A non-equivalent 3D point cloud segmentation method

Abstract

The invention provides a non-equivalent three-dimensional point cloud segmentation method. First, the original point cloud data set is constructed and the training set and the test set are divided, and secondly, the training set data is preprocessed so that the number of points in each point cloud becomes the same specification. , and then design the convolutional neural network model to calculate the distance between each point in the point cloud and the point that is not in the same segmented area, and select the smallest distance value between the two points to construct the distance matrix. By designing a penalty function and acting on The distance matrix, so that the value of the point close to the junction of the segmentation area becomes larger, and the value of the point away from the junction becomes smaller, and the penalty distance matrix is ​​multiplied by the model loss function, so that the segmentation model optimizes the loss during training Calculate, strengthen the error feedback of points at the junction of the segmentation area, improve the accuracy of point cloud segmentation, and finally use the test set to evaluate the segmentation performance of the designed model. The invention can realize rapid segmentation of non-equivalent three-dimensional point clouds, and has high segmentation accuracy.

Description

technical field [0001] The invention relates to the technical field of three-dimensional point cloud segmentation, in particular to a non-equivalent three-dimensional point cloud segmentation method. Background technique [0002] Point cloud segmentation is the process of dividing the point cloud into multiple homogeneous regions. Points in the same region will have the same attributes. There are two main methods: the first method is to use pure mathematical models and geometric reasoning techniques for segmentation , such as region growth or model fitting, etc. This method is simple in principle and easy to program, but the degree of automation is not high, and the segmentation results are not accurate enough; the second method is to use machine learning technology for segmentation, which improves the accuracy of point cloud segmentation. degree, but it takes a long time and may lead to excessive segmentation results. [0003] In recent years, deep learning algorithms repr...

AI Technical Summary

Problems solved by technology

At present, most of the research on point cloud segmentation using deep learning is carried out for the variable and disordered number of point clouds, and does not consider the non-equivalent problem of point cloud in the actual segmentation process, that is, point cloud segmentation Points at the boundary of the region have a greater influence on the segmentation result than points far from the boundary

Therefore, when using deep learning methods for point cloud segmentation, it is necessary to solve the impact of non-equivalent point clouds on segmentation accuracy

Images