Local reference point-based quick point cloud registration method

Abstract

The invention belongs to the technical field of three-dimensional reconstruction, and discloses a local reference point-based quick point cloud registration method. The method comprises the steps of performing down-sampling on input original points cloud to obtain sparse point clouds corresponding to the original point clouds; obtaining local included angle features by taking points in the sparsepoint clouds as reference points; establishing a kd-tree by the local included angle features of a view angle; searching for nearest neighbor features of the local included angle features of the otherview angle; taking a k1 pair with a minimum Euclidean distance error in the nearest neighbor features as a seed matching pair; searching for other matching pairs under the reference from neighborhoods; calculating out an adjacent view angle transformation matrix by using rigid body transformation to finish rough registration of the point clouds; and optimizing a rough registration result by usinga partially overlapped ICP registration algorithm. According to the method, initial matching is established by utilizing quick included angle features, and the matching neighborhoods are used for propagation, so that the registration precision is ensured while the point cloud registration speed is increased.

Description

technical field [0001] The invention belongs to the technical field of three-dimensional reconstruction, in particular to a fast point cloud registration method based on local reference points. Background technique [0002] At present, the existing technologies commonly used in the industry are as follows: Vision is the most important way for humans to perceive the world, and video plays a key role in various vision-related applications. In the 3D reconstruction algorithm based on RGB-D images, the structure of the target object is still restored according to the motion information of the depth camera, that is, the structure is recovered from the motion. This technology is the same as the 3D reconstruction algorithm based on image sequences, but the two reconstruction algorithms estimate the There is a big difference in the way of motion information. The former first calculates the point cloud information of the target at each angle of view during the camera movement proces...

AI Technical Summary

Problems solved by technology

[0003] In summary, the problems in the prior art are: The existing three point cloud registration algorithms have small motion and high overlap rate; it is necessary to establish a description for each point in the point cloud, which is time-consuming

Images