In this paper we present a new methodology to extract the small error-prone motion recorded during a Computed Tomography (CT) imaging directly from the sinogram. In comparison to the use of the usual reconstructed images, this approach has the advantage to minimize the reconstruction errors and artifacts. Moreover, the B-spline modeling of the motion has brought about the required accuracy for the detection of such a small motion. The windowing used can act as a kind of time regularization, in case of volume motion detection.