwanted Control Points before adding new ones, which did take some time.
The times for processing the images through PTGui and Pano2VR will depend on the specification of the computer used and the dedication of the computer to the task. If other programs are being used or running in the background, or the computer is on-line, then the processing will take longer.
To give an idea of the difference in time for more images the following times are for processing the 12 images using the Nikon 10.5mm lens compared with 56 images using the Nikon 18-200mm lens @ 18mm:
The operations to Open Images and Align Images are pro rata with the number of images, whereas the creation of the Panorama from the 56 images only took twice as long as the creation of the Panorama from the 12 images.
These times are for the automatic processes without having to manually add Control Points, which can be very time consuming, especially if there are a large number of images.
The Sigma 8mm lens is available to fit a variety of SLR and DSLR cameras including the Canon and Nikon range, so is often recommended. This lens has an Angle of View of 180° on the long side of the image when used with an APS-C sensor. As mentioned in Making use of the Nodal Point, 6 images @ 60° intervals will cover the subject, but 8 images @ 45° intervals gives more overlap between the images, so this was used in the investigation.
Three arrangements were used:
8 Horizontal Images @ 45° intervals: No extra Control Points were required, but there is a "hole" and some mismatching at the zenith.
8 Horizontal Images @ 45° intervals + two images @ 90° @ a 90° Interval: Additional Control Points were required for the two zenith images. This was probably due to the "symmetrical" nature of the ceiling tiles and lights as this arrangement does not usually need additional Control Points.
8 Images @ +10° @ 45° intervals: Tilting the camera up by 10° eliminates the "hole" at the zenith and does create a slightly larger "hole" at the nadir, but no additional Control Points were required.
The Nikon 10.5mm lens has an Angle of View of 180° across the diagonal of the image when used with an APS-C sensor, so covers a smaller area than the Sigma 8mm.
Two arrangements were used:
8 Horizontal Images @ 45° intervals: No additional Control Points were required, but there is a larger "hole" at the zenith and nadir than with the Sigma 8mm.
8 Images @ -10° @ 45° intervals + 4 Images @ +50° @ 90° intervals: No additional Control Points were required. This arrangement gave the best result.
The Nikon 18-200mm zoom lens with an APS-C sensor is equivalent to a 27.4mm in 35mm photography terms.
Three arrangements were used:
16 Horizontal Images @ 22½° intervals: Additional Control Points were required in areas where the small coverage by the lens meant that there were very few common points in adjacent images. This arrangement is often all that is required for some HDS applications and having less images means less processing times.
16 Horizontal Images @ 22½° intervals + 16 @ +35° @ 22½° intervals: Additional Control Points were required in areas where the small coverage by the lens meant that there were very few common points in adjacent images, but having the two tiers of images helped. This arrangement could be more useful when scanning where a single tier of horizontal images does not fully cover the height of the subject.
4 tiers of 16 Images @ 22½° intervals with the tiers @ -35°, 0°, +35° and +70°: Additional Control Points were required in areas where the small coverage by the lens meant that there were very few common points in adjacent images and in the images @ +70° because of the symmetry of the ceiling tiles and lights. It is often possible to only have 8 images @ 45° intervals for the +70° tier, but this did not work on this subject.
The Panasonic Lumix cameras advertise that they have wider angle zoom lenses than most compacts. The 6mm lens with the small sensor has 4.796 magnification factor so is equivalent to a 28.8mm lens in 35mm photography terms, which is slightly longer than the Nikon 18-200mm lens at 18mm. The results using the Panasonic are almost identical to those using the Nikon 18-200mm lens with the exception that with the 16 Horizontal Images @ 22½° intervals + 16 @ +35° @ 22½° intervals, no additional Control Points were needed.
The Panasonic lacks the manual control that is possible with a DSLR so the flash was used to take the images. The camera's ¼” Whitworth mount is offset so this was taken into consideration when mounting it on the Panosaurus. I did not calibrate the Nodal Point as I did not want to bombard the sensor with a laser, but estimated it from the geometry of the Nikon 18-200mm lens.
The use of this camera was an afterthought and was possible because of the very stable concrete pillar in the room used as the subject.
Three arrangements were used:
16 Horizontal Images @ 22½° intervals: Additional Control Points were required in areas where the small coverage by the lens meant that there were very few common points in adjacent images.
16 Horizontal Images @ 22½° intervals + 16 @ +35° @ 22½° intervals: No additional Control Points were required. This was probably due to a different plan orientation of the images so that there was sufficient detail in the overlaps to provide enough Control Points automatically, and having two tiers of images with good overlap helped.
4 tiers of 16 Images @ 22½° intervals with the tiers @ -35°, 0°, +35° and +70°: Additional Control Points were required in areas where the small coverage by the lens meant that there were very few common points in adjacent images and in the images @ +70° because of the symmetry of the ceiling tiles and lights.
The Cube Images were "fitted" in Cyclone using the minimum three points required, with one on each of the Front, Left and Back images. The same points were used in each set and the points chosen were very clear in both the Cube Images and the Scan World.
In the Right Cube Image is a green radio standing in front of the grey wall making it ideal for checking that the relevant points in the Point Cloud have been coloured correctly.
All the options fitted the Point Cloud exactly showing that whichever arrangement was used for taking the images provided the same result. It has to be said that using the the Nikon 18-200mm lens and Panasonic camera did require a lot more work and time when additional Control Points were needed, but these sets of images did allow higher resolution Equirectangular Images to be created.
An Equirectangular image has the property that the height is exactly half the width. When using arrangements such as the 16 Horizontal Images @ 22½° intervals or 16 Horizontal Images @ 22½° intervals + 16 @ +35° @ 22½° intervals, PTGui may need to be manually set to achieve this as creating Cube Images from an Equirectangular image with an incorrect aspect ratio will result in distorted Cube Images. When working with such arrangements click the Advanced >> option and then the Panorama Settings tab and ensure that the Projection: option is set to Equirectangular (for spherical panoramas) and set the Field of View: (vertical) value to 180 before opening the Project Assistant tab again and using Create Panorama...
Creating 360° Panoramas from large set of images, such as with the Nikon 18-200mm lens, is more time consuming and requires more effort than using a fisheye lens, especially as it is probable that additional Control Points will required, which needs manual intervention, but does enable higher resolution Equirectangular Images.
One of the concerns with using a lot of images was that the resulting Equirectangular Image may not represent a true sphere in the same way that the fewer legs in a conventional Angle and Distance Survey Traverse the better the "closure". This investigation found that this concern was unfounded and that all the Cube Images produced by the different arrangements fitted the Point Cloud in the HDS Scan World exactly.
The following images show the different amounts of coverage that the different arrangements give in the Equirectangular images.