| CPC G06T 5/50 (2013.01) [G06V 10/44 (2022.01); G06V 10/803 (2022.01); G06V 10/82 (2022.01); H04N 5/145 (2013.01); G06T 2207/20221 (2013.01)] | 17 Claims |
|
1. A method of forming an output image, the method comprising:
extracting a plurality of features from a plurality of images, the plurality of features comprising a plurality of first features associated with a first image of the plurality of images and a plurality of second features associated with a second image of the plurality of images;
performing a global motion estimation based on the plurality of features to obtain a global optical flow estimate;
performing an optical flow estimation based on the plurality of features to obtain a local optical flow estimate;
fusing the global optical flow estimate and the local optical flow estimate to obtain a fused optical flow estimate; and
forming the output image based on the fused optical flow estimate, wherein the performing the global motion estimation comprises:
estimating, based on the plurality of features, fitting weights using a first neural network to obtain robust weights; and
motion fitting, using a direct linear transform (DLT), based on the robust weights and based on the local optical flow estimate to obtain the global optical flow estimate, wherein the global optical flow estimate comprises a fundamental matrix.
|