Partition method for orderly extracting point cloud on ground

Abstract

The invention discloses a partition method for orderly extracting point cloud on ground. The method comprises the following steps: (1) a depth sensor progressively scanning scenes to be detected, setting the number of columns j to be equal to zero in a first column; (2) obtaining the part, belonging to the ground, before a cutoff point i and a local corresponding coordinate of a column point in the depth sensor depth information in the column according to the angle change relationship between adjacent points under a coordinate system of the depth sensor; (3) under the coordinate system of the depth sensor and a local coordinate system, sequentially judging whether a point after the cutoff point i belongs to the ground; (4) judging whether each column of depth sensor depth information in the scenes to be detected is processed, if the information is processed, point cloud ground part extraction is finished, and conducting a step (5), and if the information is not processed, enabling j be equal to j-1, and returning to the step (2); (5) conducting cluster partition based on distances on non-ground point cloud, and separating objects in the scenes. The method is used for partitioning point cloud on multiple road conditions, and enables the objects in the scenes to be separated from each other.

Description

technical field [0001] The invention relates to the field of point cloud segmentation, in particular to a point cloud segmentation method for orderly extracting ground. Background technique [0002] The use of depth sensors and position and attitude sensors to obtain 3D point cloud information of surrounding scenes based on fixed stations or mobile platforms has developed rapidly in recent years due to its high reconstruction accuracy and good reconstruction perspective. Since the scanning scene involves various types of objects, such as buildings, ground, and objects on the ground such as trees, pedestrians, and vehicles, it is necessary to separate the above-mentioned parts from each other by point cloud segmentation before point cloud modeling. The previous method of processing point cloud segmentation considers the data collected by the depth sensor as a large number of disordered discrete point clouds, while ignoring the useful information that the depth sensor scans co...

AI Technical Summary

Problems solved by technology

But for outdoor scenes, the ground must be extracted before the clustering method can be applied for clustering and segmentation, otherwise the existence of the ground will make all other objects on the ground connected together, and the segmentation cannot be completed

The current ground extraction method uses the Random Sampling Consensus Algorithm (RANSAC) to directly obtain the ground as the largest plane in the current scene, but this ignores the undulations of the ground and the inevitable slope ground conditions, which also affects the following The effect of clustering and segmentation on this basis is to separate the objects on the ground by clustering the distance between each other.

Images