节点文献

矿区变形监测精密近景摄影测量关键技术研究

Research on Key Technology of Precision Close-Range Photogrammetry for Deformation Monitoring in Mining Area

【作者】 孙久运

【导师】 高井祥;

【作者基本信息】 中国矿业大学 , 地图制图学与地理信息工程, 2010, 博士

【摘要】 矿区地质灾害在世界范围内都是令人关注的问题,矿区沉陷和变形监测是矿山安全生产的重要技术保障手段。论文根据矿区变形监测的问题与需求,结合计算机视觉、数字图像处理,与坐标及其置信区间估计理论,对矿区变形监测中精密近景摄影测量的有关理论与方法进行研究。高质量影像是精密近景摄影测量的基础。论文在分析CCD/CMOS成像原理的基础上,探讨了利用相机原始信号和多余影像互补信息,在影像质量评价指导下,高动态范围影像获取方法;对于数据预处理,应用巴特沃斯滤波器改进各向异性扩散方程,实现了边缘保持的影像降噪效果。高精度定位是精密近景摄影测量的关键。本文对定位算子展开了深入分析,改进了椭圆拟合和巴特沃斯拟合法圆点定位,使之计算简单、鲁棒性好,提出了统计和几何分析相结合的定位策略,并引入了两种易于实现的几何分析法定位。根据相关法定位可以达到较高的相对位移精度,提出了匹配、超分辨率重建、定位三步结合的高精度定位策略。影像特征定位作为一个估计问题,在确定目标位置的同时,若能给出精度及其置信区间则更为可信。本文结合蒙特卡洛计算方法,探讨了CRLB一定置信度下,对目标定位理论精度与置信区间,以及实际定位算子的理论精度与置信区间进行估计的理论方法。为实际定位算子的性能比较提供基准,也为设计新的更高性能的评估器提供准则,也可避免研究者做无用功寻求超过限值的不可能结果。微小位移和裂隙监测需要能反映纹理细节的更高分辨率影像,但目前依靠硬件改进提高分辨率的方法已近极限。本文分别针对高精度定位和全画幅重建的要求,提出了局部重建的Log-Polar与DT-CWT内插结合的高分辨率重建方法,和基于LSTM匹配的自适应全幅重建两种方法。传统近景摄影的局限性不仅来自于数字图像处理与模式识别算法问题,解析模型也影响极大。本文把计算机视觉中自标定理念,和多视几何方法引入近景摄影测量,改进了顾及镜头系统误差影响可应用于平面监测的EDLT方法;和应用序列图像进行在线标定与监测同步进行的光束法模型。论文最后把所探讨方法,应用于矿山相似材料模型室内实验、滑坡体采集和巷道裂隙勘查等实际工程,获得了较高的精度。验证了所提出方案的可行性,和算法的精确性。

【Abstract】 Geological disasters in mining area get special attention over the earth, anddeformation monitoring is importance indemnificatory ways and means. Based on theclose-range photogrammetry developing trends, integrating the researching results ofcomputer vision, digital image processing, and statistical theory; as well as aiming atthe request of high precision in deformation monitoring, this paper focuses on theprecision measurement by considering many methods from multi-tache to enhanceprecision. According to the proper sequence of collection, process, location,super-resolution restoration, resolution model analysis, and engineering applicationpractice, to outspread the research.High quality image is the basic requirement. Based on analyzing theCCD/CMOS imaging theory, this paper discusses using originality signal of cameraand complementary information superabundance images, and under the guidance ofthe image quality evaluation to obtain high dynamic range image; with regard toimage noise reduction, the paper realizes the purpose of edge retentive noise reductionby using butterworth filter to improve the anisotropic diffusion equation.High precision location is the key problem of precision close-rangephotogrammetry. Based on deeply analyzing the location operator, the ellipse-fittingmethod and Butterworth Tepuy fitting were improved, which are more simple androbust. The paper brings forward location strategy which integrates statisticalapproach and geometric approach, and introduces into two kinds of geometricapproach. Based on the characteristic of Correlation Locating Algorithm, it can getrelatively high precision displacement and it brings forward a new high precisionlocation strategy by matching, super-resolution restoration, and location.As a low error bound of parameter estimating, CRLB is frequently used toevaluate arithmetic estimating performance. Combining with Monte Carlo algorithm,this paper discusses the theoretic approach to estimate the academic precision andconfidence interval of target location in certain confidence, and the academicprecision and confidence interval of fact location operator. It provides the benchmarkfor performance compare of fact location operator and the guide line for high poweredalgorithm design.Sometimes close-range photogrammetry not only requires getting the position ofmonitoring points, but also the texture compare analysis is important too. Aiming athigh precision location and full image application, it brings forward integrating Log-polar and DT-CWT super-resolution restoration method for part restoration, andself-adaptive method based on LSTMThe insufficiency of traditional close-range photogrammetry not only comesfrom the problem of digital image processing and pattern recognition algorithm, butalso is influenced by resolution model. Self-calibration method and multiple viewgeometry are introduced into the dissertation. This paper brings forward EDLTmethod for plane monitoring which considers the system error, and bundle adjustmentmodel which can be calibrated online and synchronized monitoring using imagesequences.Finally, the paper applies the method which has been discussed to similarmaterial model test and coast, and gains the better results. It proves the feasibility ofthe scheme and the accuracy of the algorithm.

节点文献中: