| 主权项 |
1. A method for estimating a size of an object present on two images representing the same scene from different points of view, the method comprising:
determining an AAM image model; matching the AAM image model to the images to find a set of parameters describing the shape of the model (pc) and the appearance of the model (λc) for which minimal matching errors for both images occur by:
calculating the matching error for the first image as a sum of two summands: a first summand, related to the first image, calculated as a sum of squares of the difference between the appearance of the AAM image model for the first image at the pixel (x) and the intensity of the input first image (IL) at the pixel determined by the affine warp function ((x,pL)), which warps the actual shape for the first image described by parameters (pL) onto the mean shape (SL0) of the first image; a second summand, related to both images, calculated as a sum of squares of the difference between the appearance of the AAM image model for the first image at the pixel (x) and the intensity of the input second image (IR) at the pixel determined by the modified affine warp ((x,pR)), which warps the actual shape for the second image described by (pR) parameters onto the mean shape (SL0) of the first image;
calculating the matching error for the second image as a sum of two summands: a third summand, related to the second image, calculated as a sum of squares of the difference between the appearance of the AAM image model for the second image at the pixel (x) and the intensity of the input second image (IR) at the pixel determined by the affine warp function ((x,pR)), which warps the actual shape for the second image described by parameters (pR) onto the mean shape (SR0) of the second image; a fourth summand, related to both images (left and right), calculated as a sum of squares of the difference between the appearance of the AAM image model for the second image at the pixel (x) and the intensity of the input first image (IL) at the pixel determined by the modified affine warp ((x,pL)), which warps the actual shape for the first image described by (pL) parameters onto the mean shape (SR0) of the second image; and estimating the size of the object based on the differences in locations of points of the AAM image model in the images. |
