CASIA OpenIR  > 毕业生  > 硕士学位论文
基于SLAM的增强现实跟踪注册技术研究
靳杰1
Subtype工程硕士
Thesis Advisor蒋永实
2018-05-25
Degree Grantor中国科学院大学
Place of Conferral北京
Keyword即时定位与地图构建 增强现实 跟踪注册 视觉-惯性测量 语义地图
Other Abstract

       增强现实通过给真实世界附加虚拟数字信息提供良好的人机交互,在科普教育、互动娱乐、文化旅游、医疗卫生、工业维修等领域具有广泛的应用前景。跟踪注册技术是增强现实的关键技术之一,决定了增强现实系统的性能,直接影响用户的体验。针对当前面向未知环境的增强现实跟踪注册技术存在的跟踪鲁棒性差、精度低,缺乏环境的语义感知等导致虚实融合视图真实感不强的问题,本文通过深入分析即时定位与地图构建(Simultaneous Localization and Mapping, SLAM)对跟踪注册性能改变的机制,构建了解决未知环境下增强现实跟踪注册技术的方法体系。本文的主要工作如下:

  1. 针对纯视觉方法容易出现跟踪丢失的情况,提出了一种改进的视觉-惯性数据融合的SLAM方法。首先,针对惯性传感器利用离线标定结合在线标定技术,使得惯性传感器的定位精度提高了10%;其次,对视觉图像特征进行统计计数并采用点、线特征结合的方式,优化了视觉信息的使用;最后,在基于优化的框架下使用松耦合的方式融合视觉-惯性数据,系统的鲁棒性得到了提升,在PennyCOSYVIO数据集上进行了实验验证,系统在三个方向上的平均绝对误差在45cm以内,达到了业界先进水平,部分指标优于现有的成熟系统。

  2. 针对缺乏环境的高层次语义感知信息导致虚实融合真实感不强的情况,提出了一种新的三维语义地图构建及应用方法。首先,通过语义检测器获取语义线、语义面进行点采样及曲线拟合构建出三维语义地图;其次,定义了点到点、点到线、线到线的重投影误差,实现了基于语义地图的定位算法,将系统轨迹在三个方向上的平均绝对误差限制在20cm以内,提升了系统的定位精度。最后,基于三维语义地图的增强现实应用实验,实现了基于语义物体的确定位置的虚实融合,提升了系统的真实感。

  3. 在上述方法基础上,设计并实现了一个面向未知环境的移动增强现实原型系统。该系统能够在未知环境下跟踪注册,通过语义分类器对环境中的物体分类,在真实物体表面叠加三维语义标签,实现了具有较好真实感的虚实融合效果,提供良好的人机交互体验,为进一步扩大增强现实技术的应用领域和范围奠定了坚实的基础。

;

    By adding virtual digital information to the real world, augmented reality improves the experience of human-computer interaction, which reveals a bright prospect in popular science education, interactive entertainment, cultural tourism, medical service and industrial maintenance. Tracking and registration are key technologies of Augmented Reality, which are the bases of system performance and user experience. In sight of the low robustness, low accuracy and the lack of semantic perception in recently augmented reality systems for unknown environments, which cause the poor sense of reality, this thesis builds a methodology that solves the tracking and registration problem for unknown environments by analyzing the internal mechanism of SLAM-based tracking and registration. Main contributes of this thesis are as follows:

  1.  For the reality problem of tracking lost of pure visual method, an improved visual and inertial data fusion method is proposed. Firstly, the combination of offline calibration and online calibration improves the accuracy of inertial measurement unit by 10%. Secondly, feature counter and combination of point and line features makes deep use of visual data. Finally, inertial data and visual data are fused in a loosely-coupled manner using optimization to improve the robustness. Experimental results on PennyCOSYVIO dataset show the improvement of mean absolute accuracy, which are below 45 cm in all three dimensions and are superior to the state of art in some measure.

  2. Aiming at adding high-level semantic perception to get realistic experience, a semantic map building and applicating method is proposed. The proposed method first builds the map using semantic landmarks, and then defines different reprojection errors for different semantic landmarks. Next, a localization method based on semantic map is implemented, which improves the accuracy of the system to less than 20 cm. Finally, the preferable sense of reality for augmented reality application is achieved by setting the exact positions of virtual objects based on semantic map.

  3. A prototype system has been designed and implemented, which integrates above methods with several modules and a friendly graphical user interface. The system runs in unknown environments with SLAM-based tracking and registration, and adds virtual 3D semantic labels on the top of the real objects, which demonstrates preferable sense of reality and provides good human-computer interaction experience and builds a strong foundation for expanding application field and range of augmented reality.

Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/20953
Collection毕业生_硕士学位论文
Affiliation1.中国科学院大学
2.中国科学院自动化研究所
Recommended Citation
GB/T 7714
靳杰. 基于SLAM的增强现实跟踪注册技术研究[D]. 北京. 中国科学院大学,2018.
Files in This Item:
File Name/Size DocType Version Access License
2015E8014661092.pdf(12634KB)学位论文 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[靳杰]'s Articles
Baidu academic
Similar articles in Baidu academic
[靳杰]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[靳杰]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.