Moving boundary guided optical flow filtering method based on collaborative deep neural network

Abstract

The invention discloses a moving boundary guided optical flow filtering method based on a collaborative deep neural network, and the method comprises the steps: constructing a moving boundary guided optical flow filtering data set and the collaborative deep neural network, inputting an initial optical flow estimation result and a moving boundary, and outputting a filtered optical flow estimation result, comprising an initial optical flow feature extraction sub-network, a moving boundary feature extraction sub-network, a first optical flow filtering sub-network and a second optical flow filtering sub-network, training the collaborative deep neural network by using the training set, and filtering an initial optical flow estimation result by using the trained collaborative deep neural network, and quickly generating an optical flow estimation result with higher precision. According to the method, the collaborative deep neural network is used for automatically learning the optical flow filtering process guided by the moving boundary, the complex function relation from variable input to optical flow filtering result output is accurately simulated, new errors are prevented from being introduced into irrelevant edge information except the moving boundary, and the efficiency and accuracy of optical flow filtering are improved.

Description

technical field [0001] The present invention relates to image processing and motion estimation technology, and specifically refers to an optical flow filtering method guided by a motion boundary based on a collaborative deep neural network. Background technique [0002] Optical flow is the two-dimensional instantaneous velocity vector field of all pixels in a video image. As one of the core issues in the field of computer vision, optical flow estimation is the basis of image processing and motion estimation. It has a very wide range of applications in object detection, object recognition, object tracking, object segmentation, video denoising, and video super-resolution. application. The motion boundary is the discontinuous boundary of the optical flow, which divides the optical flow into several regions, and the optical flow value inside each region satisfies the smoothness characteristic. Using the motion boundary to guide the initial optical flow to filter can filter out...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the above problems, the present invention aims to provide an optical flow filtering method guided by the motion boundary of a cooperative deep neural network, which uses a large number of sample data to automatically learn the structural information in the motion boundary, and guides the filtering optimization of the optical flow to solve the current problem. There are problems such as inaccurate modeling of optical flow filtering method and introduction of new errors

Images