3D Modeling For Complex Indoor Scene From Point Clouds Joint With 2D Model

With the development of lidar scanning equipment and mapping technology,point cloud has become one of the main data sources of Indoor 3D modeling.Meanwhile,the application of the indoor 3D model in indoor navigation,building information modeling,and other fields is also increasing.At present,how to reconstruct a complete and accurate indoor model from indoor 3D point cloud data has attracted extensive research interest.Generally speaking,a complete indoor three-dimensional model should include the main parts such as ceiling,wall,floor,door,and window,but not the movable objects such as table and chair.Due to the limitation of laser scanning equipment and the existence of a large number of movable objects(such as tables,chairs,etc.)in the indoor environment,the indoor 3D point cloud collected by laser scanning equipment has the problems of incomplete data and high level of data noise,thus affecting the indoor 3D modeling effect,such as the incomplete point cloud data will affect the wall modeling result,etc.Based on the current research of indoor 3D modeling,this paper studies the following three aspects from the perspective of indoor debris removal,opening area detection,and inconsistent ceiling height modeling:1.In this paper,the quality evaluation of indoor point cloud without reference data is proposed.In view of the difficulty of obtaining reference indoor point cloud,based on the characteristics of indoor point cloud data,this paper evaluates indoor point clouds from two aspects of smoothness and occlusion.2.This paper presents a method to generate two-dimensional model based on an end-to-end network.This data-driven method does not need human interaction,and it can be applied to a variety of complex indoor scenes after training.By effectively using of the information in the indoor scene,it can distinguish the wall,table,chair,and other debris,and locate the opening areas.3.This paper presents a framework of indoor 3D modeling for complex scenes.This framework does not need to extract the plane first and does not need to filter the debris manually.Even in the complex indoor scene covered by many debris,it can also make full use of the indoor scene information to filter the debris and complete 3D modeling.In this paper,we mainly use the data set of S3DIS.We also built a data set for the end-to-end network to generate a 2D model.The experimental results show that the proposed method is effective and feasible.