I have a video taken at an angle to the axis of a circular body. Since it was taken from an unknown angle, the circle appears as a ellipse.
How to find the angle of camera offset from the video? Also, Is it correct to apply the same transformation to all the frames in the video; as the video camera was in a fixed location?
For a super easy fix, go back to the scene and take the video again. This time, make sure the circle look like a circle.
That being said, this is an interesting topic in the academia. I believe there's various solutions/articles that are aimed to solve this kind of problem. Base on your reputation, I believe you already know that, but still wanted to give Stackoverflow members a shot at answering this problem. So here it goes.
For an easy fix, you can start with this function, by guessing the camera location by trial and error until you find an acceptable transformation to your image (a frame of the video). The function does not work right out of the box, you have to debug it a little bit.
If you have access to the (virtual) scene of the image, you can take an image. Base on mutual feature points from the new image and the original image, register the two images (and get the transformation) (ex1, ex2).
Finally, apply the same transformation to each frame of the video.
To answer your second question, though the camera location is fixed, there may be objects moving in the scene. So applying the same transformation to every frame will only correct the objects that are still. So it's not ideal. In the end, it depends on what the aims of the project is and how this non/correction affects the project aims.
Related
I use the Leaflet plug-in "Leaflet.ImageOverlay.Rotated.js" to use its L.imageOverlay.rotated(...) thing in order to overlay certain map pieces in various places on top of the normal map.
It does this by taking an image and having me tell it its top-left, top-right and bottom-left coordinates to figure out how to rotate, tilt and stretch/squeeze it properly.
It took me a very long time to figure these coordinates out by hand. For this reason, I'm looking for some sort of "geopositioning mode", perhaps enabled by this extension, which would simply let me click three times on the map to tell it where these points go. That would be so simple for the developers to do and would help so much. It's such an obvious thing to do that I strongly suspect it's already implemented and ready.
Is there such a "mode"? If not, how am I expected to find the positions without spending so much time and trial-and-error as I did for the first overlay map image?
Added: I should also clarify that the image should be shown in this mode so that you can re-adjust the points and watch in real time as the image bends/warps, to get it just right.
you can develop a modul for this problem.
find minimum 4 point on raster map.
click on tilemap for 4 points
than find different slope and distance same 2 points.
maybe you must rotate and use affine transformation.
If I have understand well, 3D 360 photos are created from a panorama photo, so I guess it should be possible to create a 3D photo (non 360) from a normal photo. But how? I did not find anything in Google! Any idea of what should I search??
So far, if nothing available (I don't think so), I'll try to duplicate the same photo in each eye. One of the pictures a little bit moved to the right, and the other one moved a little bit to the left. But I think the distortion algorithm is much more complicated.
Note: I'm also receiving answers here: https://plus.google.com/u/0/115463690952639951338/posts/4KdqFcqUTT9
I am in no way certain of this, but my intuition on how 3D 360 images are created in GoogleVR is this:
As you take a panorama image, it actually takes a series of images. As you turn the phone around, the perspective changes slightly with each image, not only by angle, but also offset (except in the unlikely event you spin the phone around its own axis). When it stitches together the final image, it creates one image for each eye, picking suitable images from the series so that it creates a 3D effect when viewed together. The same "area" of the image for each eye comes from a different source image.
You can't do anything similar with a single image. It's the multitude of images produced, each with a different perspective coming from the turning of the phone, that enables the algorithm to create a 3D image.
2D lacks a dimension hence cannot be converted to 3D just like that, but there are clever ways for example Google Pixel even though doesn't have 2 camera can make it seem like the image is 3D by applying some Machine learning algorithm that create the effect of perspective and depth by selective blurring.
3d photos can't be taken by normal but you can take 360 photos with normal camera ..... There are many apps via which you can do this ..... Also there are many algorithms to do it programmatically
for a project I am working on I will need to automatically decide if a video is a VR (360) video, and if so what format it is. Is there any way to tell? I was thinking the metadata but could not find any information on this.
