In 2008 I was bitten by the panoramic / QTVR photography bug (thanks Mark K.!), so most of my bargain hunting that year was for related gear. I thought I could get away with shooting by hand but frankly, the stitching software ends up having a really rough time and then I have just as rough a time trying to fix the mess, if it is even fixable. So like many other things in life, this problem is one that is easily resolved by throwing some money at it. In this case, the money bought a Manfrotto 303SPH tripod head, designed for shooting spherical panoramas. While I was feeding ailing economies here and abroad I also picked up a Canon 10-22mm EF-S lens, which would save me some time when shooting but without losing quite as much detail as would a fisheye.
The following page shows some examples of my adventures
into interactive panoramic photography:
So, about that pano hardware...
The photograph below of a Manfrotto 303SPH reveals a complex contraption meant to facilitate the capture of multiple images for 360°x180° panoramic photography. What escapes many people is why they can't just stand in one place, turn around and take a bunch of pictures, or why they can't just swivel the camera around on an ordinary tripod.
The problem boils down to parallax, and it is easy to demonstrate. Close one eye, hold out your hand and put up a thumb so that it covers something a little further away. Without moving your arm or thumb, pivot your head back and forth and notice how the obscured object re-appears. The reason for this is that your eye does not pivot directly above your neck. As your head rotates, your point of view moves. When the point of view is moved the picture looks different.
A tripod usually screws into the base of a camera body. In an SLR camera that screw often aligns with the film or sensor plane. But the camera actually sees the world from a point somewhere near the front of the lens barrel, as far as several inches forward of the axis of rotation upon the tripod.
The trouble begins when the panorama stitching software tries to line up all the pictures. It depends on finding and matching up details common in overlapping parts of each picture. It needs to match the right side of a left picture to the left side of a right picture, top of a bottom picture to the bottom of a top picture, etc. If the left picture shows a foreground object appearing to the right of a background object and the right picture shows the same foreground object to the left of the background object, the software will get very confused and it can be very time consuming to fix, if it can be fixed at all. These parallax errors are only a minor irritation for software designed for simple stitching of wide landscapes, but a complete 360° requires matching a lot more detail and those parallax errors can be a complete show-stopper.
The intent of a panoramic tripod head is to make sure that the camera's perspective of the world around it is identical for each photograph. It does this by allowing the camera to pan and tilt on the axis of that point from where the camera actually sees the picture (called the nodal point, entrance pupil or simply "no parallax point"), rather than about the axis at the back of the camera. But buying the head is still just half the task - you still have to figure out the entrance pupil of the camera and/or lens and make the correct adjustments to the tripod head.
Here's the thing about that Manfrotto spherical head. In this picture on the left (the perspective is one that would mount a camera in portrait orientation and face it slightly up to the left) you will see three sliding plates on the 303SPH. The first one on the bottom (1) adjusts the assembly for the height of the camera and lens. Then the second plate mounted to the folding vertical bracket (2) adjusts for the offset of the lens' nodal point from the back of the camera, and serves to rotate the camera up and down. Finally that third plate is what attaches to the camera (3) and is meant to allow side to side adjustment for cameras whose tripod socket is not on the same optical axis as the lens and film or sensor.
How many digital SLR cameras have you seen whose tripod socket is off-axis?
Not mine either. That third plate has to go.
I'm sure that someone must need that third plate but that someone is not me. Even my little Canon SD700 has an on-axis tripod socket.
That Manfrotto plate is about 5-1/2" x 2-1/4" - the most ungainly thing on anything less than a medium format camera, hurts when you bump it, scratches surfaces on which it rests, and is heavy. I found a web site where someone replaced that third Manfrotto plate with an RC2 quick-release clamp and plate, but since I already have some Arca-Swiss type stuff I decided to pick up a Really Right Stuff lever operated quick-release clamp and a matching plate for my Canon 20D. It makes mounting the camera much simpler and quicker, eliminates plate swapping for me, allows the head to fold and store more compactly, and saves at least four ounces too.
Of course nothing is ever quite as simple as it would seem. RRS' B2 LR II is a beautifully machined quick-release clamp but the nicely polished bottom surface meant that no matter how "farmer tight" I make the 3/8" screw, nothing will stop the clamp - and thus the camera - from rotating about the screw and wandering off-axis from the nodal point offset plate.
Nothing we can't fix with a Dremel of course (cue obligatory Tim Allen grunts here). OK, perhaps a Dremel is a bit small for Tim the Tool Man. He'd want something with more power! Suddenly the drill scene in Body Double comes to mind. Uh-oh, let's get back to business...
