Three-dimensional motion estimation of objects for video coding

摘要

Three-dimensional (3-D) motion estimation is applied to thenproblem of motion compensation for video coding. We suppose that thenvideo sequence consists of the perspective projections of a collectionnof rigid bodies which undergo a rototranslational motion. Motionncompensation can be performed on the sequence once the shape of thenobjects and the motion parameters are determined. We show that thenmotion equations of a rigid body can be formulated as a nonlinearndynamic system whose state is represented by the motion parameters andnby the scaled depths of the object feature points. An extended Kalmannfilter is used to estimate both the motion and the object shapenparameters simultaneously. The inclusion of the shape parameters in thenestimation procedure adds a set of constraints to the filter equationsnthat appear to be essential for reliable motion estimation. Ournexperiments show that the proposed approach gives two advantages. First,nthe filter can give more reliable estimates in the presence ofnmeasurement noise in comparison with other motion estimators thatnseparately compute motion and structure. Second, the filter cannefficiently track abrupt motion changes. Moreover, the structure imposednby the model implies that the reconstructed motion is very natural asnopposed to more common block-based schemes. Also, the parameterizationnof the model allows for a very efficient coding of the motionninformation