【作者】 易学锋；

【导师】 闫利；

【作者基本信息】 武汉大学 ， 大地测量学与测量工程， 2009， 博士

【摘要】 目前,单反数码相机已用于变形测量,但是,在一般情况下,数码相机的量测精度不如常规的测量仪器,数码相机在变形测量中的应用受到一定的限制。在单反数码相机的成像过程中,许多因素会引起量测误差,其中的一些误差可以采取一定的方式消除。数码相机感光像元的整齐排列,会产生莫尔效应,利用莫尔条纹可以精密测量位移量。为产生彩色,大部分单反数码相机图像传感器的表面都覆盖着一层bayer虑镜,该虑镜使感应红(蓝)色的像元隔行(列)存在,当红(蓝)色标志影像边缘落在非感红(蓝)像元上时,会产生标志位置的感应误差,感应误差在0-0.71个像元,由于每行(列)都有感绿像元,若不考虑噪声,绿色标志基本没有感应误差,白色标志的影像含有红、绿、蓝色,每个像元都能感应,所以没有位置感应误差。彩色标志的位置感应误差不仅与标志影像中红绿蓝色含量之比有关,还与RAW图像转换成黑白图像的算法有关,绿色成份多或红、蓝色含量相当的彩色标志位置感应误差小,由基色相关自适应彩色插值算法和简单平均彩色转黑白算法得到黑白图像定位精度最高。需要注意的是,常用的加权平均彩色转黑白算法不仅丢失了图像中的大量蓝色信息,有时还会影响白色标志的定位精度。图像传感器前面的低通虑镜会使标志的边缘向左和向上扩张一个像元,对于绿色和白色标志而言,它们的影像中心向左上方偏移了0.71个像元,对于红色和蓝色标志,由于存在bayer虑镜,它们的影像中心视不同的情况偏移0-1.41个像元。像元间存在缝隙,当标志边缘落在缝隙处也会产生位置感应误差。图像的gamma变换往往是非线性的,会使图像中的一些灰度层次丢失,影响标志的定位精度。相机的快门引起的震动频率较高,高速快门无法消除高频震动的影响。像元的感光性能与定位精度密切相关,当标志影像的边线位置发生很小的变化都能引起像元感应值较大的波动时,定位精度就高,以此原理可以推导出像元的物理特性与定位误差的关系式,该关系式显示：一般情况下,白色正方形标志定位精度比绿色标志定位精度高；标志的形状与精度密切相关,标志面积越大定位精度越高,正方形标志比圆形标志的定位精度高,圆形比椭圆的定位精度高；标志位置发生微小变化时,定位精度也会发生变化。对单反数码相机误差源的研究不仅使我们更深刻的认识单反相机的测量误差,而且还可以采取如下的方式提高单反数码相机量测的精度：①尽量使用RAW格式图像,在转换成黑白图像时应该使用基色相关自适应彩色插值算法和简单平均彩色转黑白算法,并且不要进行非线性的gamma变换。②选定像控点时必须优先选择白色标志,像控点标志的形状应该是正方形或圆形,避免长条形。③大量的像控点,可以减小上述误差的影响。数码相机在拍摄密集条纹时会产生莫尔条纹。长期以来,人们一直将数码相机的莫尔条纹当作有害条纹。在相机的设计上,尽量消除莫尔条纹。现代的数码相机,由于加装了低通虑镜,莫尔条纹已几乎看不见了。但是,数码相机的莫尔效应仍然存在,在条纹的影像上叠加一定的CCD光栅,可以将莫尔条纹从图像中提取并显现出来。实际上,并不是只有拍摄细条纹才能产生莫尔条纹,拍摄任何宽度的条纹都将产生莫尔条纹。如果拍摄的目标是长光栅形式的标志,将会产生计量莫尔条纹,计量莫尔条纹的灰度变化近似于正弦曲线,当被摄光栅标志移动一个栅距时,莫尔条纹移动一个纹距,相应的正弦曲线移动一个周期。若以莫尔条纹图像像元的原始感应值做为观测值,用正弦曲线函数进行拟合,计算出莫尔条纹移动前后的相位以及相位差,用相位差可求位移量。由于一个周期的正弦曲线长度在影像上要远远大于一个栅距,所以,莫尔条纹有位移放大的特性,当莫尔条纹用于位移测量时,精度将超过常规的方法。莫尔条纹影像中,每行像元都能拟合出两条相位相差180°的正弦曲线,正弦曲线总数一般都在二十个以上。每条正弦曲线都对应一个位移量,位移量的平均值可以大大减弱噪声对位移量测精度的影响。还可以用常规莫尔条纹图像处理法求位移量,但是,拍摄莫尔条纹图案得到的莫尔条纹图像与拍摄光栅标志产生的莫尔条纹图像在频域上有很大的不同,前者主要是低频莫尔条纹信号,后者低频莫尔条纹信号叠加在强烈的高频信号上,用常规滤波方法对后者进行低通滤波将会极大的减弱莫尔条纹正弦曲线的强度,所以,基于前者的莫尔条纹图像处理法不能直接用于后者,必须将后者的莫尔纹提取出来,产生新的没有高频成分的莫尔纹图像,才能用常规的方法处理。实际量测时,光栅标志与图像传感器(CCD光栅)的相对位置是无法达到常规计量莫尔条纹的要求。根据标志光栅和CCD光栅的光强分布函数以及它们的相对位置,可以推导出标志光栅与CCD光栅存在空间夹角时的莫尔条纹方程,若两张图片上的莫尔条纹发生变化,可以根据莫尔条纹方程反演标志光栅与图像传感器相对位置的变化量,当相机的位置固定时,变化量就是光栅标志的空间位移量。用莫尔条纹方程可以证明,当φ<7°时(ω、κ不限),使用正弦曲线拟合的模型误差可以忽略不计,所以,测量时对两光栅空间夹角的要求并不严格。正弦曲线的拟合一般都采用最小二乘平差计算,相位是平差计算时的一个未知数,所以,从协因数阵中可以得到相位的协因数,协因数乘以单位权方差就是相位的中误差。相位的协因数是正弦曲线的角频率ω、初始相位Φ、振幅A以及观测值的函数,通过函数关系可以分析ω、Φ、A、观测值数量对相位精度的影响,以确定最佳观测方案。镜头畸变会使正弦曲线变形,在小范围内,变形量是一定的,所以,在计算两条正弦曲线的相位差时,可以消除镜头畸变的影响。标志的光照不均匀也会使莫尔条纹的正弦曲线产生变形,并严重影响相位差的精度,解决的方法是重新设计拟合函数,在正弦曲线拟合函数中加入二次多项式,用新的拟合函数可以较好的修正低频不均匀光照产生的误差。相机快门引起的高频震动会使图像中各像点产生不均匀位移,影响标志位移测量的精度,所以,拍照时应该使用重型脚架并启动反光镜预升功能。电子快门没有震动,因此,最好采用电子快门的相机。由于莫尔条纹对标志的位移非常敏感,并且在计算时直接使用RAW格式图像,分别在三个基色中求位移量,不需要成像算法,不受bayer虑镜和低通虑镜的影响,也基本不受像元的缝隙的影响。数据处理中的平差计算、平均计算、差分计算既消除了大部分偶然误差(噪声)又消除了部分系统误差,所以,莫尔效应摄影测量法精度很高。多次试验表明,摄影距离在100m(最远试验距离),即使用廉价的低档单反数码相机,在一般的室外环境下,一个像对的位移测量精度也达到了1/50万(位移误差比物距),并且抗干扰能力强,对环境的要求不高,操作过程也不复杂。如果用多个像对计算位移量,精度超过1/100万。更多还原

【Abstract】 SLR Digital Cameras are use to deformation observation now, and in a currently instance the measurement accuracy of SLR Digital Camera is not as good as some routine measuring instruments, thereby there are some restrictions on applications received in deformation measurment field for it. In the imaging process of SLR Digital Cameras, many factors will being caused by measurement errors, some errors are taken to eliminate a certain way, Moore Effect is caused by neatly arranged of its photosensitive pixels, and using Moore Stripes can acquire precision displacement measurement.For producing color, most of the SLR Digital Cameras’s image sensor are covered with a layer of bayer filter, this filter makes red or blue pixels existence of de-interlacing. When red or blue image edge falls on signs of non-sense of red or blue will produce the error sensor location of signs, Sensor error 0～1 pixel, because each row (column) has a sense of the green pixel, if does not considering the noise, there are little green signs sensor error, panchromatic mode images marked with red, green, blue, each pixel can be induced, there is no position sensor error. Color mark sensor position error is not only relative with the content ratio of red and green and blue in logo images, but also with the algorithm for converting RAW images to black and white images, color mark sensor error is small when Multi-ingredient green or red and blue color content is considerable, by color-related adaptive color interpolation algorithm