If I had a drill press here I probably would have tapped a second shallow tripod socket into the RRS clamp. But when all you have is a hammer, all problems look like nails. Since all I have is the Dremel, dozens of Aluminum oxide cut-off discs and some basic hand tools I resorted to cutting a small piece of 1/8" thick 1/2" wide Aluminum stock, to a size that will fit snugly into the slot of the Manfrotto plate. That would be about .21" wide give or take a few thousandths. I then used Power Poxy's PLUG-N PATCH metal epoxy (other products would do fine - read the directions carefully) to bond that bit of Aluminum to the RRS clamp.
Here is what the RRS clamp looks like mounted to the
303SPH in place of that third plate. The bubble level is superfluous in
this application. This arrangement is not quite as plug'n'play as an RC2
clamp and architectural plate perhaps, but it works better with what I already have
in my bag.
The RRS clamps and plates all have very useful centering and index marks so it
only takes a few seconds to get alignment perfect when mounting the camera.
PS, note the white sticker I affixed to the slide face, to help set or re-set
its depth consistently when unfolding the apparatus.
view from the back, arrow
pointing to a .125" x .21" x .5" piece of Aluminum. In order to correctly
locate and align the piece when bonding it to the RRS clamp I mounted the clamp on the
plate, double-checked the clamp's alignment with a square, put a very thin bit of epoxy on
one side of the Aluminum piece, then with needle-nose pliers I pressed the
Aluminum into place through the plate's slot.
A picture of the bottom of the augmented RRS quick release clamp. Around the world, experienced machinists are weeping for my soul as they gnash their teeth over tool marks and the ugly epoxy. But it works fine and is perfectly solid, doesn't hurt anything or interfere with functionality, and is small enough that it should not interfere with normal ball-heads if I ever choose to move the clamp from the 303SPH.
Note well, the machining of those slots in the Manfrotto plates is not done to very close tolerances. I have found slot width
variations of several thousandths of an inch from plate to plate. So if
you plan to reproduce my modification and seek the same snug fit, don't take
my word on that .21" measurement. Measure for yourself and grind
The lens seems to one of Canon's "sleepers"...
Reviews of Canon's 10-22mm F3.5-4.5 EF-S lens are surprisingly positive, the EF-S stigma and medium price point being quietly overshadowed by qualities more typical of Canon's kilobuck L-series lenses such as UD glass and aspherical elements. A USM motor is obligatory of course. The 10-22mm is fairly light as 77mm lenses go - 13.7 oz - and the zoom shows little if any tendency to wander from gravity. My lens happened to come with the "optional" case. Perhaps because I bought it in Korea? No optional hood but the one from Canon's 16-35mm is fine. In fact, even the EW-83H hood from my 24-105mm F4 L usually doesn't cause any vignetting.
I discovered the nodal point of Canon's 10-22mm EF-S lens is fairly far forward. After lots of testing, plate nudging and more testing, that lens' NP appears to be at the crease in the zoom ring just forward of the textured rubber, about 2-3/4" (69mm) from the mounting plane. That was at 10mm - I did not test other focal lengths. And not to get into any religious arguments about nodal point versus entrance pupil versus aperture or F-stops, but I did not measure any practical difference between F3.5, F22 and the lens' alleged "sweet spot", F6.3.
Interestingly, the Entrance Pupil Database says this should be 66mm, not 69mm, and others have reported the no-parallax point to be at the gold ring which would be 73mm, so Your Mileage May Vary. But this should give you a good starting point at least.
With a 97.24° x 74.11° field of view against an APS-C sensor, this lens can capture a complete spherical (360° x 180°) panorama with as few as 12 portrait shots, though I tend to use 24 for more thorough overlap, which becomes 72 if utilizing bracketing for HDR. I have found HDR very useful for panoramic photography mostly for meeting the exposure challenges typical to wide angle photography.
I will probably keep the 16-35mm F2.8 L around for whenever I end up getting a full-frame DSLR or if I venture back into film for some strange reason but for now, this lens has a permanent place in my kit and not just for panoramic work. However a lens this wide is a learning experience, not just in composition but in terms of things like exposure and lighting. Fellow amateurs be warned!
Outside the box...
Nobody says you MUST use a wide angle or fish-eye lens for shooting panoramas. In fact if you are shooting something with lots of small and/or distant detail, the results from, a lens such as a 50mm prime can be very impressive. The difference is that on a consumer DSLR with an APS-C sensor, a 50mm lens would require at least 192 photographs to cover 360° x 180° - 24 at each of 8 different elevations - to capture a complete spherical image. 30 at each elevation if you want better overlap for the stitching software. If you bracket, that's 720 images. Bring spare flash memory cards!
Many people shoot perfectly good panoramas with small point 'n shoot cameras and home-made panoramic heads, too. You don't have to spend thousands of dollars. A camera with good manual control is a must. RAW image output helps a lot too and there are hacks available for some Canon P&S cameras to enable both.
Seems like half the time I shoot, I forget something fairly important. So here is a checklist of things to tend to before we waste time, waste all that money on nifty hardware, and possibly miss a rare opportunity. You may correctly assume that most of these items are things I have neglected to do at least once!
