【作者】 谢文寒；

【导师】 张祖勋； 张剑清；

【作者基本信息】 武汉大学 ， 摄影测量与遥感， 2004， 博士

【摘要】 人类从外界环境获取的信息中,80%来自于视觉。人的眼睛将获取的信息传入大脑,由大脑根据知识和经验,对信息进行处理与识别。随着数字技术的出现与发展,计算机视觉替代了人眼与大脑,实现了对周围环境进行简单识别与理解的功能。计算机视觉的研究目标是使计算机具有通过二维图像认知三维环境信息的能力,这种能力将不仅使机器感知三维环境中物体的几何信息,包括它的形状、位置、姿态、运动等,而且能对它们进行描述、存储,识别与理解。而若通过二维图像感知三维世界,传感器的定标(或标定)成为关键技术之一。近年来,随着数字化技术的不断进步,硬件装置价格的不断下降和计算机性能的不断提高,数码相机也得到了普及,许多计算机视觉系统也都以普通数码相机作为获取数据的主要传感器。 因此,数码相机(摄像机)的标定逐渐成为计算机视觉领域以及摄影测量领域的研究热点之一。 国内外的许多摄影测量及计算机视觉界的专家学者提出多种多样的标定方法。然而,没有哪一种标定方法可以用于所有应用,不同的用途、不同的环境就要采用不同的标定方法。 本文以基于灭点的标定方法为切入点,首先阐述几种传统相机标定方法与自标定技术以及存在的问题,随后提出了几种既不需要控制场同时精度又较高的基于多像灭点的标定方法。这些标定方法操作简易、灵活,可做到“随时随地”标定相机,若将标定结果用于人工规则建筑物的三维重建,该方法还可进行在线标定。 本文涉及的主要研究内容集中在四个方面:(1)直线检测与灭点的确定;(2)基于多像灭点的标定方法;(3)基于点线混合的相机标定方法;(4)基于多平面格网的相机标定方法。 1)直线检测与灭点的确定。灭点是空间平行线在影像上投影的交点。它作为本文所提出的标定方法的重要因素,贯穿文章始末。一般灭点位于影像边界以外,甚至是无穷远处,因此标定前确定灭点的最佳位置至关重要。灭点是直线的交点,本文研究了一种自适应的最小二乘直线模板匹配技术用于高精度的提取影像中的直线信息。由于影像中噪声的存在,使得平行线在影像上未必交于一点,因此,从摄影测量角度出发,研究了一种以影像直线为观测值,建立直线与灭点相关的平差模型确定灭点的方法。该方法既不需要将直线投影到球面坐标,也不需要对大量直线作统计特性分析,当存在大量多余直线观测时,这种平差方法可快速、准确地计算出灭点的最优位置。 2)基于多像灭点的相机标定。从单像灭点标定方法的误差分析入手,以灭点为桥梁,通过灭点的几何特性将影像中直线观测值与相机方位元素直接关联,并对多像联合建立平差模型,以解求未知的相机内外方位元素。同时,非线性畸变模型和几何约束条件也直接纳入到平差标定模型中,多种类型未知参数统一解求,使标定精度大大提高。 3)基于点线混合的相机标定。分析了长焦距相机因视场角小在应用灭点标定时遇到的问题,首次研究了专门针对长焦距、小视场角相机的基于点线混合的标定方法。该方法通过一种无位移、多方位的全景摄影方式扩展长焦距相机的视场角,并用基于点的自标定与基于线的平差模型相结合的方法建立统一模型对相机进行标定。 4)基于平面格网的相机标定。通过分析灭点与灭线的约束特性,利用两个灭点也可进行相机标定。对一个或(多个平行)平面格网或人工建筑物某一立面进行多位移、多角度拍摄,利用多像灭点约束建立标定模型,最后通过矩阵分解得到相机参数矩阵。该方法将三维方向的灭点信息发展到二维方向的灭点信息,且不需要任何物方控制点信息,使灭点标定方法得到进一步推广。更多还原

【Abstract】 Among the information obtained from outside environment, 80% comes from human visual perception. The information is delivered from eyes to human brain, then is analyzed and recognized by human brain in according with his or her knowledge and experience. With the emergence and envelopment of the Digital Technology, Computer Vision, taking the place of Humans’ cerebra and eyes, can realize the simple function of environment recognition and understanding. The aim of computer vision research is to make the computer possess the capability of obtaining 3D environment knowledge from 2D image. This capability refers to not only the objects’ geometrical information perception which includes the information of shape, position, pose and motion etc., but also information describing, storing, recognizing and understanding. Sensor calibration is one of the key techniques in three-dimension environment recognition from two-dimension images. Recently, with the rapid development of digital technique, the continuous descending price of hardware device and the promotional performance of computer, digital camera gains prevalence in practical application. Many systems and software of computer vision utilizes digital camera as their main sensor for data acquirement. Subsequently digital camera calibration gradually becomes the hotspot in the research fields of computer vision and photogrammetry.Many experts and researchers in the fields of computer vision and photogrammetry have already proposed various methods about camera calibration. However, none of them can achieve extensive application. For different purpose and distinct application we have to adopt different calibrating methods.This paper begins from the calibration method based on vanishing points. Firstly several types of traditional calibrating methods and self-calibration methods are investigated, and then the existing drawbacks of corresponding calibration methods are addressed. Aiming at these problems, the new calibrating methods are proposed in this paper, which need no control points and can obtain higher precision based on vanishing points of multi-view. These calibrating methods can be easily operated. When these camera-calibrating methods are applied in three-dimension reconstruction of regular buildings, it can also realize on-line camera calibration.Following aspects about these methods are introduced and focused on in this paper including: (1) line detection and vanishing points calculation; (2) camera calibration based on vanishing points of multi-view; (3) camera calibration based on point-line combination. (4) Camera calibration based on multi-planar grid. 1) Line detection and vanishing points calculation.Vanishing point is the intersection of the lines in image, which are parallel lines before projecting into one image. As the most important factor in the proposed calibration methods, vanishing point are addressed through the whole paper. Generally vanishing points locate at the outside of image frame, even extend into the infinite. Therefore, it is crucial for camera calibration to determine the optimal location of vanishing points. Because vanishing points locate at the intersection of lines, in this paper the technique of Self-adaption Least Square Line Modulate Matching is proposed for extracting line information from images with high accuracy. Because of the existence of image noise, the image lines in parallel usually don’t intersect into one point in practical. Therefore based on image line observation, the adjustment model of correlation between line and vanishing point is constructed to determine the location of vanishing point from the viewpoint of photogrammetry. This method needn’t project lines to sphere coordinates, also needn’t analyze statistic property of lines. When a big amount of redundant lines exist, this method can calculate the optimal location of vanishing points rapidly and accurately.2) Camera calibration based on vanishing points of multi-view.The single-view calibrating method of vanishing points is illustrated in this paper. Based on error analysis, in according with geometry property, line observation is directly related with camera orientation, through adjustment of multi images, a novel calibrating method of vanishing points based on multi-view can be recommended. At the same time, non-linear radial distortion model and geometry constraints are also introduced to this calibrating model for accuracy improvement.3) Camera calibration based on point-line combination.After analyzing the problems of camera calibrating with long lens, and aiming at the camera calibration with long lens and narrow view angle, one calibrating method based on point-line combination is addressed.4) Camera calibration based on multi-planar grid.In according with the geometry property of vanishing points, it can be seen that only two vanishing points cannot be used to calibrate camera. However, based on the constraints between vanishing point and vanishing line, camera calibration can be easily conducted.更多还原