In video processing algorithms sometimes needs to determine what type of motion was detected. We was need to classify two types of motion: motion of object and motion of camera.
Motion detection and definition of type of motion. Is it motion of object in frame or motion of camera?
The easiest way to solve this task is in using correlation between frames of video stream. But this solution require many system resources and many calculation. It hard to implement with HD videos because quantity of calculations has non-linear increase depends of video size. Also for this task should be used two areas on frame: border of frame and center part of screen. When the camera is moving, changes will be on both areas. But when only some objects is moving, in border of frame will be minimum changes.
One of good ideas was using scaling of video before processing and using not each frame but using frames with some step between them. It give ability to decrease quantity of calculation in thousends times, but in this case exists probability to lose motion of small objects on video.
Click on image to view it on full size