Model-based
Centre for Vision, Speech & Signal Processing
BASIC LIBRARIES AMMA SEARCH AMMA HOME
 

Robust model-based motion estimation package

This package is for motion estimation of image patches based on matching. The technique is known to be successful whenever there is enough detail for the required patch and if the appropriate pre-filtering has been used on the frames (please see MatchMotEstC). Relatively little knowledge of the method is required in order to be able to use it. Interested users are refered to M.Tekalp, "Digital Video Processing", Chapters 5, 6 and 7 (available in the CVSSP library).

These classes use robust optimisation to find the optimum motion model within a patch (a region) of a pair of images. The currently recommended starting point is MultiMatchMotEstC (templated on the motion model and optimiser type), which contains an example of its use. The subtree Motion Model contains the classes of the various motion models needed for the template.

The following tasks are remaining (please feel free to undertake any):-

  • perspective motion estimation
  • multiresolution framework
  • overlapping motion estimation
  • allowing for mean-level shifts
  • estimation of confidence in the motion vector
  • calculation of match measure using all three colour channels
  • optimisation of code

    • Subtopics:

        AMMA
            Image
                Motion
                    Estimation
                        Model-based
                            Motion Model

    Classes:

    Default Classes Description
    LMSRegionMatchC Computes a 2D motion vector from 2 images over a region using a least-mean-square fit
    NumCostMotEstC Optimisation cost function for image patch comparison.
    NumCostMotEstBC Implementation class for cost function used for comparing image regions.
    NumCostMotEstBaseC Optimisation cost function for image patch comparison.
    NumCostMotEstBaseBC Implementation class for cost function used for comparing image regions.
    NumOptimiseImageDescentC Gradient descent optimisation algorithm for motion estimation.
    NumOptimiseImageDescentBC Gradient descent optimiser implementation class for motion estimation
    NumOptimiseDiehlMotEstC Gradient descent optimisation algorithm for motion estimation.
    NumOptimiseDiehlMotEstBC Gradient descent optimiser implementation class for motion estimation
    MatchMotEstBC
    MatchMotEstC Class for motion estimation
    MultiMatchMotEstC Motion estimation with templated motion model
    MatchMotEstBaseBodyC Put a brief description of your class body here
    MatchMotEstBaseC Put a brief description of your class here
    MultiMatchMotEstBC
    MultiMatchMotEstBaseBodyC Put a brief description of your class body here
    MultiMatchMotEstBaseC Put a brief description of your class here
    Author: Anonymous, Generated by DocEntry: March 20, 2001