Research On Data Processing And Semantic Segmentation Algorithm Of Three-dimensional Point Cloud In Outdoor Scene

The point cloud data obtained by 3D laser scanning is the 3D data set of the real world scene in the computer,which contains a lot of semantic information.The semantic segmentation of three-dimensional point cloud has been widely concerned and played an important role in many fields.The high complexity of outdoor scene and the noise of the point cloud obtained by scanner make the semantic segmentation a very difficult challenge.In the real world,many interference factors are often included.In order to ensure the accuracy of the semantic segmentation of the 3D point cloud,the pre-processing algorithm is of great importance.Since PointNet only extracts the features of each point while the local features of the point cloud are not taken into account,a new method is needed to extract the local features of point cloud,so as to enrich the feature information of points in semantic segmentation.In this paper,the data acquisition,scene filtering,scene understanding and other core issues required by semantic segmentation of threedimensional point cloud of outdoor scenes are studied.The main contents are as follows:1.Point cloud data filtering algorithm based on the vector difference in multiple scales : in this algorithm,normal vector difference is obtained by multi-scale neighborhood in a reasonable range,so point cloud can be divided into sharp area,which contains more information,and smooth area with less information.Different filtering algorithm is adopted to the two areas respectively.As for the noise interference that will appear in the cloud of outdoor scenic spot,this algorithm can effectively suppress the noise.2.Improved RANSAC ground segmentation algorithm : the algorithm uses multi-scale normal vector diference to extract the ground points in the smooth area of the scene.The algorithm can realize the complete segmentation of ground points and effectively ensure that the details of objects in contact with the ground will not be lost.3.A point cloud outdoor scene semantic segmentation algorithm applying spatial partitioning to PointNet: to solve the problem of slow feature extraction and high resource consumption in existing point cloud semantic segmentation algorithms,a new algorithm is proposed in this paper that can quickly obtain local features of point cloud.This algorithm extracts the global feature of point cloud scene based on PointNet network,and divides the scene into multiple spatial regions to extract regional feature.The algorithm is trained and tested based on the public 3D point cloud data set.Compared with the existing algorithms,the average is improved by about 4%.4.The scenario experiment under the environment of ROS: processes including scene scanning,data acquisition,data filtering and semantic segmentation are taken to verify the effectiveness of the algorithm applying in simulated outdoor scene.Based on the Unity3 D engine,the algorithm of recognition results with scenario application is designed into a 3D scene reconstruction system.