Motion Stabilization

Abstract

Stabilization for devices such as hand-held camcoders segments a low-resolution frame into a region of reliable estimation, finds a global motion vector for the region at high resolution, and uses the global motion vector to compensate for jitter.

1 . A method of a digital signal processor for stabilizing a video sequence, comprising:
 (a) providing a low resolution version of an input picture;   (b) segmenting said low resolution version into reliable motion estimation blocks and unreliable motion estimation blocks;   (c) finding a single motion vector for the aggregation of said reliable motion estimation blocks;   (d) finding a global motion vector for said input picture from said single motion vector by scaling said single motion vector, wherein said scaling and includes a first scaling by a factor of 2 in both horizontal and vertical directions, a first single motion vector refinement by a local search, a second scaling by a factor of 2 in both horizontal and vertical directions, and second motion vector refinement by a second local search;   (e) compensating for jitter motion in said input picture using said global motion vector.   
   
   
       2 . The method of  claim 1  further comprising combining said reliable motion estimation blocks into a large block. 
   
   
       3 . The method of claim  5 , further comprising, terminating the method when the average of said relative motion vectors has magnitude larger than at least one of the threshold. 
   
   
       4 . A video camera, comprising:
 (a) means for providing a low resolution version of an input frame;   (b) means for decomposing said low resolution frame into blocks;   (c) for each of said blocks, means for computing motion vector prediction errors and when an average motion vector prediction error exceeds a minimum motion vector prediction error by a first threshold, means for designating said each of said blocks as a reliable block;   (d) for each of said blocks, means for computing a relative motion vector using a corresponding relative motion vector from a prior low resolution frame;   (e) means for taking integer N equal to the number of said reliable blocks with a relative motion vector which differs from the average of said relative motion vectors of all of said reliable blocks by more than a second threshold;   (f) when said N is larger than a third threshold, for each of said reliable blocks which has a relative motion vector greater than a fourth threshold, means for changing the designation from reliable to unreliable;   (g) means for finding a motion vector of an aggregate of said reliable blocks;   (h) means for extending said motion vector of step (g) to a global motion vector of a region in said input frame corresponding to said aggregate in said low resolution frame; and   (i) means for applying said global motion vector to stabilize said input frame.