The simplest method is to create frame difference outputs from both the videos. Frame difference is basically the subtraction of previous frame from the current frame of a video. The frame difference of a video having moving object will give you non-zero values at the position of the moving object, while the frame difference of the other video containing no moving object will give you zero (or near-zero) values.
I think what you're looking for is background subtraction. You can find a lot of information (links to publications, videos and even source code for a quite a few methods) here:
https://github.com/andrewssobral/bgslibrary
@Marek Kraft : sir please provide links for matlab codes also. Thankyou.
Check this below link.This maybe useful for tracking the object in the video and this link contains matlab code.
http://www.stats.ox.ac.uk/~wauthier/tracker/
If you compute the 3D spectrum (eg via an FFT) of the data sequence, stationary objects will have all their energy confined to the xy plane in frequency-space. Objects in motion will have their spatial frequency components "tilted" out of the xy plane, according to f_z = -v_x*f_x-v_y*f_y, where v_ x & v_y are the spatial velocities (in pixels/frame), f_x & f_y are the spatial frequencies (in cycles/pixel), and f_z is the temporal frequency (in cycles/frame). So to answer your question, videos containing no motion will contain only components with f_z = 0 in their spectra!
Actually, I have small cubes of videos and i want to find all those cubes which contains moving objects or part of the object which is moving. How can i do that?
Following is Matlab code for moving object detection using consecutive frame differencing:
close all;
clear all;
clc;
MOV = VideoReader('traffic.avi');% add your video file name here
vidFrames = read(MOV);
[row col color frame]=size(vidFrames);
implay(vidFrames);
% getting difference of consecutive image
for k=2:frame
i=vidFrames(:,:,:,k-1);
i1=vidFrames(:,:,:,k);
dif(:,:,:,p)=i1-i;
p=p+1;
end
implay(dif);
Result will be noisy so apply morphological operation at the end and use regionprops matlab function to get bounding box or centroid of cubes.
Frame differencing is better than background subtraction as far as the objective is to detect only the moving object(s), because background subtraction will give non-zero values even when the object has stopped moving (unless the background model is updated with time which increases the complexity of the method).
+1 for frame difference, obviously the difference with moving objects will be higher than non moving sequences
... If you only want to compute motion within small blocks of data, then the same relationship in the f domain applies - just compute the local DFT for that block of data. I have just written a couple of papers on this if you are interested.
Note that frame differencing only will work with a stationary camera.
If the camera has pan & tilt (or vibration) you first need to stabilize the image. Image based stabilization is hindered by moving objects, so you have to be smart here to solve the chicken-and-egg problem.
If your camera itself also is moving, and you are looking at 3D scene then it is extremely hard to differentiate between parallax effects which give apparent motion and true motion within the 3D scene. Options here include use of tools such as essential and fundamental matrices which state which motions can have occurred due to motion of the viewpoint rather than actual moving objects.
- Badges
- Science topic