面向增强现实的纹理生成与三维物体跟踪研究

发布时间：2018-06-07 01:15

本文选题：纹理映射 + 纹理修复　；参考：《浙江大学》2017年硕士论文

【摘要】：随着智能手机、平板电脑、头戴显示设备的飞速发展,增强现实技术得到了人们的广泛关注。增强现实具有虚实融合和三维注册等特点。因此,虚拟物体的真实感呈现和相机位姿的准确实时恢复是其中的两个关键问题。围绕着增强现实中的这两个关键问题,本文深入研究了三维物体的真实感纹理生成和三维物体跟踪注册,分别提出了一个半自动的高质量纹理生成方法和一个三维物体跟踪方法,有效提高了增强现实应用的用户体验。总体而言,本文的主要贡献如下:1)对于真实感纹理生成而言,它需要解决的问题是如何产生更加逼真的纹理信息,然而为了保证纹理表面的光测度一致性,目前主流的全自动纹理生成方法都会不可避免地产生纹理失真的现象。为此,本文提出了一套半自动的高质量纹理生成方法,该方法包含了自动纹理映射和交互纹理修复两个部分,通过对现有纹理映射算法的优化和加入人工的后处理流程从而可以生成更具有真实感的纹理信息。2)对于三维物体跟踪而言,它需要解决的问题是如何稳定地求解相机位姿,目前主流的跟踪算法都是基于纹理信息进行特征跟踪,然而现实生活中部分物体本身没有太多的特征,甚至表面还会存在高反光,目前的特征跟踪方法无法很好地处理这些问题。为此,本文提出了一个针对缺乏纹理信息的三维物体跟踪算法,并将其与现有的SLAM系统进行了融合,从而实现了对于陌生场景中已知三维物体的稳定跟踪。实验和对比结果表明,本文提出的半自动纹理生成方法可以有效提升纹理映射的质量,同时所提出的融合跟踪算法能够很好地对三维物体进行跟踪定位,可以满足增强现实的应用需求。
[Abstract]:With the rapid development of smart phones, tablets and headwear display devices, augmented reality technology has received wide attention. Augmented reality has the characteristics of virtual reality fusion and three-dimensional registration. Therefore, real-time rendering of virtual objects and accurate real-time restoration of camera pose are two key problems. Focusing on these two key problems in augmented reality, this paper studies the realistic texture generation of 3D objects and the registration of 3D objects. A semi-automatic high quality texture generation method and a 3D object tracking method are proposed, which can effectively improve the user experience in augmented reality applications. In general, the main contribution of this paper is as follows: (1) for realistic texture generation, the problem it needs to solve is how to produce more realistic texture information. However, in order to ensure the consistency of light measurement on texture surface, At present, the main automatic texture generation methods will inevitably produce the phenomenon of texture distortion. In this paper, a semi-automatic method of high quality texture generation is proposed, which includes two parts: automatic texture mapping and interactive texture repair. By optimizing the existing texture mapping algorithms and adding the artificial post-processing process to generate more realistic texture information. 2) for 3D object tracking, the problem it needs to solve is how to solve the camera pose stably. At present, the mainstream tracking algorithms are based on texture information. However, in real life, some objects do not have too many features, and even the surface will have high reflectance. The current feature tracking methods can not deal with these problems well. Therefore, this paper proposes a 3D object tracking algorithm for the lack of texture information, and combines it with the existing SLAM system to achieve the stable tracking of known 3D objects in unfamiliar scenes. The experimental and comparative results show that the proposed semi-automatic texture generation method can effectively improve the quality of texture mapping, and the fusion tracking algorithm proposed in this paper can track and locate 3D objects well. It can meet the application requirements of augmented reality.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】