3D Scene Reconstruction Method from Single Image

Abstract

The invention discloses a three-dimensional scene reconstruction method of a single image, comprising the following steps: an image input step, inputting each image in the image sequence; a feature extraction step, extracting features from the scene of the image, and based on the extracted features Perform scene classification and object recognition to obtain semantic information, and simultaneously extract monocular geometric information based on the extracted features, and obtain monocular geometric information; the object detection step, based on the extracted features, and refer to the scene classifying and performing object detection; a three-dimensional graphics primitive model selection step, selecting a three-dimensional graphics primitive model according to the result of the object detection; and a scene three-dimensional model generation step, according to the scene semantic prior, the three-dimensional graphics primitive model and the described The monocular geometric information is used to reason and verify the 3D model of the scene, so as to generate the final 3D model of the scene.

Description

technical field [0001] The invention relates to a three-dimensional scene reconstruction method of a single image, belonging to the technical field of computer vision and image processing. Background technique [0002] Restoring the 3D structure of a scene from a single image is a classic problem and a typical pathological problem in computer vision. The difficulty lies in that the image is projected from the 3D space to the 2D imaging plane by the scene content through the camera. Regarding the 3D structure of the scene The geometric information of is inevitably lost, which makes the problem ill-conditioned. However, human beings can accurately identify three-dimensional information such as the spatial structure and the relationship between distance and distance in the scene from an image through their own cognition. Therefore, in the field of computer vision, the 3D scene reconstruction of a single image has been committed to imitating human cognition, obtaining various i...

AI Technical Summary

Problems solved by technology

[0004] The obvious problem with the above method is that it is highly dependent on the data. Obviously, not all images will have the information they need, which makes it impossible to use the reconstruction method for reconstruction.

However, it still does not solve the above technical problems

Images