A camera does not (usually) a 6-degree-of-freedom accelerometer. Thus it as no direct way of knowing how it is moving. In principle, you could analyze multiple images acquired while the camera (or the subject) was moving, and figure out a (3D-to-2D) translation/rotation that transforms one image into another. You'll need good images and an understanding of image transformations. It is a challenging task ...
