【作者】 陈明；

【导师】 陈一民；

【作者基本信息】 上海大学 ， 计算机应用技术， 2010， 博士

【摘要】 增强现实技术(Augmented Reality,以下简称AR)是随着虚拟现实(Virtual reality,以下简称VR)技术发展起来的一种新技术。通过对VR进行建模,构成虚拟景物,并融入真实场景或物体,从而呈现给用户一个虚拟对象与真实环境融为一体的增强现实环境,扩展了虚拟现实应用。虽然目前AR技术已应用到娱乐、训练、医疗和教育等各个方面,但目前AR的进一步应用还是受到虚实交互相关的关键技术研究的限制。虚实交互技术指的是在AR应用中,虚拟物体与真实场景、物体或交互工具间的交互技术。据笔者认为,其中,主要关键问题是对象的注册跟踪与识别技术、虚实碰撞检测技术与新型人机交互技术理论与应用研究等。由于上述问题皆存在数据维度、精度与实时性要求高等问题,采用传统算法已被证明无法解决上述问题,因此本文引入多种启发式智能算法,如元胞自动机(Cellular Automata,以下简称CA)、复杂网络(Complex Network)、蚁群优化(Ant Colony Optimization,以下简称ACO)与人工免疫系统(Artificial Immune System,以下简称AIS)等,进行AR虚实交互应用的深入研究,以实现数据降维、提高精度与实时性,解决AR虚实交互中的若干关键问题,取得了如下的研究成果:(1)本文创新性地提出利用CA算法来改进基于标记的注册识别应用,同时提出并实现了一个高效的基于自然特征的注册算法(SIFT K-means Contour Tracking,以下简称SIFT-KCT)。基于CA的标记识别算法采用CA模型取代传统的标记识别方法,设计并改进了相关CA规则,对AR标记进行识别与跟踪注册,并与传统方法进行对比分析。另外,针对无标记应用的快速发展,本文利用SIFT方法、活动轮廓方法与聚类方法结合,进行了基于自然特征的对象识别、注册与跟踪注册应用研究,实现了基于自然特征的SIFT-KCT注册跟踪算法。(2)本文提出利用AIS算法检测AR应用中是否有虚拟与实际物体在交互过程发生碰撞;提出了AR碰撞后虚拟物体响应变形的体素CA变形模型(Voxel Deformation Cellular Automata,以下简称VDCA)。AR碰撞检测技术包括在AR应用中检测虚实碰撞事件的发生并在碰撞发生后进行快速碰撞响应。无论是检测碰撞的发生还是对于碰撞的响应,都要求实时处理,对于碰撞响应,更要求有较快的实时渲染算法。由于AIS算法具有分类精确,对于原始数据维数与其它要求不敏感等特点,因此,本文对于碰撞检测中的快速高维数据处理的难题,创新地引入了启发式的算法AIS,建立调整相关AIS CD的实时处理模型,很好地解决了AR应用中虚拟与实际物体交互中的碰撞检测发生的检测问题,并进行了仿真分析。另针,针对AR碰撞检测后的虚拟物体变形响应问题,本文提出了相关的快速体素CA变形模型,设计了变形模拟的相关变形响应模型,并做了仿真分析与应用实践。(3)本文提出了利用蚁群优化等技术改进AR手形识别,并在AR应用中首先引入了手掌与掌纹识别技术。虽然目前新型人机交互方式不断出现,如手形与手势交互,眼动交互等,但这些新的人机交互方式普遍存在理论研究不足,实时应用困难等问题。针对现有交互方式的不足,特别是手形识别技术在AR应用中的不足,本文提出了综合利用蚁群优化等技术进行手形识别交互的处理,并在AR应用中创新地引入了手掌与掌纹识别技术,实现了个性化的AR应用。(4)本文提出了一种基于凝视交互轨迹网络的复杂网络仿真模型(Eye-Gaze Tracking Network,以下简称EGTN),提出了一种基于EGTN的AR增强显示方法。EGTN模型首次分析了凝视跟踪交互形成的轨迹网络所具有的复杂网络特性,并进行了相关模拟的仿真分析;另外,针对EGTN应用,创新地提出了一种基于EGTN模型应用的增强显示方法,实现了基于眼动跟踪分类的增强现实显示过程。实验表明,本文提出的基于CA的标记识别算法与基于自然特征跟踪的SIFT-KCT算法精度与实时性高,能够满足AR跟踪注册要求;基于AIS的AR虚实碰撞检测算法与快速变形响应模型VDCA能够实时检测碰撞产生与响应碰撞变形,满足了AR环境下对碰撞检测的要求;基于ACO的AR的手形识别算法实时性与精度皆有了很大改进;基于复杂网络模型模拟的凝视轨迹网络EGTN模型与实际情况匹配,其AR增强显示应用算法鲁棒性强,能够实现一定的AR景深显示效果。较好地解决了AR虚实交互的一些关键问题。更多还原

