毕业生知识库

基于视觉的三维重建技术一直是计算机领域的重要前沿技术之一，近些年 来已经在很多领域取得了广泛应用，比如医学研究、文物修复、无人驾驶等； 同时三维重建技术也在助力扩展现实地进一步发展。随着扩展现实、虚拟现实 的应用深度和广度日益增加，用户对三维重建的模型提出了可交互可全表面触 碰的高要求。目前的三维重建模型无法满足这一要求，一方面三维模型重建涉 及多项复杂技术，另一方面重建模型如何满足这一要求的技术路线仍不完整。 本文旨在设计一种重建扩展现实三维实体模型的技术路线，并对其中关键技术 进行有针对性的研究，使模型满足扩展现实应用的要求。进一步地，本文自主 搭建了一套完整的扩展现实系统，以验证方案的可行性，同时为扩展现实在医 疗领域的应用与发展提供可行有效的方案。主要内容如下： 对获取物体三维点云的方法进行了研究与实验。在调查并分析主流三维重 建方法的优缺点后，针对本文目标，设计了基于单目相机多视图获取三维点云 信息的方案，验证方案并对关键步骤进行补充实验，为后续模型重建提供数据 基础。 进行了点云处理的设计与实验。本文针对模型问题，创新性的引入点云处 理到整体步骤中，先通过实验探索了常规滤波和配准的效果，并在此基础上又 提出了一套结合组合滤波器和点云配准再融合再配准的方案，很好的解决了从 三维重建点云数据生成扩展现实模型的诸多难点，为后续重建出模型的可交互 性提供了关键性的处理。 进行了表面重建的研究与实验。通过对多种表面重建方法的结果，以及重 建模型和应用模型的区别进行对比分析，基于处理后的点云数据，本文设计了 基于点云的表面重建方法，极大的优化了网格重建效果，减少三角面，同时使 得表面完整，使得重建模型满足应用要求。最后给出了可视化结果并验证了模 型完整的物理特性，证明本文方法的有效性。 为进一步论证本文方法，本文结合医学研究的现实需求，设计并开发了一 套以本文模型为核心的扩展现实战救出血控制医疗培训系统。在微软 Hololens2 上搭建了系统，完成了从基本信息层到设计开发应用层的全部具体实现。并对 系统进行了性能分析，模型应用流畅，系统效果显著。 本文成果为扩展现实模型提供了新的技术方案，同时为包括但不限于医疗 培训领域的扩展现实应用提供了新的可行参考案例。

Vision-based 3D reconstruction technology has been one of the important frontier technologies in the computer field, and has been widely used in many fields in recent years, such as medical research, restoration of cultural relics, unmanned vehicles, etc.; at the same time, 3D reconstruction technology is also helping the further development of extended reality. With the increasing depth and breadth of the application of extended reality and virtual reality, users have put forward high requirements for the 3D reconstructed models that can be interacted with and touched by the whole surface. The current 3D reconstruction model cannot meet this requirement. On the one hand, the 3D model reconstruction involves several complex technologies, and on the other hand, the technical route of how to reconstruct the model to meet this requirement is still incomplete. The purpose of this paper is to design a technical route for reconstructing 3D solid models of extended reality, and to conduct targeted research on the key technologies to make the models meet the requirements of extended reality applications. Further, this paper independently builds a complete extended reality system to verify the feasibility of the scheme, and at the same time provides a feasible and effective solution for the application and development of extended reality in the medical field. The main contents are as follows: The method of acquiring 3D point clouds of objects is studied and experimented. After investigating and analyzing the advantages and disadvantages of mainstream 3D reconstruction methods, a scheme for acquiring 3D point cloud information based on multiple views of a monocular camera is designed to address the objectives of this paper, validate the scheme and conduct additional experiments on key steps to provide a data base for subsequent model reconstruction. The design and experiments of point cloud processing are carried out. In this paper, we innovatively introduce the point cloud processing into the overall steps, firstly, we explore the effect of conventional filtering and alignment through experiments, and on this basis, we propose a set of combined filter and point cloud alignment and then fusion and re-alignment scheme, which well solves many difficulties in generating extended reality models from 3D reconstructed point cloud data, and provides the key processing for the interactivity of the subsequent reconstructed models. The study and experiments of surface reconstruction were conducted. The research and experiments of surface reconstruction are conducted. By comparing and analyzing the results of various surface reconstruction methods and the differences between the reconstructed model and the applied model, based on the processed point cloud data, this paper designs a point cloud-based surface reconstruction method, which greatly optimizes the mesh reconstruction effect, reduces the triangular surface, and makes the surface complete at the same time, so that the reconstructed model meets the application requirements. Finally, the visualization results are given and the complete physical properties of the model are verified to prove the effectiveness of the method in this paper. To further demonstrate the method of this paper, an extended reality combat rescue bleeding control medical training system with the model of this paper as the core is designed and developed in this paper, taking into account the realistic needs of medical research. The system was built on Microsoft Hololens2, and all specific implementations from the basic information layer to the design and development application layer were completed. The performance analysis of the system was also conducted, and the model was applied smoothly with significant system effe