Recently, the problem of mesoscale estimation in visual tracking has been solved by combining scale space theory.
First post a scale space blog: Click to open the link
Then there are several methods based on the scale of the mean_shift, although now Mean_shift has no OVA.
Paper 1:
Mean-shift Blob Tracking through scale Space, cvpr,2003, should be considered in this direction very early paper, recently in research, may be finishing a technical document.
Paper2&3:
Mean-shift BLOB TRACKING with ADAPTIVE FEATURE SELECTION and scale adaptation & scale and Orientation ADAPTIVE MEAN Shift Tracking
Two domestic articles, the latter being the Zhanglei team, have code:http://www4.comp.polyu.edu.hk/~cslzhang/soamst.htm
Paper 4
Robust scale-adaptive Mean-shift for Tracking ideas are similar to the above.
The main focus of these days is to see these several papers