Research On Semantic Segmentation Algorithm Of 3D Point Cloud Based On Context Awareness

In the field of computer vision research,3D scene understanding has always been an important research direction.With the continuous development of 3D point clouds acquisition equipment,Scene understanding based on 3D point cloud has attracted more and more researchers' attention.Semantic segmentation of 3D point clouds is an important basic problem for 3D scene understanding,which is widely used in actual scenes such as automatic driving,robot navigation,indoor scene analysis and so on.However,due to the sparsity,disorder,and unstructured characteristics of 3D point clouds,as well as the influence of related point clouds acquisition equipment,the problem of 3D semantic segmentation based on 3D point clouds still faces many difficulties and challenges.This paper mainly studies the semantic segmentation task of 3D point clouds,and proposes a semantic segmentation model of 3D point clouds based on local perception.In order to make full use of the global features of point clouds,the global feature extraction module and global attention module are added to the semantic segmentation model of 3D point clouds based on local perception.And then,a 3D point clouds semantic segmentation model based on the combination of global perception and local perception is designed.The specific work of this paper is as follows:(1)Aiming at the deficiency of Point Net++ model in the aspect of local feature extraction and encoding of point clouds,a 3D point cloud semantic segmentation model based on local perception is proposed.In this model,the local perception module is used to extract the local features of point clouds.This module can dynamically adapt to the local structure of point clouds and automatically learn the fine local features of point clouds through the attention mechanism.At the same time,the local perception loss function is used to refine the local features.This model can better learn and perceive the local structure of point clouds and improve the accuracy of semantic segmentation.Compared with Point Net++ model,the proposed algorithm improves the overall accuracy and mean class intersection-over-union of S3 DIS dataset are improved by 5.6% and 4.4%,respectively.(2)Based on context feature fusion,a 3D point cloud semantic segmentation model based on the combination of global perception and local perception is proposed to address the shortcomings of the previous model that does not make full use of the global features of point clouds.The model adopts a global feature extraction module to extract the global information of the whole scene of point clouds.Meanwhile,the global attention module is applied to fuse the global features and local features of point clouds.The model can perceive the local structures of point clouds and the global scene features at the same time,which improves the performance of semantic segmentation.Compared with the semantic segmentation model based on local perception,the overall accuracy and mean class intersection-over-union of this model are improved by 1.8% and 1.6% respectively for S3 DIS dataset.