Automatic reconstruction method for large scene based on multi-site point cloud stitching

At present, the three-dimensional (3D) reconstruction system is based on hand-held camera which causes certain problems, such as needing for a large number of human-computer interactions, demanding on high quality image data and insufficient accuracy of 3D reconstruction. Aiming at these problems, a fully automatic reconstruction of large scene based on multi-site point cloud stitching is presented in this paper. The proposed method uses the Kinect sensor for image acquisition. Since, there are multiple sites in the room, each site of image data is processed using a separate model in order to get good 3D point cloud data. Then, these sites are used to constitute a local area network, and the method of bundle block adjustment is employed to stitch each site point cloud data. The proposed method achieves a high-degree automation and provides a high-precision 3D reconstruction which has two main advantages: (i) the reconstruction process is fully automatic, without any human-computer interaction; (ii) the automatic reconstruction is robust. Experimental results show that proposed automatic reconstruction method is convenient and practical, and can provide better 3D reconstruction model than commonly used methods. Moreover, it can be applied to virtual reality shopping malls, Virtual Reality (VR) and other fields.