The paper derives an optimal fixed interval smoother for the linear time invariant output estimation problem. The smoother employs forward and adjoint Kalman predictors. The suboptimal time-varying case is discussed which includes the use of forward and adjoint extended Kalman predictors within a nonlinear smoother. The efficacy of the smoother to tracking points within sequences of noisy images is investigated. The results of a simulation study are presented in which it is demonstrated that a Kalman filter can outperform a so-called matched filter, and a fixed interval smoother can provide a further performance improvement when the images are sufficiently noisy.