UNSUPERVISED STEREO MATCHING APPARATUS AND METHOD USING CONFIDENTIAL CORRESPONDENCE CONSISTENCY

摘要

The present invention includes two convolutional neural networks (CNNs) having the same structure and the same weight and pre-trained by an unsupervised learning method, between an encoder and feature maps that extract feature maps from an input stereo image. A disparity for minimizing the matching cost volume among the disparity candidates having a matching maximum disparity range and a matching cost calculation unit for calculating the matching cost volume, and generating a disparity map from the obtained disparity It includes a parity map acquisition unit, and the two CNNs estimate a positive sample based on a correspondence point consistency according to an epipolar constraint for a disparity map obtained from a stereo image input during learning, and the estimated positive sample as an adjacent pixel It is possible to provide a stereo matching apparatus and method for learning by back propagating an error between learning maps generated by propagation and disparity maps.