1

Is there a way to take long horizontal images using techniques similar to a scanner NEARLY INSTANTLY

In other words I want to be able to point a camera out the left or right window of my car, drive a few blocks in one direction, and 30 seconds later post a picture that is an orthographic projection of the view out the side of the car for that 30 seconds. The same way I can rotate the phone and a few seconds later post a panorama. Anything slower is not answer to this question (manually taking 500 to 1000 pictures and loading them into some software over hours is not a solution).

Such an solution would either take video or use a live camera and using motion detection pull out a center column of each frame to build an indefinitely long image.

The difference between this and panorama is panorama works by rotating the camera. This works by moving the camera in one direction so imagine putting the camera facing out the side of a car or train and making a long flat projection photo of an entire city block.

I know I could do this in Photoshop potentially but it would be extremely tedious as only a small column from center of each image can be used otherwise there would be perspective distortion issues so I'm wondering if there are other solutions.

Trying to make it clearer the result of the technique I'm looking for would generate an image with no horizontal perspective. Imagine you just try to take a picture of a bookshelf using 2 pictures, one for the left 60%, another for the right 60%. They overlap in the middle. In both images you'll get perspective. You'll see the sides of the bookshelves perspecting toward the center and you see that panorama software can't fix this problem.

On the other hand if you took 1000 images and only took the center 1% from each one you'd get a flat image, at least in the vertical plane. Because stitching that many (or far far more) images is way too much work, ideally you want a solution that using a live camera only takes 1 vertical pixel column from each image, saving on the storage and processing costs.

This is how smart phone cameras take panoramas (at least iphone). It only takes thin columns as you rotate the camera. Unfortunately it uses the compass sensor to know that you're rotating. There is no position sensor accurate enough to know you're sliding the camera left to right so instead you'd need to use image motion detection tech.

gman
  • 528
  • 2
  • 14
  • This is still a panorama — a panorama is any particularly wide-angle view of a scene. The method of making one by rotating a camera is just one approach. You describe another. – mattdm Jun 21 '19 at 02:56
  • 2
    Closely related: How can I stitch a panorama correctly if I moved the camera along the horizontal axis? – scottbb Jun 21 '19 at 04:16
  • Also related (not intuitively, but still a narrow-slice-panorama): How is this changing vertical perspective effect achieved? – scottbb Jun 21 '19 at 04:19
  • 1
    Possible duplicate of How can I stitch a panorama correctly if I moved the camera along the horizontal axis? – mattdm Jun 21 '19 at 04:58
  • Also, you're wrong about how smartphone panoramas work. If they did that, they would be far too sensitive to tiny misalignments. – mattdm Jun 21 '19 at 05:01
  • They compare one column to the next and find the best match of alignment. Shake as you take the panorama as you'll see the columns are extremely thin. – gman Jun 21 '19 at 05:07
  • Have you ever tried taking a photo of the roadside from a moving car with a phone? They make lovely diagonal patterns, as the scan takes a noticeable distance to complete. You'd need a good fast shutter speed to do it anyway. – Tetsujin Jun 21 '19 at 09:17
  • I'm reminded of this question - https://photo.stackexchange.com/questions/108999/how-can-i-create-a-distortion-free-high-resolution-panoramic-image-of-a-three-di – Tetsujin Jun 21 '19 at 10:54
  • @Tetsujin, thanks for the reference. Yea, that's basically what I'm looking for, just wishing someone brought the tech to 2019 so it was fast like taking a panorama is on a phone today vs what it was like in 2003. Some phones do 240fps video so maybe the could do this if someone tried. Was hoping someone already had. – gman Jun 21 '19 at 13:09
  • So, this is basically an "is there an app for that" question, which is off-topic here. This site is really better for learning how to actually do something rather than doing product searches for you. You might try [softwarerecs.se] instead. – mattdm Jun 22 '19 at 17:11
  • No, it's not an "is there an app for that question" as there could be cameras made for this purpose. In fact there is one at the museum of the future in Tokyo but of course that's not a consumer camera – gman Jun 23 '19 at 01:20
  • Sorry, that wasn't clear. Whether a physical app or a consumer product, shopping questions are generally off-topic. See this discussion on meta. Questions about how to do something have a much longer shelf-life and are more broadly applicable. – mattdm Jun 23 '19 at 01:47
  • Fine, worded to make on topic. – gman Jun 23 '19 at 01:58
  • 1
    This is still an 'is there a tool I can buy to do this' question. Clearly there is a technique for doing this (take an image: if it's the first image keep it, otherwise match it with the previous image and use suitable chunks of it to extend the previous image): the question boils down to 'has anyone burnt this algorithm into a camera or into an app for a phone', and that's still not on topic here. –  Jun 23 '19 at 21:23
  • There are plenty of questions of the form “how do I do X?”. My question is “Who do I do X” where “X” = “take a long orthographic panorama in a short amount of time, without minutes or hours of post processing”. That difference is important as being able to make them fast will lead to making 100s of them instead of < 10 of them as well as opening them up to more people. – gman Jun 24 '19 at 02:36

