Edge Detection-Based Optical Flow Estimation Method

Abstract

Due to the edge jagged and blurred problem in conventional deep learning-based optical flow estimation methods, an edge detection-based optical flow model (EDOF) is proposed in this paper to improve the accuracy of optical flow estimation. In this model, the feature extraction module EDNet is used to obtain the features with the edge information of the objects in images, while the other feature extraction module OFNet extracts the convolutional features with other common features such as the texture and color information of the object and others. The two kinds of features are fused to EstiNet, and then the estimation result is obtained by input the fused features into the optical flow network. Experiments on the public MPI Sintel and Flying Chairs datasets show that the EDOF method can reduce the average endpoint error of optical flow estimation.