Resealrch On 3D Reconstruction By Lidar Point Cloud Data Of Indoor Scene

Three-dimensional(3D)reconstruction for indoor scenes,that is 3D modeling for indoor scenes which contain multiple targets,as well as walls and floors.3D modeling is the process that use the 3D topology data of one target to reconstruct the geometric topology of it under the three-dimensional space.The 3D reconstruction for indoor scenes has played an important role in the field of military and civil activities.In particular,it can be used in combination with the Virtual Reality(VR)technology,which is more and more popular,to greatly expand its application value in the field of real estate,3D games,surveying and mapping,as well as tourism.On the one hand it can provide precise information such as the distance,azimuth and topological structure of the indoor scene so as to effectively assist the relevant surveying work.On the other hand,the visualization of the result of 3D reconstruction can greatly facilitate and enrich people's life.Among the sensors used for scanning the indoor scene,lidar is more suitable because of its higher ranging accuracy,stronger anti ?interference ability and larger visual range.At present,the main difficulty of 3D reconstruction contains outlier removal,registration,and 3D reconstruction.We firstly set up the scanning platform with single-line lidar and turntable to obtain the point cloud data of indoor scenes,then we analyzed 3D reconstruction with these data.To achieve a better result of 3D reconstruction for indoor scenes,the following improvements based on the current algorithms have been done:Firstly,aiming at the problem about outliers among the point cloud with uneven distribution,an algorithm based on adaptive clustering was proposed,which improved the adaptability about the neighborhood radius of DBSCAN algorithm,and optimized the search scope of neighborhood.According to the simulation results of indoor scene,the detection rate of outliers was increased by 2.7%,and the efficiency was increased by 9.7%.Secondly,aiming at the problem about registration of multi-view point clouds,given the fact that the registration complexity of ICP is high and the accuracy always depends on the initial distance,an algorithm was proposed,which combined the coarse registration based on PCA with the fine registration based on ICP.According to the simulation results of indoor target,the efficiency has improved by around 26.7%.Finally,aiming at the problem of 3D reconstruction of point cloud,two methods,greedy projection 3D reconstruction and poisson 3D reconstruction,was comparatively analyzed,.What's more,the node function of the octree in Poisson 3D reconstruction was optimized with the use of NURBS basic function.According to the simulation results of indoor target,it preserves the hermeticity and smoothing of Poisson 3D reconstructed,meanwhile improves the detail of the reconstructive surface.