OK, leveling the tripod isn't the end of the world but it gives you less headaches in the software phase. RAW or TIFF recording is important because JPEG compression artifacts can make it more difficult for the stitching software to accurately locate control points, and because the panorama software will usually compress the final image and when you further compress an already compressed image you lose much more detail and color subtleties with little reduction in data size. The manual focus, white balance and exposure necessity should be a no-brainer, as you need consistent exposure from shot to shot (or set to set if bracketing). If you are shooting in close quarters then a very small aperture will help depth of field but keep in mind what opening your lens "likes" in terms of least image distortion, chromatic aberrations, corner falloff etc. And the bracketing and continuous mode is something that is particular to certain cameras such as my Canon 20D which, if AEB and Continuous are both set, will conveniently snap all three exposures in succession as long as the shutter release is held down, and then will stop.
There are those who insist on using mirror lock-up for the best picture clarity. Test your combination of camera, lens and tripod before you bother doing so in production. In my case for example, I found no discernable difference with anything less than a 200mm lens. Others maybe more sensitive.
I have begun carrying around a couple of pieces of blue painter's tape to help make sure the zoom ring on my 10-22mm lens stays put. The adhesive is very benign yet still fairly secure.
I've also taken to nearly always shooting bracketed exposures. Storage is cheap and getting cheaper every day, and the flexibility can be desirable later on, even if only for some artistic license.
Left the white balance on "auto", did ya'? If you shot RAW, no problem. Use your camera's PC-based RAW processing software to uniformly apply the correct WB setting to all the pictures. RAW also permits more latitude for exposure correction if necessary.
You forgot to remove the tripod and take a nadir (straight down) shot? Compose a nice logo for yourself, circular perhaps, and use it to cover up the tripod! Somewhat more time consuming would be to use Photoshop or GIMP to get rid of the tripod, or take the lazy way out and set a limit to the negative tilt when you create the panorama.
I am losing count of how many times I have failed to correctly re-set the upper slide on that Manfrotto head after unfolding it at a shoot. Aggravating if you are still on-site, but gut-wrenching if you don't discover the goof until the stitching software throws fits about the parallax problems.
Try really hard not to bump into the tripod during shooting sequences. Even tiny movements can cause a lot of grief during cubic or spherical stitching.
Watch your tripod equipment setup in general. The pictures below are from the bottoms of shots taken at -50 degrees from horizontal. On the left, the Manfrotto 303SPH's bottom plate locking knob intrudes inward into the composition. Bad move. On the right the camera was rotated the other way (more like the configuration in this page's first picture) so the knob stays behind and out of the way.
The difference in the prominence of the tripod head's features in the finished stitch is alarming. If you have no nadir shot, this means an uglier nadir in the panorama or the need for a larger logo or more work editing your source images before stitching.
You can also see some benefit in keeping the tripod legs angled closely in, and/or extending the tripod's column for required height, but be mindful of stability and vibration issues respectively.
I am omitting nitty-gritty details about the workflow elements since that information is best found elsewhere and varies greatly by need and situation. This is meant to serve mostly as a conceptual example.
How many pictures must we shoot?
You want to be sure the images have a good amount of overlap - at least 30% or so - so the software can successfully stitch the photographs together. All we need is some information and some math. It's all about the camera and lens. As mentioned previously, the Canon EF-S 10-22mm lens has a 97.24° x 74.11° (107.47° diagonal) maximum FOV on an APS-C camera. You can find out these measurements from your lens manufacturer. Remember to compensate for your camera's sensor size if not a "full size" 35mm sensor. Then note whether you are shooting in portrait or landscape orientation. The Manfrotto spherical tripod head forces portrait orientation.
Where 0° = horizontal, +90° = zenith, -90° = nadir, and rounding to the nearest 0.5°...
Vertically, the 97° FOV allows us to shoot the full 180° from nadir to zenith with just two shots, or two rows of shots. But the overlap would only be 14°, or about 14% of the frame, and that's assuming your apparatus allows perfect alignment to shoot the bottom row at precisely +48.5° (to cover -7° to +90°) and -48.5° (to cover +7° to -90°).
In this case we are better off shooting three rows of shots for more overlap, and this way we get one highly useful "straight ahead" row.
In portrait the overall FOV would be -48.5° to +48.5° from horizontal. Axis tilted up 55°, AOV is +6.5° to +103.5°. Likewise, axis down 55°, AOV is -6.5° to -103.5°. So using -55°, 0° and 55° tilts gives us a generous 42°, or 43% overlap of the frame vertically. Overlaps at zenith and nadir are 13.5°.
If shooting a full spherical panorama inside a structure where the stitching oftware could benefit from zenith and nadir image data, we could use 60° or even 65° tilts for 18.5° or even 23.5° overlaps at the zenith and nadir, and still have more than 33% overlaps with the 0° row.
