Lsd-slam-based real-scene three-dimensional reconstruction method

Abstract

The invention discloses an lsd-slam-based real-scene three-dimensional reconstruction method. The method comprises the steps of S1, recording a to-be-modeled scene, and collecting video data that covers the to-be-modeled scene; S2, resolving a key frame gesture from a key frame to obtain the restored images of the three-dimensional information based on the lsd-slam modeling technology, and generating a three-dimensional space point cloud; S3, selecting images for being spliced into a panoramic image out of all the restored images of the three-dimensional information according to the space location relationship, and generating the panoramic image; S4, putting the panoramic image and the three-dimensional space point cloud into the same spatial reference system, and subjecting three-dimensional data to the monomer treatment process. In this way, the key frame can be obtained through acquiring and analyzing video data. After that, the three-dimensional space point cloud and the panoramic image are generated through analyzing the key frame. The three-dimensional reconstruction is achieved based on the three-dimensional space point cloud and the panoramic image. Therefore, the calibration of equipment is not required and the method can be implemented without any hardware equipment. Meanwhile, panoramic data and model data are collected, so that the problems during the existing three-dimensional reconstruction process are solved.

Description

technical field [0001] The invention relates to a three-dimensional reconstruction method in the field of computer vision, in particular to a real scene three-dimensional reconstruction method based on lsd-slam. Background technique [0002] Three-dimensional reconstruction refers to the establishment of mathematical models suitable for computer representation and processing of three-dimensional objects. The usual 3D reconstruction method is to calibrate the camera first, that is, to calculate the relationship between the camera’s image coordinate system and the world coordinate system. Then use the information in multiple 2D images to reconstruct 3D information, and the 3D reconstruction based on photogrammetry Method In the beam adjustment process of photos, the distribution of the residuals leads to the distortion of the light, which makes the texture of the 3D model distorted and deformed, and the intuitive display effect is not as good as that of the panoramic image. ...

AI Technical Summary

Problems solved by technology

The usual 3D reconstruction method is to calibrate the camera first, that is, to calculate the relationship between the camera’s image coordinate system and the world coordinate system. Then use the information in multiple 2D images to reconstruct 3D information, and the 3D reconstruction based on photogrammetry Method In the beam adjustment process of photos, due to the distribution of residuals, the light is distorted, which makes the texture of the 3D model distorted and deformed, and the intuitive display effect is not as good as that of the panoramic image.

[0003] In order to ensure that the effect of 3D reconstruction meets the display requirements, panoramic images are usually used for 3D reconstruction at present. However, the 3D reconstruction of panoramic images requires the use of depth information. At present, the acquisition of depth information basically depends on panoramic cameras and RGB-D cameras. As a result, the current 3D reconstruction of panoramic images is highly dependent on hardware equipment, requiring panoramic cameras and RGB-D cameras (cameras with ranging sensors) are difficult to popularize

Secondly, the current 3D reconstruction of panoramic images still needs to calibrate the panoramic camera and RGB-D camera, which increases the workload and accumulates errors

Moreover, at present, panorama data and model data for 3D reconstruction of panoramic images are collected separately, which is inefficient, and the calculation speed based on sift algorithm is relatively slow

Finally, the current 3D reconstruction results of panoramic images have not been singled out, and the object management of 3D entities is not possible, which is difficult to meet the deep-level application requirements.

Images