Optical Flow Estimation in EC

Optical flow estimation can be used in motion trajectory, which has the following expression:

function x stand for intensity distribution of video sequence.

See reference here：

http://www.doc88.com/p-571889099829.html

The whole process of optical flow method in ec.

Very clearly and easy.

OF estimation and Temporal Regularization

Using the constant speed model of estimation, hehe, it was MVE indeed. I just laugh!!!!!

Spatial Regularization of FMV Field
One 7 X 7 median filter for intra pixel withou FMV.
Two Then a two-dimensional median filter in 16X16 window and kernel size larger than 10X10.
Three Rounded to 1/2 pixel.

Projection Onto Missing Frame
With 1/2 pixel decision, if multiple contribution to the same bin, average them.

Interpolation of Missing Pixels
Median of the pixels in a neighboring centered in the missing one.
The critical size 9 X 9 kernel.

Filtering and Downsampling
2 x 2 average kernel and then down sampling by two.

Extension to Loss of Multiple Frames
Regard the estimated frame t-1 as a good received frame.

Extension to the second frame
As in IPP… the second P is estimated from the previous P frame in front of I to reconstructed. The author has the same idea with me. Hahahaha…….

Concealment of Whole-Frame Losses for Wireless Low Bit-Rate Video Based on Multiframe Optical Flow Estimation