Method for motion vector estimation

Patent metadata

Field	Value
Publication number	US-9584824-B2
Application number	US-201314401145-A
Country	US
Kind code	B2
Filing date	Jun 25, 2013
Priority date	Jun 25, 2012
Publication date	Feb 28, 2017
Grant date	Feb 28, 2017

Abstract

A motion vector estimation method in video encoding: First, a feature point is detected in a video frame. Next, motion estimation is performed on the feature point to obtain a motion vector of the feature point. Next, the feature point is mapped to a feature image block, and the motion vector of the feature point is used as an initial motion vector of the feature image block. Finally, a distance between each image block and the feature image block is calculated, motion estimation is performed on the image blocks in an ascending order of the distances between the image blocks and the feature image block, and an obtained motion vector of an image block is used as an initial motion vector of an image block that is adjacent to the image block and has not undergone motion estimation, until motion estimation is completed for the image blocks in the entire video frame. The method can make an improvement to a conventional motion estimation method, mitigate influence of local optimization on a search method, and increase the accuracy of a motion vector obtained through motion estimation.

First claim

What is claimed is:

1. A motion vector estimation method, comprising:
detecting a feature point in a video frame, and performing motion estimation on the feature point to obtain a motion vector of the feature point;
using a window region that has the feature point being the center and has a preset pixel size as a feature block, and using the motion vector of the feature point as a motion vector of the feature block;
dividing the video frame into image blocks, determining whether an overlapping area between an image block in the video frame and the feature block is greater than a preset threshold value, and if yes, determining the corresponding image block as a feature image block;
using a weighted average of motion vectors of feature blocks having an overlapping relationship with the feature image block as an initial motion vector of the feature image block; and
calculating a distance between each image block and the feature image block, performing motion estimation on the image blocks in an ascending order of the distances between the image blocks and the feature image block, and using an obtained motion vector of an image block as an initial motion vector of an image block that is adjacent to the image block and has not undergone motion estimation, until motion estimation is completed for the image blocks in the entire video frame,
wherein the using a weighted average of motion vectors of feature blocks having an overlapping relationship with the feature image block as an initial motion vector of the feature image block comprises: calculating a sum of overlapping areas of the feature blocks having an overlapping relationship with the feature image block, calculating a ratio of an overlapping area between each feature block and the feature image block to the sum of overlapping areas, using the ratio as a weight coefficient, and using the weighted average of the motion vectors of the feature blocks as the initial motion vector of the feature image block.

2. The method according to claim 1 , wherein the detecting a feature point in a video frame comprises detecting a feature point in the video frame by using the FAST detection method.

3. The method according to claim 2 , after the detecting a feature point in the video frame by using the FAST detection method, further comprising: using the feature point detected by using the FAST detection method as an input to further detect the feature point in the video frame by using the Harris detection method.

4. The method according to claim 1 , wherein the performing motion estimation on the feature point to obtain a motion vector of the feature point comprises: performing motion estimation on the feature point to obtain the motion vector of the feature point by using a sparse optical flow method.

5. The method according to claim 1 , wherein the calculating a distance between each image block and the feature image block, performing motion estimation on the image blocks in an ascending order of the distances between the image blocks and the feature image block, and using an obtained motion vector of an image block as an initial motion vector of an image block that is adjacent to the image block and has not undergone motion estimation, until motion estimation is completed for the image blocks in the entire video frame comprises:
calculating the distance between each image block and the feature image block; calculating priority information of each image block, wherein an image block at a smaller distance from the feature image block has a higher priority; inserting each image block in a priority queue according to the priority information, wherein an image block having a higher priority is inserted at the head of the priority queue; performing motion estimation on the image blocks in the priority queue sequentially according to the priority queue, and using a motion vector of a current image block as an initial motion vector of an image block that is adjacent to the image block and has not undergone motion estimation, until motion estimation is completed for all image blocks in the priority queue.

6. The method according to claim 1 , wherein the calculating a distance between each image block and the feature image block, performing motion estimation on the image blocks in an ascending order of the distances between the image blocks and the feature image block, and using an obtained motion vector of an image block as an initial motion vector of an image block that is adjacent to the image block and has not undergone motion estimation, until motion estimation is completed for the image blocks in the entire video frame comprises:
calculating the distance between each image block and the feature image block; inserting the feature image block in a first-in-first-out queue, and performing motion estimation on an image block at the head of the first-in-first-out queue to obtain a motion vector of a current image block; and determining whether an adjacent image block of the current image block has not undergone motion estimation and has a distance from the feature image block being greater than the distance between the current image block and the feature image block, if yes, using the motion vector of the current image block as the initial motion vector of the adjacent image block, and inserting the adjacent image block in the first-in-first-out queue, and if not, ending motion estimation of the current video frame.

7. The method according to claim 2 , wherein the calculating a distance between each image block and the feature image block, performing motion estimation on the image blocks in an ascending order of the distances between the image blocks and the feature image block, and using an obtained motion vector of an image block as an initial motion vector of an image block that is adjacent to the image block and has not undergone motion estimation, until motion estimation is completed for the image blocks in the entire video frame comprises:
calculating the distance between each image block and the feature image block; inserting the feature image block in a first-in-first-out queue, and performing motion estimation on an image block at the head of the first-in-first-out queue to obtain a motion vector of a current image block; and determining whether an adjacent image block of the current image block has not undergone motion estimation and has a distance from the feature image block being greater than the distance between the current image block and the feature image block, if yes, using the motion vector of the current image block as the initial motion vector of the adjacent image block, and inserting the adjacent image block in the first-in-first-out queue, and if not, ending motion estimation of the current video frame.

8. The method according to claim 3 , wherein the calculating a distance between each image block and the feature image block, performing motion estimation on the image blocks in an ascending order of the distances between the image blocks and the feature image block, and using an obtained motion vector of an image block as an initial motion vector of an image block that is adjacent to the image block and has not undergone motion estimation, until motion estimation is completed for the image blocks in the entire video frame comprises:
calculating the distance between each image block and the feature image block; inserting the feature image block in a first-in-first-out queue, and performing motion estimation on an image block at the head of the first-in-first-out queue to obtain a motion vector of a current image block; and determining whether an adjacent image block of the current image block has not undergone motion estimation and has a distance from the feature image block being greater than the distance between t

Assignees

• Univ Peking Shenzhen Graduate School

Inventors

• Wang Ronggang
• Tang Chengzhou
• Gao Wen

Classifications

• H04N19/54

  using feature points or meshes · CPC title

• H04N19/513 Primary

  Processing of motion vectors · CPC title

• H04N19/56 Primary

  Motion estimation with initialisation of the vector search, e.g. estimating a good candidate to initiate a search · CPC title