基于自然特征的增强现实虚实注册技术研究

发布时间：2018-04-03 19:01

本文选题：增强现实　切入点：特征点提取与匹配　出处：《华中科技大学》2016年硕士论文

【摘要】：虚实注册是增强现实技术需要解决的主要问题,直接决定了增强现实系统在实时性、鲁棒性、运行速度和占用存储空间等方面的性能。目前发展成熟的增强现实应用主要依靠识别标志物来叠加虚拟信息,在运算速度和实时性方面有较好的表现,但使用场景有限。在基于自然特征的增强现实系统中,利用传统的方法难以满足终端应用实时重建的需求。因此如何保证自然场景注册的精度、效率和鲁棒性,成为增强现实应用需要着重考虑的问题。本文提出了实现基于自然特征的增强现实虚实注册技术的新方案,旨在满足实时性和降低内存消耗的需求,为移动增强现实的实现提供思路,并利用本文的方法理论构建了基于自然特征的增强现实原型系统,证明了方案的可行性。传统基于自然特征的增强现实系统对场景图像提取特征点进行匹配,利用匹配点对重建三维点云,本文分析了主流特征点提取和描述符生成算法在速度和存储空间占用方面的限制,提出加速分割测试特征算子(Features From Accelerated Segment Test,FAST)与局部差分二进制(Local Difference Binary,LDB)结合算法对该过程加速,其中LDB算法通过生成二进制特征点描述符大大减少描述符占用的存储空间。系统分为两个阶段,离线重建阶段和在线注册阶段,通过对两个阶段不同处理来降低终端设备的负担,保证系统运行实时性。实验证明该方案在重建结果相似的情况下,特征点提取速度比原来提升约3倍,匹配速度提升4倍,存储空间占用减少66%。在虚实融合方面,系统存在的缺陷是简单叠加了虚拟物体,没有实现虚实遮挡。
[Abstract]:Virtual reality registration is the main problem that needs to be solved in augmented reality technology, which directly determines the performance of augmented reality system in real-time, robustness, running speed and storage space.At present, the mature application of augmented reality mainly depends on the recognition markers to stack virtual information, which has good performance in computing speed and real-time performance, but the use of the scene is limited.In the augmented reality system based on natural features, the traditional method is difficult to meet the needs of real-time reconstruction of terminal applications.Therefore, how to ensure the accuracy, efficiency and robustness of natural scene registration becomes an important problem to be considered in augmented reality applications.In this paper, a new scheme for realizing augmented reality virtual reality registration based on natural features is proposed, which aims to meet the requirements of real-time and reduce memory consumption, and to provide ideas for the realization of mobile augmented reality.An augmented reality prototype system based on natural features is constructed by using the method theory in this paper, and the feasibility of the scheme is proved.The traditional augmented reality system based on natural features matches feature points extracted from scene images and uses matching points to reconstruct 3D point clouds.In this paper, the limitations of mainstream feature point extraction and descriptor generation algorithms in terms of speed and storage space are analyzed. An accelerated test feature operator, Features From Accelerated Segment Test (FAST) and a local differential binary binary Local Difference binary LDBs (LDBs) are proposed to accelerate the process.The LDB algorithm reduces the storage space greatly by generating binary feature point descriptors.The system is divided into two stages: off-line reconstruction stage and online registration stage. The two stages are processed differently to reduce the burden of terminal equipment and ensure the real-time operation of the system.The experimental results show that the feature point extraction speed is about 3 times faster than the original one, the matching speed is 4 times higher, and the storage space consumption is reduced by 66 times when the reconstruction results are similar.In the aspect of virtual reality fusion, the defect of the system is that the virtual object is simply superimposed, and the virtual reality occlusion is not realized.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.9

【参考文献】