Research On Related Algorithms Of 3D Reconstruction Based On Kinect

3D Reconstruction refers to the building of digital models that are fit to be represented and processed by computers,allowing it to be analyzed,operated and processed in a computer environment,which is a key technology to represent obje-ctive world in computers.With the development of low-cost depth cameras such as Kinect and Xtion,3D reconstruction based on depth cameras has become a po-pular direction of research,which has broad application in the field of Simultaneo-us localization and mapping(SLAM),Virtual Reality(VR)and Augmented Reality(AR).In this paper,three-dimentional reconstruction research of the scene and rotati ng object based on Kinect depth camera is studied.The main contents are follow s:(1)The structure and working principle of depth camera are introduced.this paper analyzes the current research situation of 3D reconstruction based on depth camera and some of the key techniques and related algorithms are summarized.Advantage,dis-advantage and application fields of these algorithms are analysed.As for 3D image preprocessing,A depth image super-resolution algorithm based on color image texture guidance is proposed(2)Aiming at the problem of the keypoint detection in the process of 3D reconstruc-tion,a Smallest Univalue Segment Assimilating Nucleus(SUSAN)algorithm based on point cloud is proposed.The proposed algorithm has been tested on the surface of point clouds with different scales and types and compared with existing point cloud feature points extraction algorithm(3d-SIFT,3d-Harris,ISS,etc.).This paper measured the algorithm's performance from aspects of efficiency&precision of initial registration.experiments proved that the proposed algorithm has high efficiency and can satisfy the requirement of feature point detection and matching in 3D reconstruction process under the premise of ensuring the initial registration's effect.(3)Aiming at the problem of the feature description in the process of 3D reconstruction,a improved Fast Point Feature Histogram(FPFH)combined with texture information is proposed.The proposed algorithm has been tested on the Kinect data set and the corresponding precision has been count.Experiment verified the effectiveness of the proposed algorithm.(4)A prototype of 3d-pointcloud-processing software.We designed the interface framework based on Point Cloud Library(PCL)and Qt.It implemented plenty of functions such as reading/saving,visualization,feature extraction/description,segmentation and clustering of 3D point cloud.