Designated target three-dimensional reconstruction method and system

Abstract

The invention discloses a designated target three-dimensional reconstruction method and system. The method comprises the steps of obtaining a multi-view image; determining an appointed target image needing to be reconstructed; performing scene recovery on the multi-view image by adopting SFM to obtain a scene sparse structure of the multi-view image, and marking a part of 3D points in the scene sparse structure of the multi-view image to obtain a training sample set; constructing a strong classifier by adopting an AdaBoost algorithm based on the training sample set; classifying the rest 3D points in the scene sparse structure by adopting a strong classifier to obtain a sparse point cloud structure of a specified target; determining a dense point cloud structure of the scene of the multi-view image by adopting an MVS based on the sparse point cloud structure; and based on the dense point cloud structure, taking the training sample set as training data, and determining a specified targetdense point cloud structure by adopting a multi-decision tree judgment strategy. According to the method, three-dimensional reconstruction of the designated target can be achieved only through a small number of sparse points under one viewpoint, the calculated amount is small, and precision is high.

Description

technical field [0001] The invention relates to the field of three-dimensional reconstruction, in particular to a method and system for three-dimensional reconstruction of a designated target. Background technique [0002] Image-based 3D object and scene reconstruction is one of the important research directions in the field of computer vision, whether it is based on single-viewpoint or multi-viewpoint images, there are extensive researches. Due to the lack of depth information of the target in single-view images, reconstruction often requires varying degrees of human interaction, or the use of modeling rules to infer the target structure. Mapping spatial structure from a single-viewpoint image is a typical ill-conditioned problem, and more methods are based on multi-viewpoint images. However, most of the methods are designed for the reconstruction of the whole scene. The reconstruction result is that if the whole scene wants to obtain the model of the specific objects in t...

AI Technical Summary

Problems solved by technology

However, most of the methods are designed for the reconstruction of the entire scene. The result of the reconstruction is that if you want to obtain the model of the specific objects in the scene as a whole, you need to further perform target recognition on the scene model.

Like image target recognition, 3D target recognition also needs to collect enough images, depth maps, etc. as target samples in advance to train the model. When the number of samples is small or even non-existent, it is difficult to recognize such targets, which also leads 3D models of specific objects cannot be obtained from the overall scene

Therefore, the target specified in the scene cannot be reconstructed in the prior art

Images