单幅近景图像空间信息提取及误差分析

发布时间：2018-12-17 18:58

【摘要】：随着新一代信息技术的发展及图像采集设备的普及,图像数据的数量与日俱增,其中不乏大量与空间信息有关的图像,有待深入利用,以扩展空间信息来源。众所周知,图像是对客观世界最形象、最真实的记录与表达,其中不仅蕴含了丰富的光谱、纹理、几何形状等表层信息,而且还蕴含了物体的尺寸、深度变化等深层次的空间信息。 图像空间信息提取,在图像测量、目标识别、机器人导航、应急救援等领域具有重要的应用价值。另外,在某些仅存在有限图像资料的场合,如历史场景、事故现场等,基于单幅图像的空间信息提取成为唯一的方法。在此背景下,本文以单幅未标定图像为研究对象,实现未知距离的几何量测,恢复图像所含场景的深度信息,评价几何量测和深度计算的精度。 针对几何量测,本文研究了基于交比原理的单幅图像几何量测方法,并对该量测结果精度进行评定；分析了误差来源,建立了误差预测与修正模型。主要研究内容和成果如下： (1)实现了基于交比的单幅图像跨平面距离量测。研究了基于交比的单幅图像二维度量信息提取方法,将共线距离量测算法模型和共面距离量测算法模型推广至跨平面距离量测算法模型,减少了已知条件的数量。 (2)分析了单幅图像几何量测的误差来源,并提出了相应误差修正方法。本文从拍摄相机镜头畸变、灭点计算及其误差传播、像素拾取偏差、已知条件数量、已知条件分布等方面分析了几何量测误差的来源和影响机理。研究发现：1)量测误差与待测线段到已知矩形中心的距离具有强相关性；2)当已知条件数量超过三个时,采用三组量测结果的均值可显著提高量测精度；3)量测误差的大小与已知条件的分布没有显著联系。 (3)建立了几何量测误差预测和修正模型,实现了几何量测误差定量描述。本文依据量测误差与距离(待测线段到已知矩形中心)的强相关性,构建了单幅图像几何量测误差预测与修正模型,提高了量测结果的整体精度,实现了单像几何量测误差定量分析。 图像量测仅仅恢复了图像的度量信息,图像深度也是图像空间信息不可或缺的组成部分,同时图像量测和深度计算也是互为条件和互为补充的。因此,本文完善了现有深度计算方法,实现了绝对深度计算,评价了深度计算结果的精度,提出了改善方法。主要研究内容和成果如下： (1)完善了深度计算模型,实现了图像连续深度半自动化估计。本文完善了基于针孔成像模型的地面点深度计算模型,根据图像中地物分割结果和空间位置关系,分区域实施了深度计算及估计,实现了单幅图像连续深度半自动化估计。 (2)提出了基于深度差的相机高度恢复方法,改善了深度计算结果。本文分析了深度计算模型的误差影响因素,发现相机高度误差是深度计算精度的主要影响因素之一,基于该结论本文提出“基于地面点已知深度差的相机高度恢复方法”,可显著提高深度估计结果的可靠性。
[Abstract]:With the development of new-generation information technology and the popularization of image-collecting equipment, the number of image data is increasing, and there are lots of image related to spatial information, which is to be used in-depth to extend the spatial information source. It is well known that the image is the most image and the most true record and expression of the objective world, which not only contains abundant spectrum, texture, geometric shape and other surface information, but also contains the deep space information such as the size and the depth of the object. The image space information extraction has important application price in the fields of image measurement, target recognition, robot navigation, emergency rescue, etc. a value. In addition, in some situations where only limited image data is present, such as a history scene, an accident scene, etc., spatial information extraction based on a single image becomes the only party. In this background, in this paper, a single uncalibrated image is used as the study object, and the geometric measurement of the unknown distance is realized, the depth information of the scene contained in the image is recovered, and the precision of the geometric measurement and the depth calculation is evaluated. In this paper, the method of measuring the geometric quantity of a single-amplitude image based on the cross-specific principle is studied, and the accuracy of the measurement results is evaluated. The error source is analyzed, and the error prediction and repair are established. Positive models. Primary study content and The results are as follows: (1) The single image cross-level based on the intersection ratio is realized. The method of the two-dimensional measurement of single-amplitude image based on the intersection ratio is studied. The model of the linear distance measurement and the model of the total-surface distance measurement are extended to the model of the cross-plane distance measurement and the known method is reduced. (2) The error source of the single-image geometric measurement is analyzed, and the phase is presented. In this paper, the errors of geometric measurement are analyzed from the aspects of camera lens distortion, point-of-point calculation and error propagation, pixel pick-up deviation, known condition number, known condition distribution, etc. the results show that 1) the measured error has a strong correlation with the distance between the line segment to be measured and the center of the known rectangular center; 2) when the number of known conditions exceeds three, the mean value of the three groups of measurement results can be displayed to improve the precision of measuring the measuring precision; 3) the magnitude of the error of measurement and the point of the known condition (3) The model of geometric measurement error prediction and correction is established, and several methods are implemented. In this paper, based on the strong correlation between the measuring error and the distance (the line to be measured to the center of the known rectangle), the error prediction and correction model of single-image geometric measurement is constructed, and the overall precision of the measurement result is improved, and the single image is realized. The quantitative analysis of the measurement error of the image is that the measurement of the image is only the measure information of the image, and the depth of the image is an integral part of the image spatial information, and the image measurement and the depth calculation also Therefore, the present depth calculation method is improved, the absolute depth calculation is realized, and the depth calculation results are evaluated. The accuracy of the method is improved. The main research contents and results are as follows: (1) The depth calculation model is improved and the implementation is realized. The image continuous depth semi-automatic estimation is presented. The ground point depth calculation model based on the pinhole imaging model is improved, and the depth calculation and estimation are carried out on the basis of the object segmentation result and the spatial position relation in the image. image continuous deep semi-automatic estimation. (2) a high-depth-based camera is proposed In this paper, the error influence factor of the depth calculation model is analyzed, and it is found that the height error of the camera is one of the main influencing factors of the depth calculation precision. Based on this conclusion, the 鈥淐amera height recovery method based on ground point known depth difference鈥，

本文编号：2384660