and a simple algorithm for the average black and white color to be black and white images, the highest positioning accuracy. It should be noted that commonly used in the weighted average algorithm for color to black and white not only lost a large number of blue image information, sometimes affect the positioning accuracy of the full color logo. Image sensor in front of a mirror will consider low-pass mark to the left and up the edge of a pixel expansion, the square green signs and panchromatic, their image centers offset to the upper left corner 0.71 pixel, to the red and blue logo, as a result of taking into account the existence of mirror bayer, their image centers depending on the situation of migration 0～1.41 pixel. Pixel gap existed, When the mark falls on the edge of the gap may also have a position sensor error. Gamma images are often non-linear transform, making some of the image gray-level lost, marked impact on positioning accuracy. Camera shutter caused high frequency vibration, high-speed shutter can not eliminate the impact of high-frequency vibration. The photosensitive pixel performance is closely related to positioning accuracy, when images of the line marks the location of the changes occurred in small and pixel sensor can cause a greater fluctuation in value and high positioning accuracy, this principle can be derived from the physical characteristics and the relationship of pixel positioning errors, the relationship between the display:under normal circumstances, square full color logo positioning accuracy higher than the Green Label positioning accuracy; The shape is closely related with the accuracy, marked the higher the positioning accuracy of the greater area, square than circular logo sign high positioning accuracy, elliptical than circular high positioning accuracy; Small changes in location of signs occur, the positioning accuracy will also change.SLR digital camera to study the source of errors allows us not only to a better understanding of the measurement error SLR, but also can take the following ways to improve the accuracy of measurement of single-lens reflex digital cameras:1 To make full use of RAW format images, in black and white into color images should be related to the use of adaptive color interpolation algorithm and a simple algorithm for the average color to black and white, and not to non-linear transform of the gamma.2 Control points selected as the preferred choice to be full color signs, control points as the shape of signs should be square or round,but avoid to strip; 3 As a large number of control points can reduce the impact of the above-mentioned error.There will be moire fringes when taking photos of intensive pinstripes using digital cameras. The moire fringes have been always considerated harmful for a long time and people are trying to clear them up through the design of cameras.For the modern digital camera,the moire fringes almost can not be seen as a result of low-pass filter.However,they still exist and they can be extracted from images through the superposition of some kind of CCD grating.As a matter of fact, it is found that there will be moire fringes no matter how width the stripes are. When taking photos of a mark in the form of long grating,the metrologic moire fringes will be generated,whose changes in gray are similar to a sinusoid.When the mark moves by a grid distance,the moire fringes move by a groove spacing and the corresponding sinusoid will move by a period.Based on this,if the original gray values are taken as observations,then the sinusoid function can be used to fit to these values to calculate both the former and the latter phase for the further calculation of displacement using phase difference.Furthermore,because the length of a sinusoid in one period is much more larger than a grid distance,the moire fringes have the characteristics of displacement amplification.That is exactly why higher accuracy will be gotten when the moire fringes are used in the displacement measurement.In the images of moire fringes,two sinusoids with phase difference of 180 degrees can be fit to the pixels in each line.Considering the total number of sinusoid is generally more than 20,the average of the corresponding displacement could be used to reduced greatly the noise impact on the