Complex large-scale scene-oriented three-dimensional reconstruction method based on GMS feature matching

Abstract

The invention discloses a complex large-scale scene-oriented three-dimensional reconstruction method based on GMS feature matching. The method comprises the following steps of 1, obtaining an image byadopting an SLAM method fused with GMS feature matching, and obtaining an initial value of a camera pose; step 2, optimizing an initial value of a camera pose by adopting a two-three-dimensional feature point combined cluster adjustment method; and step 3, by using the optimized camera pose, carrying out back projection on pixels of each image to a world coordinate system to obtain point cloud data, and carrying out point cloud fusion by using a surfel deformation graph construction method to obtain a three-dimensional reconstruction model. According to the method, more complex large-scale scenes can be well coped with during three-dimensional reconstruction, good camera pose estimation can be achieved in a texture sparse area, and large rotation and translation of the camera can be wellachieved.

Description

technical field [0001] The invention belongs to the technical field of three-dimensional reconstruction, and in particular relates to a complex large-scale scene-oriented three-dimensional reconstruction method based on GMS feature matching. Background technique [0002] At present, the 3D reconstruction of scenes and objects has a wide range of applications in many fields such as human-computer interaction, virtual reality, games and movies. There is also an increasing demand for more complex and realistic 3D models. However, most of the current 3D reconstruction systems are based on the method of laser scanning. Although this method can obtain a relatively ideal 3D model, the cost of the system is very expensive. If more complex and realistic 3D models can be obtained as easily as photos and videos, then 3D models will be used in more fields, such as drones and unmanned vehicles to perceive their own orientation in real time when they are in motion, and carry out plannin...

AI Technical Summary

Problems solved by technology

However, the efficiency of the bundle adjustment algorithm of this method is not high, and the algorithm takes too long

However, as the reconstruction scene continues to expand, the accumulation of errors will gradually increase, and the bundle adjustment will easily fall into a local optimal solution, and the errors will be difficult to eliminate

Images