【Abstract】 Augmented Reality (AR) technology is a novel technology with the advance of Virtual Reality (VR) technology. AR presents a seamless augmented reality environment to users and expands VR application with virtual object and reality environment from modeling VR objects and generating virtual scene and fusing reality scene or objects.Although AR technology has been applied to entertainment, training, medicine, education etc., the further application of AR is limited to the key technology of virtual and real human-computer interaction (V&R HCI). V&R HCI is a kind of interaction technology, which is about the interaction between virtual objects and real scenes, or some objects and interaction tools, in AR application. The main problems are object registering, tracking and recognition technologies, virtual and real collision detection technology, novel HCI theories and applications and so on. The above key technical problems are all multidimensional and accurate and real-time. Solution of these problems with traditional algorithms has been proved infeasibility. So we import some heuristic intelligence algorithms such as Cellular Automata (CA), Complex Network (CN), and Ant Colony Optimization (ACO), etc. to solve those problems. The specific researches are listed as follow:(1)This paper proposes an improved marker recognition algorithm based on CA model and a SIFT K-means Contour Tracking (SIFT-KCT) registering algorithm based on natural features. The CA marker recognition algorithm replaces traditional marker recognition algorithms with CA model. It improves the CA rules and uses it to register and recognize and track marker in AR application. This paper analyzes the marker recognition and SIFT-KCT algorithm and compares it with traditional algorithms. Besides, according to the development of markerless application, the SIFT-KCT algorithm based on nature features is proposed which combines SIFT operator with active contour model and clustering method.(2)An AIS collision detection (AIS CD) algorithm and a Voxel Deformation Cellular Automata (VDCA) model are proposed in this paper. AR collision detection (AR CD) is a technology about detecting the virtual and real collision events which have happened and responding the event afterwards. It is necessary to meet real-time feature for AR CD. For responding process, the fast render algorithm is more necessary. This paper introduces and designs an AIS CD model to solve the problem which is detecting the happened collision event. Because AIS algorithm has accurate classify ability and insensitivity for the dimensions of initial data sets. We create and adjust the AIS model for solving and completing simulation. And a VDCA model is proposed to solve the virtual object deformation problem after CD. The model is designed and implemented for deformation simulation.(3)An improved hand shape recognition algorithm is proposed based on ACO, and the palm print recognition technology is applied to AR firstly. With the development of HCI, the novel HCI methods occur with an increasing frequency, such as hand shape and gesture and so on. But there are some shortcomings about these technologies, such as inadequate theoretical research, real-time problem, etc. This paper proposed a method which uses multi-model HCI method to solve AR natural interaction problem especially hand shape in AR. The method uses ACO model to recognize hand shape and uses palm-print recognition in AR firstly, and in addition brings personality in AR application.(4)This paper creates a simulating eye gazing tracking network (EGTN) model for novel gazing control technology, and proposes an augmented display method for depth of field effect. And we simulate and analyze the EGTN model with complex network conception. The EGTN and Ferns method are applied in the augmented display method to implement the display process.Experiments demonstrate the marker recognition algorithm based on CA and SIFT-KCT algorithm based on natural features presented by this paper have good features of real-time and accuracy and achieve the demands of AR tracking and registering. The AIS CD and VDCA model can detect the collision in real-time and implement the deformation which happens in collision event and meet the demands of collision detection in AR environment. The hand shape recognition algorithm based on ACO has great improvements in accuracy and real-time feature for AR application. The EGTN based on complex network models successfully simulates the factual network features of eye gazing tracking network, the EGTN augmented display method is robust and meets the depth of field requirements in AR application. Some key problems of AR virtual and real interaction are solved by this paper.更多还原