3D-RGB point cloud registration method based on local gray scale sequence model descriptor

Abstract

The invention belongs to the technical field of computer vision, image processing and three-dimensional measurement, and particularly relates to a 3D-RGB point cloud registration method based on a local gray scale sequence model descriptor. The method comprises the following steps: 1, calculating a four-neighborhood gray average value of each point in two point clouds; 2, dividing adjacent pointsof the key points into six parts according to gray values, and finally connecting the feature vectors of the six parts in series to form a key point feature descriptor; 3, constructing a point-to-point mutual corresponding relationship between the source point cloud and the target point cloud according to a nearest neighbor ratio method and an Euclidean distance threshold, and removing an error corresponding relationship by utilizing random sampling consistency and color consistency; and 4, solving a conversion matrix between the source point cloud and the target point cloud by using the corresponding relationship, and performing spatial transformation on the source point cloud to complete registration of the point cloud. According to the method, the influence of unobvious geometric information and light intensity change on point cloud registration can be effectively reduced. The method has a wider application range. The precision and the robustness of three-dimensional point cloud registration are improved.

Description

technical field [0001] The invention belongs to the technical fields of computer vision, image processing, and three-dimensional measurement, and in particular relates to a 3D-RGB point cloud registration method based on a local grayscale sequential model descriptor. Background technique [0002] The 3D reconstruction of real objects is an important research topic in the fields of computer vision, computer-aided geometric design, and computer graphics. It involves many fields such as computer graphics, image processing, and pattern recognition. research hotspots and difficulties. When using the acquisition equipment for point cloud acquisition, affected by the spatial position, geometric shape and measurement method of the measured object, a single scan can only obtain the local point cloud data of the object. Generally, it is necessary to scan the object from different perspectives to obtain The complete point cloud data of the object, and the point cloud data measured fro...

AI Technical Summary

Problems solved by technology

However, these algorithms only use the spatial shape information of the point cloud, and need to estimate the local consistent direction of adjacent sampling points, such as normal, curvature, local coordinate system, etc. to achieve rotation invariance, and the estimation of the direction is prone to errors. There are deficiencies in anti-noise performance and registration accuracy when the geometric shape information of the point cloud to be registered is not obvious and some point clouds are missing.

Images