Algorithm for determining the size of a room from a video stream

Question

Does anyone know about an image analysis algorithm with which I can determine how large (approximately, in real measurements, let it be width in meters or something else), the room is not included in one (or several) videos of this room

I currently use OpenCV as my optional image library, but I'm not very far from learning image analysis algorithms, just losing a name will be wonderful.

thanks

Edit: Well, a little clarification I just got from people. I do not control how the video stream is taken, and cannot guarantee the availability of several data sources. However, I have a certain place in the room, and I have to put something in this regard. Therefore, I would probably look at trying to determine the edges of the room, and then determine how far this space is in the room, and then guess how large the room is.

Answer 1

A terribly difficult (but interesting!) Problem.

If you are thinking of doing this in a fully automatic way, I think you will have many problems. But I think this is doable if the operator can mark the control points in the image set.

Your problem can be formulated in general terms as a search for the distance between two points in three-dimensional space, when you have only the locations of these points in two or more 2D images taken from different points of view. The process will work more or less as follows:

• Images will be displayed with information about camera location and orientation. For example, let's say that you get two photos with the same camera orientation and where two photos were taken with the camera at a distance of three feet from each other horizontally. You will need to determine the reference source for the 3D space in which the cameras are located, for example, you can say that there is a camera (0,0,0) in the left picture and the right image in (3,0, 0), and both will be directed forward, which will be the orientation (0,0,1). Or something like that.
• Now the operator comes and marks the two corners of the room in both pictures. Thus, you have 2 sets of 2D coordinates for each angle.
• You need to know the details of your camera and lens (field of view, lens distortion, aberration, etc.). The more you know about how your camera deforms the image, the more accurately you can make your assessment. This is the same as panorama firmware to achieve a better seam. See PanoTools for information on this.
• Here is the interesting part: now you will do a back projection for each of your 2D points. Perspective projection occupies a point in three-dimensional space and camera definition and calculates a two-dimensional point. This is used to represent three-dimensional objects on a flat surface, for example, on a computer screen. You do the opposite, for each two-dimensional point you try to get a three-dimensional coordinate. Since there is not enough information in the 2D point to determine the depth, the best thing you can do from one two-dimensional point is to get a line in three-dimensional space passing through the lens and through this point, but you do not know how far the point is from the lens. But you have the same two-dimensional point in two images, so you can calculate two 3D lines from different places on the camera. These lines will not be parallel, therefore they intersect at one point. The point of intersection of 3D lines will be a good estimate of the location of the 3D point in space, and in the coordinates of your reference camera 3D space.
• The rest is easy. When you evaluate the 3D locations of two points of interest, you simply calculate the 3D distance between them and the amount you want.

Pretty easy, huh?