Automatic subarea selection for image registration

A computer-implemented method for optimally registering a current real-time image to a reference image, said method comprising the steps of:

• retrieving a reference image from a storage device;
• filtering the reference image using a plurality of nonlinear filters to produce an inverse covariance matrix for each of a plurality of image subareas of the reference image;
• processing the plurality of inverse covariance matrices to select a set of candidate image subareas based upon the relative values of the inverse covariance matrices for use in registering the current real-time image to the reference image;
• processing the set of candidate image subareas to generate a list of subareas that optimally register the current real-time image;
• calculating covariance propagation weights for the reference image using a predetermined uncertainty model, which weights are indicative of the error covariance between the reference image and the current real-time image;
• processing the list of subareas, the inverse covariance matrices, and the covariance propagation weights to produce signals indicative of a set of reference subareas that is used to optimally register the current real-time image;
• selecting a set of subareas from the current real-time image that correspond to the set of reference subareas;
• correlating the selected set of subareas from the current real-time image to the set of reference subareas to produce difference signals indicative of the difference in registration between the reference image and the current real-time image.

A computer-implemented method for optimally registering a current real-time image to a reference image, said method comprising the steps of:

• retrieving a reference image from a storage device;
• filtering the reference image using a plurality of nonlinear filters to produce an inverse covariance matrix for each of a plurality of image subareas of the reference image;
• processing the plurality of inverse covariance matrices to select a plurality of candidate image subareas based upon the relative values of the inverse covariance matrices for use in registering the current real-time image to the reference image;
• calculating covariance propagation weights for the reference image using a predetermined uncertainty model that is indicative of the error covariance between the reference image and the real-time image;
• calculating the covariance of polynomial error coefficients of each of the candidate subareas using a previous covariance of polynomial error coefficients and the calculated covariance propagation weights;
• calculating a total mean squared registration error for each of the candidate subareas in terms of position and predicted measurement covariance;
• and
• outputting a list of subarea positions derived from the total mean squared registration error that are used to correlate the candidate subarea& which correlated subareas are used to register the current real-time image to the reference image.

A computer-implemented method for guiding a vehicle toward a target using video imagery, and wherein a reference image corresponding to the target has been stored, said method comprising the steps of:

• processing the reference image to determine the relative uncertainty in locating the actual position of an area of interest to provide selected subareas in the reference image that are optimal in registering the live image;
• processing the live imagery using a subarea extraction procedure to extract subareas from the live imagery that correspond to the selected subareas;
• correlating the selected subareas and the extracted subareas to provide output signals indicative of the error therebetween;
• coupling the output signals to the vehicle to provide guidance signals thereto;
• and
• steering the vehicle in response to the guidance signals, which guidance signals optimally register the live image to the reference image that corresponds to the target.