Recovering camera motion from video relies on SfM (Structure from Motion) or visual odometry techniques, whose core principle is to detect and match feature points across consecutive frames and use epipolar geometry constraints to estimate the camera's 3D pose changes.