Pulse or digital communications – Bandwidth reduction or expansion – Television or motion video signal
Patent
1997-11-21
2000-12-05
Le, Vu
Pulse or digital communications
Bandwidth reduction or expansion
Television or motion video signal
3484161, 348699, H04N 712
Patent
active
061576778
DESCRIPTION:
BRIEF SUMMARY
FIELD OF THE INVENTION
This invention relates generally to the parameterization of each of a set of large, related data records. More specifically, it concerns improved motion estimation in sets of related signal records, e.g. video frames.
BACKGROUND OF THE INVENTION
Within video modelling and compression, motion estimation and motion compensation is important. Without it moving objects and other motions are difficult to describe efficiently in applications like video compression and interactive video games. Singh Ajit (1991, Optical Flow Computation. IEEE Computer Society Press) describes general methods for motion estimation.
Motion estimation is usually done from one frame to another frame, say, from a frame m to a frame n, for whose intensities we use the term I.sub.m and I.sub.n.
When good statistical precision and accuracy of the motion estimation is required, it is important to use all the available information efficiently in the motion estimation. This means that if moving physical objects or phenomena are repeatedly observed in several frames, increased precision and accuracy may be attained if the motion estimation is coordinated between these repeated observation frames.
However, motion estimation is normally a computationally demanding operation, in particular when full motion fields, with one vertical and horizontal motion parameter for each individual pixel in I.sub.m or I.sub.n are to be determined.
Motion estimation can also be very memory demanding. Any simultaneous motion estimation for many frames is bound to be exceptionally demanding.
On the other hand, full motion field estimation on the basis of only two individual frames is underdetermined: For many pixels, an equally good fit can be found with a number of different motion estimates, although only one of these corresponds to the original physical movement of the objects imaged.
When physical objects can be observed to move systematically over several frames, their motions are generally such that if their true two-dimensional (2D) motion fields had been known, these would have systematic similarities from frame to frame. Due to these systematic similarities, the motion fields of a number of related frames could theoretically be modelled with relatively few independent parameters. This modelling would in turn have led to very efficient compression and editing techniques for video.
However, in practice, the true motion fields cannot be determined from empirical data. First of all there will be more or less random errors in the obtained motion fields due to more or less random noise contributions in the raw data. Worse, due to the underdetermined nature of full motion estimation the probability of finding the `true` motion field is low. A different set of spurious false motion estimates may be chosen for each frame.
Thus, existing methods and apparatuses for determining motion for a number of frames, based on individual frame pairs, have several drawbacks:
1. The lack of coordination in the motion estimation for the different frames makes it difficult to model the set of motion estimation fields efficiently and hence attain good compression of these without loss of fidelity, and good editability control.
2. The motion estimates are unnecessarily imprecise due to sensitivity to random noise in the images, since the methods do not employ the stabilizing fact that the same non-random objects or phenomena are seen in several frames.
3. The motion estimates are unnecessarily inaccurate due to the underdeterminate nature of many motion estimation problems. The imprecise, inaccurate results represent an over-parameterization that may fit the individual pairs of frames well, but have bad interpolation/extrapolation properties, and do not give good approximations of the true, but unknown physical motions.
4. Attempts at coordinating the motion estimation by treating many frames in computer memory at the same time are computationally and memorywise very demanding.
OBJECTS OF THE INVENTION
It is therefore an object of the present invention
REFERENCES:
patent: 4717956 (1988-01-01), Moorhead et al.
patent: 5392072 (1995-02-01), Rodriguez et al.
patent: 5983251 (1999-11-01), Martens et al.
Soatto et al, "Dynamic Visual Motion Estimation From Subspace Constraints", IEEE, 1994.
Davis et al, "Equivalence of Subpixel Motion Estimators Based on Optical Flow and Block Matching", IEEE, 1995.
Martens Harald Aagaard
Reberg Jan Otto
IDT International Digital Technologies Deutschland GmbH
Le Vu
LandOfFree
Method and apparatus for coordination of motion determination ov does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for coordination of motion determination ov, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for coordination of motion determination ov will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-968077