2 Answers2

1

This is how smart phone cameras take panoramas (at least iphone). It only takes thin columns as you rotate the camera. Unfortunately it uses the compass sensor to know that you're rotating. There is no position sensor accurate enough to know you're sliding the camera left to right so instead you'd need to use image motion detection tech.

You may be mistaken. My smartphone (Xiaomi Note 7) doesn't need a rotation since I just took a linear panorama of kitchen appliances by dragging it along the edge of my kitchen counter. How it works is anyone's guess, it possibly looks at some features of the image and how they move across the frame. Of course when you are standing, the only way to have features of the image change sufficiently fast is to rotate the camera, but if you are close you can just walk as I did.

Looking at the result show the limits of the method, the camera cannot tell the instant speed, so unless you maintain a very steady speed the image is likely to be shrunk or stretched at random horizontally, so I doubt you can use the technique in a car in traffic.

Experiment: "scanning" a magazine cover using the panorama mode (the original picture is 14000 pixels high):

enter image description here

xenoid
  • 21,297
  • 1
  • 28
  • 62
-1

As noted in a comment, this is still a panorama. I'm going to leave arguing that point out of my answer in the assumption that the question will eventually be edited to fix that.

While I'm not aware of any phone apps, the Hugin program commonly used for panoramas and other image stitching on Linux can do this easily. One key term you might search for is "linear panorama" — here's a tutorial I found by searching for that.

To eliminate the (horizontal) perspective distortion you are concerned with, do some combination of these three things:

  1. Shoot with a narrow (long focal length) lens.
  2. Only move a small amount between shots so there is a lot of overlap.
  3. Pre-crop your images to narrow vertical strips.

I haven't done this at particulary large scale — point 3 may get you easier results, although given enough time and compute power, and the right settings, not doing that and trusting point 2 will be sufficient.

mattdm
  • 143,140
  • 52
  • 417
  • 741
  • This is not what I'm looking for. The problem with panaroma software is it takes too large an area from each image. To create what I'm talking about requires effectively a scanner taking vertical pixel from each image the same way most phones take panorama. They don't stitch 10 x 1000x1000 images into 10000x1000 image. They stitch 10000 1x1000 images. Otherwise there will be perspective artifacts because the only part of each image that is correct is the center line (no perspective). – gman Jun 21 '19 at 03:55
  • Depends how far you move the camera between frames and how much overlap there is. – mattdm Jun 21 '19 at 04:48
  • It sounds like you're talking about a slow and deliberate complicated solution like panoramas were in 2003. I'm talking about a fast easy instant solution like panoramas in 2019 on a smartphone. Point the camera out the window, drive for 30 seconds in one direction, 5 seconds later post image. – gman Jun 21 '19 at 05:00
  • @gman The concept is the same. You could do it with frames from a video. – mattdm Jun 21 '19 at 05:02
  • 2
    Downvote or not, the concept is still the same. Sorry you don't like to hear that. – mattdm Jun 21 '19 at 05:21