Design And Implementation Of Point Cloud Semantic Segmentation Network Point At 3D Point Cloud Data

With the development of semantic segmentation network,how to directly deal with the 3D point cloud data has become a hotspot.Point cloud is a direct and simple representation of three-dimensional objects.As a collection of coordinates of many three-dimensional points,point cloud can reflect the geometric information of the scene or object.In recent years,there are more and more ways to get point cloud data.Starting from deep learning,this paper looking at those semantic segmentation network which is using 3D point cloud data directly,and design a new semantic segmentation network based on deep learning by analyzing the existing semantic segmentation network.The main research work of this paper is as follows:First of all,this paper designed a new semantic segmentation network,which uses deep learning technology to segment 3D point cloud data.In the architecture design of the network,the commonly used Encoder-Decoder architecture is used,and the whole network is divided into four parts: input layer,convolution layer,upper sampling layer and full connection layer.At the same time,the semantic segmentation network is optimized according to the disorder of 3D point cloud data.Secondly,in order to solve the problem of insufficient local feature information extraction when directly processing point cloud,this paper designs a new local feature extraction module which extracts the density features of points.In this module,the density of points is realized by kernel density estimation.The higher the density is,the lower the local weight is,and the input features are transformed into density features.At the same time,the whole process of local feature extraction is divided into five steps: key point sampling,local region construction,density feature calculation,feature fusion and feature extraction.What's more,to experiment and test the designed semantic segmentation network,this paper evaluate and compare the proposed the designed semantic segmentation network by using the current point cloud dataset and evaluation methods.The experimental results show that the proposed semantic segmentation network can achieve better results.