Now to figure out how many shots on the longitudinal axis...
With a 74° horizontal FOV, six shots at 60° longitudinal increments would give only 14° or 19% of frame overlap horizontally. Eight shots at 45° longitudinal increments allow 36% overlap horizontally. Much better!
Eight shots at each of three latitudes = 24 shots. With bracketing for HDR, that would be 72 or 120 shots depending on how many bracket exposures you elect to capture.
I always take photographs in RAW mode and for stitching, I process them into TIFF format using Canon's Digital Photo Pro software. Other camera manufacturers have similar software for working with their cameras' raw image files, i.e. Nikon Capture. Some panorama software can read RAW photos directly but software like DPP makes it easy to apply color and exposure corrections uniformly to all the photographs, then batch process the raw files into other formats such as TIFF, JPEG, etc. RAW interpretation tends to vary from product to product so even if your panorama software can read RAW files, TIFF is the best way to achieve consistent results. I place the resulting TIFF files into an appropriately named folder for the project, i.e. "Grand Canyon Panorama".
Tip: I have found that depending on the scene being photographed, the stitching software sometimes works better with 8-bit TIFFs than 16-bit. If you are are having a rough time with one, try the other.
Since TIFF files support layers and masks, they are also an ideal format for editing to correct or cover the tripod fixtures at the nadir, and to rectify image anomalies in the resulting panorama caused by movements of people and vehicles between the time that adjacent pictures were captured. You cannot do this well, if at all, with JPEG files. Look up how to add alpha masks to TIFF files in Photoshop or Gimp, and how to use them to mask off areas that you want the stitching software to omit.
Remember that if you shoot multiple exposures for HDR imaging, you need to duplicate the alpha masks to each of the bracketed exposures, or add an alpha mask to the stitched image before the blending steps and rendering it to its final projection and/or presentation form.
The TIFF images must be fed into software that can stitch the photographs together into a single image. Some photographers are happy with "flat" static panoramic images, but I always found that a bit weird. The software I am using can do flat images but also renders the stitched image into a virtual reality product that can allow you to pan the image around as if you were standing in the middle of it. This solution revolves primarily around Quicktime VR, which uses Apple's Quicktime Player for playback.
I compile my Quicktime VR (.mov) files using PTGUI. PTGUI can take some fiddling to make it sing but with the exception of going full-on hard-core and using the free Panorama Tools, it's also one of the least expensive packages, weakness of the US dollar to the Euro notwithstanding. PTGUI can stitch photographs together into various flat images and produce QTVR files. It runs best on a very fast PC or on an Intel -based Mac with plenty of free space on fast disk drives.
Since not everyone has the Quicktime Player software loaded, I utilize Flash Panorama Player to integrate the QTVR files into web pages which can be viewed by nearly any web browser that supports the Adobe (previously Macromedia) Flash plug-in. FPP still has a few major shortcomings. The VR components of Flash are not supported on any of today's mobile devices so your panoramas cannot be viewed on smart phones, iPads, etc. And FPP takes some skill to exploit its potentials and it has some issues if you're concerned about people stealing your work. A product called Flashificator is supposed to help take the grunt-work out of making FPP sing, however I have not yet tried it. There are also many additional FPP plug-in enhancements for many purposes - most of them free - available from many people. Flashificator attempts to support most 3rd-party plug-ins. I find it curious that Flashificator is nearly 40% more expensive than FPP, the product it enhances!
The most recent versions of PTGUI have been getting dramatically better at stitching up tricky sets of photos, 16-bit images, etc. The software still lacks an easy way of patching up a nadir or zenith but does have tools for masking off inconsistent parts of adjacent images without having to feed all your work through Photoshop.
Immervision Pure Tools is highly recommended and can give you output using a Java wrapper (vs QTVR and a Flash wrapper). Java should be even more platform -independent than Flash, but still seems subject to limitations of the "Flash Lite" built into most PDAs and smart phones. Pure Tools produces proprietary format ".IVP" files which require Immervision's software to display. Pure Tools still requires a stitcher and all the components of Immervision's software are sold separately. One unique advantage is that the .IVP files can be played back on Windows Mobile devices. I have not personally tried Pure Tools.
Panoweaver was also recommended but it is very costly and its copy protection method is utterly vile and entirely unacceptable. My advice: Do not trust software that does not trust you. I refuse to even evaluate this software.
I did evaluate Autodesk's (formerly Realviz') Stitcher Unlimited. I consider its copy protection worrisome as well. Stitcher Unlimited features ease-of-use that is unexpected from a company that built its reputation on AutoCAD - a software package that takes most people years to learn. On the other hand that ease of use limits its flexibility, and I was not satisfied with the image quality of the QTVR output.