Rapid selection and registration method for 3D laser radar point cloud data prediction points

Abstract

A rapid selection registration method for 3D laser radar point cloud data prediction points comprises the following steps: equally dividing a whole point cloud data set scanned by a laser radar into some sub-data sets, and carrying out regular division processing on each data point in the sub-data sets; and then based on the rule of dividing the data points in the sub-data sets, respectively carrying out rapid registration processing on the test points in each sub-data set, and verifying and selecting prediction points. According to the invention, the overall registration rate of the prediction point set in the three-dimensional point cloud data set is improved. The rapid selection and registration method for the three-dimensional laser radar point cloud data prediction points has a great reference value for point set registration processing of point cloud data in laser SLAM mapping.

Description

technical field [0001] The invention relates to the technical field of real-time positioning and mapping of three-dimensional laser radar, in particular to a method for quickly selecting and registering prediction points of 3D laser radar point cloud data. [0002] technical background [0003] In the map construction step, in addition to determining the map representation for SLAM (Simultaneous Localization And Mapping), there is another important goal: to control the newly constructed map elements at ideal positions, so as to facilitate the establishment of map elements between two frames of maps The corresponding relationship, which is also a preparation for the pose estimation step. However, the 3D environment is more complex than the 2D environment, and the number of 3D point sets is much larger than that of 2D point sets. Direct expansion will bring great problems. Because the algorithm complexity of the "prediction" step in the map reconstruction method is O(n 3 ), t...

AI Technical Summary

Problems solved by technology

However, such methods have the following problems: first, such inventions usually use traditional machine learning algorithms to identify prediction points, which are less efficient due to the irregular arrangement of prediction points; second, in these studies , they all model the structure of the application scene as a single surface. When the mapping area is too large, the error will be amplified, and the selection of prediction points between different training points will be affected; and thirdly, when When the mapping area is too large, the calculation speed of the algorithm is slow in the face of a huge amount of data, and the mapping speed becomes relatively slow

Images