A Research On Semantic Segmentation Method Of 3D Point Cloud For Interior Decoration Environment

The promotion and application of decoration robots is an important means to improve the quality and efficiency of the real estate industry and solve the shortage of human resources.At the same time,3D point cloud is an important data form for obtaining environmental information,which is widely used in robotics,autonomous driving and other fields.Effective understanding of environmental semantic information is the basis for the work of decoration robots,and the study of semantic segmentation methods for3 D point cloud data has important theoretical value and practical significance.Due to the unstructured nature of point cloud data and the high cost of acquisition,the research on traditional point cloud processing technology is limited to a single field,and the research on point cloud deep learning is in the ascendant.The existing methods are not very versatile,difficult to realize the autonomy of the decoration robot,and are not suitable for the dynamically changing decoration environment.Therefore,this paper proposes a point cloud semantic segmentation method oriented to the interior decoration environment,which can improve the data utilization efficiency and cope with the dynamic changes of the decoration environment.The specific work is as follows:1.Obtain environmental point cloud data on the spot,and obtain point cloud data of indoor decoration scenes with complete information and uniform density after preprocessing.Combining the characteristics of the data and based on the excellent algorithms in PCL,a point cloud semantic segmentation method based on the PCL library is proposed.The point cloud semantic segmentation task is divided into two stages: point cloud segmentation and semantic matching,and the advantages of traditional methods that do not require annotated data are fully utilized,which reduces the difficulty of implementing the point cloud semantic segmentation task of indoor decoration scenes.This method is used as an efficient semantic annotation method to make point cloud data sets,and realizes the point cloud data set of indoor decoration scenes from scratch.2.In order to break through the limitations of traditional methods that require manual intervention,a 3D point cloud semantic segmentation method based on deep learning is proposed.The versatility of deep learning helps to realize the autonomy of decoration robots.Among them,Point Net is a widely used point cloud deep learning classic model,and MAML is a model-independent meta-learning algorithm.First,the MAML algorithm based on the Point Net model is used for experimental verification on the indoor point cloud data set S3 DIS.Successfully extended the MAML algorithm from two-dimensional images to three-dimensional point clouds,applied to point cloud semantic tasks,and learned general capabilities through MAML algorithms.Improve the generalization ability of the model to cope with dynamic changes in the scene,and improve the efficiency of data utilization to reduce the model’s dependence on new data.Then a kPoint Net model based on k NN algorithm to extract point cloud neighborhood information is proposed to improve the local feature extraction ability of the Point Net model and improve the semantic segmentation performance of the model.Finally,combining the kPoint Net model and the MAML algorithm,a 3D point cloud semantic segmentation method based on deep learning is proposed.The prior knowledge is learned on the S3 DIS data set,and it can be adapted to the interior decoration environment through a small amount of annotation data.A good segmentation effect is achieved on the scene point cloud data set.By comparing the experimental results of different meta-learning settings,the optimal settings for the practical application of the method are obtained.