imagesAlignSeq

 Stabilize image sequence using coarse optical flow estimation.

 Perform weak image sequence stabilization as described in:
  D. Park, C. Zitnick, D. Ramanan and P. Dollár
  "Exploring Weak Stabilization for Motion Feature Extraction", CVPR 2013.
 The approach stabilizes coarse motion due to camera movement but leaves
 independent object motions intact. This code performs weak sequence
 stabilization only (no feature extraction), see section 3.1 of the paper.
 Please cite the above paper if you end up using the stabilization code.

 Optical flow is computed between all pairs of frames using opticalFlow.m
 with params 'pFlow'. Flow across multiple frames is accumulated and flow
 at integer locations at each stage is obtained either using either
 nearest neighbor (if type==1) or bilinear interpolation (if type==2)
 which is slightly more accurate but slower. Finally, all images in I are
 warped to the last frame of the sequence using the accumulated flows. If
 type==-1, a homography computed via imagesAlign.m with params pFlow is
 used instead. If videos have black boundaries use bndThr to ignore dark
 boundaries for flow estimation (with average pixel values under bndThr).

 USAGE
  [J,Vxs,Vys] = imagesAlignSeq( I, pFlow, [type], [bndThr] )

 INPUTS
  I          - HxWxN or HxWx3xN input image sequence
  pFlow      - parameters to use for optical flow computation
  type       - [1] interpolation type 1:nearest, 2:bilinear
  bndThr     - [0] fill in black areas around image boundaries

 OUTPUTS
  J          - HxWxN or HxWx3xN stabilized image sequence
  Vxs        - HxWxN-1 x-components of flow fields
  Vys        - HxWxN-1 y-components of flow fields

 EXAMPLE
  I = seqIo(which('peds30.seq'),'toImgs'); I=I(:,:,:,1:15);
  pFlow={'smooth',1,'radius',25,'type','LK','maxScale',1};
  tic, J = imagesAlignSeq( I, pFlow, 1, 20 ); toc
  playMovie([I J],15,-10,struct('hasChn',1))

 EXAMPLE
  % Requires Caltech Pedestrian Dataset to be installed
  [pth,ss,vs]=dbInfo; s=randi(length(ss)); v=randi(length(vs{s}));
  nm=sprintf('%s/videos/set%02i/V%03i.seq',pth,ss(s),vs{s}(v));
  f=seqIo(nm,'getinfo'); f=f.numFrames; f=randi(f-30);
  I=seqIo(nm,'toImgs',[],1,f,f+9);
  pFlow={'smooth',1,'radius',25,'type','LK','maxScale',1};
  tic, J = imagesAlignSeq( I, pFlow, 1, 20 ); toc
  playMovie([I J],15,-10,struct('hasChn',1))

 See also opticalFlow, imtransform2, imagesAlign

 Piotr's Computer Vision Matlab Toolbox      Version 3.24
 Copyright 2014 Piotr Dollar.  [pdollar-at-gmail.com]
 Licensed under the Simplified BSD License [see external/bsd.txt]