An efficient large displacement flow estimation method

The present invention discloses a kind of large displacement efficient optical flow estimation method, comprising: obtaining two successive images from the video, the two images are in chronological order and labeled I I; I I and to each of the bottom and the image pyramid structure the image pyramid is generated on each of the same number of seed points, matching the initialization seed point on the top layer and a random value; seed point obtained sequentially from top to bottom to match the image pyramid layer by layer, each layer It matches the seed point as the corresponding seed point lower initial value; matching result the bottom of the seed point by an edge interpolation algorithm sensitive interpolated, the interpolation result as optical flow estimation initial value, then the variational energy optimization model optimized to give a final optical flow estimation result a large displacement. The present invention has a more efficient, more flexible effect, in accordance with any number of different control scenarios seed point, to produce different efficiencies, different precision optical flow results.