Shape Analysis Based On Point Cloud Data

In recent years,with the continuous development of 3D data acquisition technology,the way to obtain 3D data is also showing more and more diversified trends,which makes a good foundation for research based on 3D point cloud data processing.Three-dimensional reconstruction,registration,shape classification and scene segmentation based on 3D point cloud data have become hot research issues for many scholars and experts at home and abroad,and have important applications in submarine military,robot vision and automatic driving.Shape analysis mainly estimates the boundaries of objects,analyzes the shape characteristics of these boundaries,and determines the original objects based on the shape characteristics.Point cloud data is a flexible geometric representation of the shape of an object.For shape analysis,the current point cloud shape analysis process generally performs classification and identification of point cloud data containing shapes or other aspects after filtering and reducing point cloud data.At present,research has been lacking a priori judgment on the shape detection of point cloud data;by a priori judgment on the shape of point cloud data,neglecting invalid operations of point cloud data without shape can avoid unnecessary resource waste in point cloud shape analysis system.At present,the research on the classification of point cloud data shape has experienced the development process from the beginning of multiple perspectives,voxels and the recent point cloud classification directly on point cloud data,however,most of the currently proposed models have more parameters and longer training time.The main research contents of this paper include the following two aspects.First,for the current research,there has been a lack of a priori detection for determining the existence of shape for a given point cloud data,the shape detection method is designed based on Ripley's K function in spatial point analysis.Experiments were carried out on 2D and 3D point cloud data to verify the validity of the shape detection method based on Ripley's K function design for point cloud shape judgment.Secondly,in view of the fact that the current point cloud classification model has more parameters and longer model training,a lightweight point cloud classification model Light Se Dg CNN is proposed.Based on the current latest DGCNN model,the structure of the model is optimized by experiment,and the SENet model is introduced to improve the influence of the model classification ability reduction caused by parameter reduction.Finally,experiments on the Model Net dataset show that Light Se Dg CNN can achieve 89.81% classification accuracy with less model parameters and training time.