Multi-view stereo vision method integrated with spatial propagation and pixel-level optimization

Abstract

The invention discloses a multi-view stereo vision method integrated with spatial propagation and pixel-level optimization, which belongs to the field of computer vision. The multi-view stereo visionmethod comprises the following steps: camera parameters are obtained according to a plurality of pictures of a scene, a depth image of each picture is extracted according to the camera parameters, andspatial propagation is utilized to update the depth image of each picture; pixel-level optimization and filtration are iteratively carried out on the updated depth images, so that processed picturesare obtained; and depth image fusion is concurrently carried out on each pixel in the processed pictures, so that a three-dimensional point cloud of the scene is obtained. In initial depth image extraction, the multi-view stereo vision method utilizes spatial propagation to reduce redundant computation between images, thus increasing efficiency; in addition, by iterative depth image optimization and filtration, while noise points are reduced, the integrity of the depth images is increased, and ultimately, while the efficiency is increased by highly efficient depth image fusion, the density ofthe point cloud is ensured.

Description

technical field [0001] The invention belongs to the field of computer vision, and more specifically relates to a multi-view stereoscopic vision method combining spatial propagation and pixel-level optimization. Background technique [0002] Reconstructing dense 3D models of scenes from multiple images has been a hot topic in computer vision for the past few decades. The general technical process of this problem is very mature: input a set of pictures of a certain scene, first obtain the camera parameters through calibration or motion recovery structure (Structure From Motion, SfM); then use the estimated camera parameters to obtain the dense three-dimensional scene Point cloud, this technology is called multi-view stereo vision (Multi-View Stereo, MVS). The resulting dense point clouds can be used in many fields: classification, image-based rendering, localization, archaeological research, artistic creation, and more. Although the MVS algorithm has been studied for decades...

AI Technical Summary

Problems solved by technology

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a multi-view stereo vision method combining spatial propagation and pixel-level optimization, thereby solving the problems of low efficiency, incomplete and incomplete 3D point cloud obtained in the prior art. Technical issues with low precision

Images