Not sure what got me excited about creating 'immersive photos', but recently I have been hip deep in it. There is something compelling about being
able to look around the photo and understand the relationship between elements. I’m sure everyone has taken a photo with the
standard 50mm lens on their point-and-shoot and thought, “that’s a good photo,
but it totally didn’t capture the environment.” Immersive photos put the viewer in the
center and let them interact with the photo to look in any direction. Like anything that looks hard to do, it is,
unless you have the right equipment.
Click the photo to view in 360 degrees.
The basic process of creating immersive photos is to take
multiple digital photos that cover a 360 degree field of view and then stitch
them together into a single large photo. That photo is then converted into QuickTime
VR or another competing technology that applies real-time perspective to the
photo through a visible portal.
While these basic steps seem pretty achievable, having the
correct equipment is required if you want to complete a photo in under a
lifetime. One thing you will notice
immediately when you enter this world is most web sites and software companies
are not located in the US. Based on my investigation this appears to be
because a small US
public company IPIX has been very aggressive about enforcing a patent they hold
with regard to ‘de-fishing’ fisheye images with focal lengths of less then
10mm. I’m not sure who is right or wrong
in this argument but I will save that debate for another upcoming post I plan
to write on why patents do not serve the greater good of mankind ( little heavy
for a photo blog).
Like most things in the computer world, everything starts
with software. I ended up selecting
Panoweaver. Panoweaver performs many
steps in one pretty easy to use piece of software. I downloaded the demo version of the software
to give it a try. Word of advice, don’t
start buying things until you are comfortable with the entire process and tool
chain. This is one of those ‘happy path’
kind of processes that require all things to work together to end up with
something that doesn’t look like a funhouse mirror.
I have a Canon 20D Digital SLR camera. Digital SLR’s typically don’t have a full
size imaging frame like film SLR’s do. For this reason they create what’s called a drum type image when used
with a wide angle fisheye lens (basically the top and bottom of the image
circle are cut off). So this results in
nearly 180 degree view in one axis and 140 degree field of view in the
other.
So, in order to cover a full 360 degree field of view (with
overlap) you need four images shot with the camera in a vertical orientation
each 90 degree to each other, plus an optional photo for the top and bottom. This creates the following set of images.
The next step is to load these images into Panoweaver. Panoweaver has ‘de-fish’ processing built in (not
all software does) so as it loads it converts the highly distorted images into
rectilinear images. This process
utilizes the meta-data stored in the photos to determine the lens focal length
used during image capture.
Once loaded the software presents adjacent photos
side-by-side and asks the user to identify matching points on the two photos. This process is repeated for all adjacent
photos. I typically identify between 10
and 20 points per photo pair, making sure to cover the entire field of
intersection.
By selecting these points it helps the software morph each
image slightly to make them align better. The software is pretty good with pure horizontal and vertical alignment errors,
but can’t deal well with angular errors along the lens axis at all.
On the hardware side you will need a good rotational head and
a quality tripod. Jinky versions of
either lead to visible errors that the software cannot correct. The trick to shooting these photos is to make
sure the lens’s focal node is exactly at the point of rotation. The focal node is where the light crosses
inside the lens. The farther you get
from this being exact the more visible errors exist in the final experience.
I use the Sigma 8mm fisheye lens and unfortunately it does
not have a mark on it for the focal node. I have been guessing the focal node is about halfway down the lens
barrel, but it’s clearly not ‘exactly there’ as I still have very small errors
in the end result.
After stitching is complete, you can save the resulting
image in three different formats,.
The
first two, spherical and cubic are flat images that can be further enhanced in
Photoshop or whatever tool you use. This
is helpful for adjusting white balance, etc.
The final format is QuickTime VR
Some final important points. When shooting the actual photos you generally should not use auto focus
as that changes the focal node of the lens. The best approach is to try to achieve the largest possible depth of
field, alleviating the focus problem. With regards to lens aperture, you guessed
it, that must be manual as well. Any
differences in exposure will be visible in the finished product. On the same theme, you also need a camera where
you can preset the white balance. The
cameras auto white balance setting will change for each photo taken which will
become amazingly obvious in the end product.
Although it seems like a lot of work, once you get the right
piece-parts and understand the issues at hand you can generally shoot the
photos in about 5 minutes, stitch in 20 minutes and fix any fixable problems in
about 20 minutes. All in all about an hour per 360 degree photo is where it
comes out.
To keep up to date with new web discoveries I make along the way check my http://del.icio.us/toddvernon/qtvr page or simply subscribe to its RSS feed..
if you had 4 or so cameras and a staggering amount of storage you could make a 360 degree movie, 24 frames per second or whatever (even low frame rates could be cool). Now you get to choose your point of view. You can watch the same movie over and over and have different experiences each time. Share views with friends (after the car chase, look behind the red building...)
If someone makes 360 degree movie codec (wonder if someone has already for VR?) then video games could use it for immersion.
If it goes personal, vacation movies would be totally different. Instead of "capture the memories," when you get home you can see things you missed while there since you only can look in one direction.
Posted by: rando | 2005.09.25 at 11:38 AM