Indoor Scenes Reconstruction Technique Based On Depth Information

With the development of science technology and the improvement of people’s living standards, people gradually come to a world of three-dimensional reconstruction. Three-dimensional scene reconstruction makes a digital representation of the real world possible, and three-dimensional reconstruction of the interior scenes has great significance. With the application of laser scanning device to obtain depth information of the scene, we can realize the target scene reconstruction. It has broad application prospects in the protection of cultural relics, the virtual military scenes, 3D effects of video entertainment, virtual fitting and other fields. The advent of laser scanning devices and the rapid development of scanning technology make the indoor scene reconstruction techniques rapid development possible.The Kinect depth sensor is used to move before the target scene and scan the scene in formation. And the Open NI interface is used to collect depth information and color inform ation by driving Kinect. Through the coordinate transformation among world coordinate system, physical coordinate system, and camera coordinate system, the depth information of the target scene is converted into the corresponding 3D point cloud model.According to a lot of noise in the depth information and compare of the advantages and shortcomings of the traditional filtering algorithm, this paper presents a filtering algorithm based on trust point. Through margin calculation between mid-value of depth value of multi-frame image and multi-frame depth value, if the result is not more than a given threshold, we can defined it as trust point. Finally we can achieve filtering de-noising by taking the value of the trust point depth value.The traditional registration methods were studied. The traditional ICP algorithm is used on the PCL(point cloud library) platform for point cloud registration. With the sampling of point, the selection strategy of the matching point, the assignment for points, the removal of the problem point, the selection of objective function, and to determine the methods of minimization and so on, point cloud finalized registration.For the 3D reconstruction, we can fuse and correct the different point clouds, and complete the scene reconstruction with PCL visualization related classes.