accuracy of displacement measurement.Besides,the displacement could also be gotten using the normal processing methods of images of moire fringes.However,there is a big difference in the frequency domain between the images by taking photos of moire fringe mark and the images from the raster mark. And the main signals in the former images are low-frequency,while for the latter ones they are formed by the low-frequency signals superimposed on the strong high-frequency signals.When low-pass filter is used to the latter signals through normal filtering methods,the strength of the moire signals will be greatly reduces.So,based on the point that the processing ways used for the former moire fringes can not be used directly to the latter ones, the new moire fringes deprived of high-frequency signal must be extracted before they could be processed using the normal methods.When the actual measurement, grating mark with the image sensor (CCD grating) the relative positions can not be achieved by conventional measurement requirements Moire fringe. Grating and the CCD in accordance with signs of the light intensity grating distribution function, as well as their relative position, signs can be derived from the existence of grating and CCD grating angle when space equation Moire fringe, if the two pictures on the Moire fringe changes, and based on the changing value of Moire fringe under the sign inversion grating equation and the relative position of image sensor, when the location of fixed cameras, change is the grating space marked displacement. Moire fringe equation can be used to prove, whenφ<7°(ω、Kopen), sine curve fitting using the model error can be ignored, so, when measuring the angle between the two grating space requirements do not strictly. Sine curve fitting using least square adjustment is generally calculated, phase is unknown in an adjustment calculation, so, factor from the HS phase array can be co-factor, Association factor multiplied by the unit weight variance is Phase error. Co-factor is the function of the phase sinusoid, including angular frequencyω, the initial phase f，amplitude A, and observations, function can be analyzed throughω、f、A、A、, and the number of observations on the effects of phase-precision, to determine the best film program. Lens distortion will result in sinusoid deformation, in a small area is a certain degree of deformation, so, two sinusoids in the calculation of the phase, can eliminate the effects of lens distortion. Signs of uneven illumination of the Moire pattern will also have a deformed sinusoid, and seriously affect the accuracy of phase difference, the solution is to redesign the fitting function, adding quadratic polynomial in the sine curve fitting function, fitting function with new amendments can be good low-frequency error resulting from uneven illumination. Camera shutter high-frequency vibration caused non-uniform pixel displacement in images, affect the accuracy of displacement measurement marks, so, it Should be the use of heavy camera tripod and mirror pre-start up function. Electronic shutter does not shock, so, using the the camera with electronic shutter is the best choice.Moire pattern of signs as a result of the displacement is very sensitive, directly in the calculation and the use of RAW format images, each of the three primary colors in order to displacement, imaging algorithm does not require, Bayer not consider low-pass mirror and consider the effects of mirror, also basically pixel not be impacted by the the gap, adjustment or average or Difference calculation of data processing computing, the elimination of the majority of accidental error (noise), and the elimination of some systematic errors, so, Moore photographic effect of the high precision measurement. Many experiments show that the Photo from the 100m (the longest distance test), that is, the use of cheap low-end single-lens reflex digital camera, in the general outdoor environment, as the displacement of a measurement accuracy has reached the 1/0.5 million (from the displacement of error than the material), and anti-interference ability is robust, less demanding on the environment, operating process is not complicated. If using more than a piece of images to calculate the displacement, accuracy will surpass 1/1000000.更多还原