is there a computer program to extract values from a video?

reki

is there a computer program to extract values from a video? (like size, speed of an object, dimensions, etc.)

mathman

Are you asking about the video parameters or something about the contents? If it is the contents I don't know of any, but it sounds like a hard problem.

As for the video parameters, many of the video playing programs, such as Window Media Player or Realplayer provide much of the information, depending partly on the video format.

reki

I'm asking about objects from the content of the video.
Saw it in a couple of movies and it could be helpful for me in a personal project.
Should exist, maybe in criminology...

hamster143

At the present level of technology, video is not coded in terms of objects. It's just a flat rectangular picture and a set of rules that prescribe how to construct frame #N+1 from frames #N, #N-1, etc. The best you can do is calculate the speed of movement of some point of the video, in pixels per second. You can't backtrack from that to physical movement. Most of the time, if you want physical movement, you need videos from two cameras with predetermined distance between them, and predetermined optical properties such as field of view angle and optical distortion characteristics (though there are tricks that will let you deduce field of view and optical distortion from the video).

Things you see in movies often have no relation to the state of technology in the real world. It always gets me in movies and detective shows, when "technicians" take a grainy 320x240 shot from an ATM camera and use it to zoom in onto a license plate in the background. CSI and Law & Order are often guilty of that. I actually work on digital video for a living and I like to think that I know what can or can't be done.

There's been a shift towards three-dimensional video recently (heard of Avatar?) and, with that kind of video (which is basically two parallel tracks from two different fixed cameras), you can do some 3D reconstruction using 3D algorithms. Even then it's not easy. There's an open-source library called OpenCV, but I haven't personally worked with that one. In my opinion, 10 years from now there will be a universally adopted true 3D standard, but, for now, that work is in its infancy.

rcgldr

Using multiple cameras along with knowledge of ball size and the court layout, systems like Hawk Eye can track objects very accurately.

http://en.wikipedia.org/wiki/Hawk-Eye

hamster143

Tracking a small unobstructed ball with multiple cameras is a relatively easy task.

Now imagine doing the same with a human. Step 1 is to identify views of the same features through different cameras. You'll get a good fix on some sharp features such as eyes, corners of the mouth, and buttons. The computer will draw a blank in all flat areas. It can't match contours, unless it has a built-in 3D model of the object (easy with a rigid ball, harder with, say, a car, very hard with a fully clothed human). If cameras are too close too each other, precision of the reconstruction goes down. If cameras are too far, some features end up visible through only one camera.

When Hollywood CG designers do motion tracking, they use 5+ cameras and many compact reflective markers on the actor's body, to aid the computer.

nbo10

Some of the high end high speed camera manufacturers, provide software that have the ability to perform the tasks you ask.

dE_logics

There are various player that can do this...for e.g smplayer (on Linux).

Install mplayer and see.