Research On 3D Laser Point Cloud Semantic Segmentation Algorithm Based On Deep Learning

Because 3D point cloud data has the advantages of high density,high precision,initiative,and rich spatial information,it is widely used in areas that require high scene interpretation requirements such as artificial intelligence,autonomous driving,and smart cities.Semantic segmentation is the foundation of many applications.With the deepening and maturity of deep learning research,its application in 3D point cloud data processing has received extensive attention,especially in the application of point cloud semantic segmentation.This paper conducts in-depth research on the problems of insufficient utilization of point cloud features and low accuracy in point cloud semantic segmentation.The main contents of the research are as follows:(1)Since the point cloud data collected by the 3D laser scanner at the ground station decreases continuously with increasing distance,in order to avoid the problem of inconsistent feature scales due to the point cloud density change in the K-nearest neighbor method,a spatial index based on spherical neighborhoods was studied,according to the combination of different neighborhood radii to effectively extract the characteristics of point cloud data at multiple scales,including curvature,roughness,perpendicularity,linearity and other features.(2)The construction of a point cloud semantic segmentation neural network model with multiple features is discussed.The study selects BP neural network,convolutional neural network and long-term short-term memory neural network for experiments to determine the basic types of deep neural network models.For the overfitting problem,Dropout,Batch Normalization and other methods are used to optimize,and then further comparison the structure of the model,including the number of layers,the number of neurons in each layer,the activation function,the loss function,the optimizer and the classifier,were determined by experiments to form a complete point cloud semantic segmentation algorithm.(3)In order to verify the feasibility and accuracy of the algorithm proposed in this paper,semantic segmentation using Semantic-3D open source data set is compared with point cloud semantic segmentation algorithms such as Point Net.The overall accuracy rate and average intersection ratio are improved,reaching 86.6% and 55.0% respectively.In order to further verify the generalization ability of the proposed algorithm,the point cloud data of some scenes of Zhengzhou University were obtained using Riegl VZ2000 i three-dimensional laser scanner,and relevant pre-processing was performed to meet the required standards for the experiment.The proposed algorithm was applied to the data set.Semantic segmentation has also achieved good results.Experimental results show that the proposed algorithm can make full use of point cloud features and improve the accuracy of semantic segmentation.