Depth map estimation method and system

Abstract

The invention provides a depth map estimation method and system. The method comprises the following steps: carrying out multi-scale depth feature extraction on an image; uniformly sampling in a scene depth range to obtain depth hypotheses under different scales; under the depth hypothesis of each scale, the depth features of the neighborhood view are transformed to a reference view through microhomography transformation, and a cost body is constructed through group correlation measurement; regularizing the cost body based on a three-dimensional convolutional neural network, obtaining a depth probability body and an uncertainty probability body through a logistic regression algorithm, and respectively estimating a corresponding depth map and an uncertainty map; performing up-sampling and normalization on the uncertainty map and the depth map to obtain a sampling interval and a depth hypothesis of a next scale; and sampling depth hypotheses of different scales, and training and supervising the depth map and the uncertainty map under each scale through an uncertain sensing loss function. According to the scheme, the accuracy and reliability of image depth estimation can be effectively improved.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a depth map estimation method and system. Background technique [0002] Multi-view depth estimation aims to build dense correspondences from multiple images with known camera poses, thereby recovering the dense geometry from the reference viewpoint. In recent years, deep learning techniques have greatly promoted the development of multi-view depth estimation. Since the deep learning-based multi-view depth estimation method needs to use the 3D convolutional neural network to regularize the 3D cost volume, this makes the network memory consumption and computation time increase cubically as the input image resolution increases. [0003] In order to be able to estimate high-resolution depth maps and improve computational speed, the "coarse-to-fine" strategy is widely used in deep learning-based multi-view depth estimation. Such methods usually first perform depth sampling...

AI Technical Summary

Problems solved by technology

[0004] In view of this, an embodiment of the present invention provides a method and system for estimating a depth map, which is used to solve the problem of doubtful accuracy and reliability of the existing depth map estimation

Images