Ground imaging from an airborne synthetic aperture radar (SAR) and conversely aircraft imaging from a ground based radar (ISAR) require the knowledge of the motion of the antenna relative to the target, with an accuracy of a fraction of the wavelength upon a time span of seconds. This information is not necessarily directly available from navigation/tracking data and must be recovered from the radar data itself using the technique well known as autofocus. The approach described in this paper divides the problem into two time scales. The fast disturbances (antenna vibrations, tracking noise) of the motion with respect to the nominal trajectory are handled before range compression by the complex valued correlation technique. The slow disturbances (inertial drifts, error accumulation) are handled by a frame drift-like technique, derived from image sequence analysis. As side effect, this method may also provide alternative techniques for moving target detection and for accurate odometry (measure of the ground speed of the carrier aircraft).