Optical motion tracking based on stereo geometry is useful in medical imaging applications to provide input for downstream processing tasks such as motion compensation. For tracking markers containing closely spaced target points, pose measurements exhibit greater jitter than for markers with larger distributions of target points. This is especially true for out-of-plane rotational motion. Three time-domain filtering techniques for small markers were considered with the aim of improving motion-corrected positron emission tomography (PET) studies of conscious rats: a finite impulse response (FIR) pose filter, Kalman filter, and Kalman smoother. Actual measured rat head motion was applied to a phantom using a precise, 6-axis robot. Motion was measured using a tracking system with both large and small markers and both sets of data, with and without filtering applied, were used for motion correction. Correction based on the large marker resulted in the best agreement with a motion-free reference image. Blurring and degradation were evident when the small marker data were unfiltered prior to motion correction. Image contrast, resolution and peak activity concentrations were all improved using the FIR filter. The Kalman filters did not improve the correction, appearing to over-smooth. This would likely be improved with pose-by-pose noise modeling rather than using an average noise estimate as was done here. In conclusion, the results emphasize the importance of filtering noisy motion data derived from small markers in order to obtain more robust estimates of pose for motion correction.