【作者】 席志红；

【导师】 张曙；

【作者基本信息】 哈尔滨工程大学 ， 通信与信息系统， 2006， 博士

【摘要】 全方位视觉感知，又称全景视觉、全视觉，是指一次获得大于半球视场(360°×180°)的三维空间的全部视觉信息，是近几年被广泛关注、热门研究的视觉感知技术。由于其视域开阔，对于民用、军事还有宇航空间领域中的依赖于视觉信息做出决策的各行业都具有非常重要的意义。 由于全方位视觉是一门新兴的技术，目前对于其基础理论的研究还不成熟，应用方面也有许多关键技术有待解决。针对国内外全方位视觉的研究现状，本文开展了全方位视觉技术及应用的研究。首先，对现有的全方位视觉成像方法进行了简单评价，对于利用反射镜的单视点折反射成像模型与约束条件进行了理论推导，从原理上解释了利用曲面反射镜可以获得视场增强效果的物理本质。 针对已有的全景图像柱面展开逐点解算的方法，进行了解算效率的分析，对原算法进行了修正，提出了快速二次逆向还原解算法，在不改变图像还原效果的前提下，减少了解算的次数，提高了解算的速度。 为解决全景图像柱形平展解算应用于开阔场景时，对于任意分布的局部景象存在比较严重的失真问题。本文提出了基于主视点的视窗平面图像还原解算法，使得解算后图像在这一情况下的失真得以改善。实验结果显示，相对于视觉系统主轴而言呈非直立分布的景物，视窗平面内的还原图像优于柱形还原成像的视觉效果，而直立分布景物的还原效果与柱形还原相同，同时还实现了对局部敏感区域的放大观察。并将本文提出的图像还原解算方法应用于实验室内人体运动目标的识别与跟踪实验的显示界面，对采集到的连续视频图像实时进行解算，实现了对视场全局的观察和运动目标的局部监视。在此基础上进一步，又提出了球形还原成像法，给出了解算的具体算法。使得还原解算后的图像对于视场空间内非整体结构零散分布的景物，更容易估计其方位信息，成像的局部效果接近视窗平面还原法。 为将全视觉技术应用于智能移动机器人，论文针对机器人定位与导航问题开展了研究。根据全景视觉的特点，提出了利用单帧全景图像中多路标特更多还原

【Abstract】 The omni-directional vision, in other name is panoramic vision, of which the visual area could be more extensive than a hemisphere (360 °×180° ), acquires all visual information of the three-dimensional space once, is a popular and widely concerned new visual perceiving technology in recent years. Due to its’ extensive visual area, it is getting more important to the fields in which decision is made depending on visual information, such as in civil, military and space exploring fields.Because the omni-directional vision is a new developing technology, its basic theory is still unripe now, and a lot of problems need to be solved too. According to the domestic and overseas current situation of the omni-directional vision, this thesis focused on the study of the omni-directional visual technologies and applications.First part, briefly introduced the existing methods of omni-directional vision imaging, derived theory model of the imaging system with a reflector which has a single view point, explained the physical essence that visual area could be extended by using a reflector with curved surface.Aim at the existing method that revert a panoramic image to a spread cylinder image, efficiency of calculation is analyzed, the original algorithm is revised, a new algorithm named twice quick reverting arithmetic is proposed. The number of calculation is reduced, and speed of reverting is advanced on the premise of not changing the effect of reverted image.In order to solve the problem that serious distortedness are remained in the spread cylinder image of a panorama ,when the reverting method is used in extensive space imaging, and substances are distributing in any direction, this thesis propose a new method named flat window imaging which is based on the main view point of reflector, designs the reverting algorithm. By using this algorithm, the distortedness in reverted image is improved. The results of更多还原