A continuous-discrete filtering strategy is proposed for cardiac kinematics estimation from periodic medical image sequences. Stochastic multi-frame filtering frameworks are constructed to deal with the parameter uncertainty of the biomechanical constraining model and the noisy nature of the imaging data coordinately. For the system with continuous dynamics and discrete measurements, the state estimates are predicted according to the continuous-time state equation between observation time points, and updatedwith the new measurements obtained at discrete time instants, yielding physically more meaningful and more accurate estimation results for the continuously evolving cardiac dynamics. The strategy is validated through synthetic data experiments to illustrate its advantages and on canine MR phase contrast images to show its clinical relevance.