How do I turn 24 photos into a 360º panorama?

Question

I have gone around a house and very quickly taken pictures of every room in 24 different directions (~45° turn each click), and I've been very amateur about it. I've eyeballed the angles and had automatic on most of the time (and I have accidentally kicked the tripod on many occasion). I know the numbers seem odd, but what I did was that I pointed the camera as straight as possible at a wall with features on it, then tilted it up until the top-most feature ended up at the bottom of the view, then took 8 pictures in a circle, tilted it back down to being straight, took another 8 pictures in a circle, and then tilted it down so that the bottom-most feature ended up at the top of the view and took the last 8 pictures in a circle. 45° x 8 = 360°, 8 x 3 = 24 pictures.

Now I wish to turn these 24 pictures into a 360° panorama picture so that one can look around the room on the web (Myst 4 point-n-click style). Many years ago I had access to Photoshop at a work place and one of the features was to turn a set of photos into a panorama and the results were pretty decent. However, I can't afford an Adobe Suite like that, and have no idea if it will play nice with my Linux system (native or wine).

So the question becomes if there is any software out there (preferably free and works in Linux/Wine) that can work its magic on a folder of pictures and turn them into a nice 360° panorama? Or any php or JS library that can do the magic "on the fly"?

Answer 1

Hugin would be the open source software package of choice for this... if you'd shot the images correctly.

See also:

• How are virtual tour photos taken?
• What are the best techniques to take 360° panoramas?

The chances of your producing a usable equirectangular 360°x180° panorama out of the images you've shot, however is incredibly low, from your description of how you did it. But you might want to try, anyway.

Indoors (in smaller spaces), parallax becomes an issue that can keep simple image warping from being able to cleanly stitch together images without showing seams. To do this properly, you'd have found the no-parallax point of your lens, and used a two-arm panorama head to rotate around that point in yaw and in pitch to cover the sphere.

You'd also have figured out the FoV of your lens and shot sufficient coverage with sufficient overlaps. If you used a fisheye lens you might have had sufficient scene coverage and overlap to stitch.

But with, say, an 18-55 kit lens set to 18mm, to cover the sphere with 25% overlap (and you'd probably prefer something closer to 30% for safety and ghost/clone removal), the shooting pattern you'd need to cover the sphere with the camera in portrait mode (by this calculator) would be:

Row	Elevation	Positions
1	72º	4
2	36º	9
3	0	11
4	-36º	9
5	-72º	4

So, 37 images, in five rows. With precise rotation and tilt in pitch around the no-parallax point. (I.e., to cover the whole sphere, you have to have gotten the ceiling above you and the floor beneath you in the shots, too).

When I shoot this type of panorama, I use a tripod, a panohead set for the NPP of my lens, and a fisheye lens, and I can cover the entire sphere in 10 shots: 6 shots with the camera in portrait orientation, rotated every 60º + a zenith (straight up) and nadir (straight down) shot.

But typically for tripod removal in post, I'll need three nadir shots: two on the tripod rotated 180º from each other, and a handheld shot with the tripod removed.

Shooting 360ºx180º panos is a relatively involved and meticulous process that requires quite a bit of practice and post-processing study and learning and practice, which is why 360º cameras are now a thing. :D

Answer 2

I confirm the use of Hugin for merging photos, and I suspect the previous dubt about the result. In fact the calculation were correct and a true precise panorama need to be calculated "a priori". So, try Hugin for the result.

Answer 3

Maybe https://github.com/alicevision/meshroom ? I used it to get a 3d model of objects, but if there are enough overlap to pinpoint the camera position, it would probably work.