Image quality of volumetric image reconstructions is often degraded by residual patient motion when using slowly rotating C-arm systems or radiotherapy linear accelerators. If the patient does not manage to hold his or her movements during data acquisition, the image resolution is impaired by strong motion blur artifacts. This work proposes a method to detect and to estimate arbitrary patient motion. The motion information is used within a motion-compensated variant of the FDK algorithm to improve the image quality. Results of two related experiments are shown. First, the motion is estimated in projection space by optical flow-based elastic image registration, using reference projections of a known static image. The motion-compensated FDK reconstruction from a simulated free breathing acquisition shows very good agreement with the objective image. Second, the motion is estimated from approximate reference projections that are computed by the SART method, using solely the acquired data.