Checking size is pointless, most properly encoded movies are using standard sizes like 1080p (1920x1080), WQHD (2560×1440) or 4K (3840×2160) because of better hardware decoding. To do that they don't have square pixels. Therefore you shouldn't guess anything by ratio.
What you should do, is check the presence of zenith and nadir. That is check for the topmost and bottommost region of image if it is the same color (assuming the most standard equirectangular projection).
This approach will need some adjusting if you have stereoscopy involved. You would have to repeat this procedure for each eye region. As a bonus, you can also deduce some stereoscopy types - for example you could differentiate top-bottom, mono and left-right. Unfortunately you couldn't guess which image is for which eye, so you would have to assume the more common situation where the left eye is usually the top or left one in the image.
There is an RFC for Metadata to describe spherical data in MP4 videos:
https://github.com/google/spatial-media/blob/master/docs/spherical-video-v2-rfc.md
This includes the introduction of a new spherical video header box, svhd, which you can test for the presence of to detect if a video is a VR 360 video.
This is not ubiquitous yet but it does have support from key players like Google, and as you have discovered, something like this is necessary so it seems likely its use will spread.
I am trying to do a number of things via MATLAB but I am getting a bit lost with what techniques to use. My ultimate goal is to extract various measurements from a users fingerprint presentation, e.g. how far the finger over/undershoots, the co-ordinates of where the finger enters, the angle of the finger.
In my current setup, I have a web camera recording footage of a top down view of the presentation which I then take the video file and break down into individual frames. https://www.dropbox.com/s/zhvo1vs2615wr29/004.bmp?dl=0
What I am trying to work on at the moment is using ROI based image processing to create a binary mask around the edges of the scanner. I'm using the imbw function to get a binarised image and getting this as a result. https://www.dropbox.com/s/1re7a3hl90pggyl/mASK.bmp?dl=0
What I could use is some guidance on where to go from here. I want to be able to take measurements from the defined ROI to work out various metrics e.g. how far a certain point is from the ROI so I must have some sort of border for the scanner edges. From my experience in image processing so far, this has been hard to clearly define. I would like to get a clearer image where the finger is outlined and defined and the background (i.e. the scanner light/blocks) are removed.
Any help would be appreciated.
Thanks
I am working on a iOS 5 iPhone project where users can choose an image on their device, then trace an object inside of the picture (trace an apple out of a picture of a fruit basket), and then the picture needs to be uploaded with the "tagged" object so it can be pulled down later. Other people will then pull down the image and try to find where the tagged object is in the picture (Think "Where's Waldo?").
I have been trying to figure out the best way of tracing the object. Before, I had a user press the top left, top right, bottom left, and bottom right points around the object and create a square view around the object. The info for that view was uploaded and then pulled down for the user to find the object. The downside is that all objects are obviously not squares/rectangles so I need to do a free form shape.
I was thinking of allowing the user to draw over the object and then somehow I need to be able to tell what is inside of the trace (For example, inside of a circle that they traced), but a problem I forsee is making sure the trace they made is closed so I can fill in the shape (which is a whole not problem).
Any advice welcome on the best way of starting this.
Thanks!
UIBezierPath might be a very useful friend here. It allows you to create any shape you need, and it supports both drawing and hit-detection. I recently did an implementation for a storybook where I could trace out a shape with their finger, freeze it, and then use the shape for tap detection.
The basic idea is this:
When your finger touches the screen, start recording positions. Discard any positions that are too close to the previous position (eg, only record a point if it is >min-distance from the last recorded point). While doing this, you can draw the UIBezierPath so you can see what you are tracing out. Modify the UIBezierPath by adding points to it, instead of recreating it every time.* When you lift your finger, close the bezier path. Quite simple.
Now, this will result in a polygon (ie, straight edges). If your min-distance is low enough or if you are using it for hit-detection (as you say), it won't really matter. However, if you want to smooth the path, you have to use the curve-to methods, which slightly complicate it - but should you wish to follow up on this more, read up on splines and spline generation from a point series.
*note: otherwise you'll get lag while drawing large shapes because recreating a bezier path from an increasingly large series of points gets expensive. Modifying the existing path makes it much, much, much faster.