点云变形序列特征提取及可监测性评价研究

The accurate extraction of deformation sequence characteristic is the important content of deformation monitoring. This subject studies the theory and method of extraction of point cloud deformation sequence characteristic according to the computation geometry, numerical analysis and error entropy. First, the noise removal fusion algorithm of point cloud based on the tensor voting and incremental finite-element theory is studied according to the influence of deformation sequence characteristic extraction by different point cloud noise, and the removal of typical and non-typical noise is realized. Second, study the relationship between error entropy and the capability of deformation extraction. The index of deformation monitoring capability is constructed based on the error entropy according to the scanning array and error entropy weight of different point. Finally, the adaptive extraction fusion algorithm of point cloud deformation sequence characteristic based on voting strategy is proposed according to the different deformation feature. The extraction algorithm of point deformation characteristic is studied based on the Hausdorff distance and cosine similarity model. The determination of level set threshold based on the pattern classification algorithms is studied, and the extraction of linear deformation characteristics based on fast level set is constructed. The total least squares conformal transformation based on Gaussian markov model is studied, and six parameters model of the global deformation is constructed. The objective of this subject will improve the accuracy and efficiency of extraction of deformation sequence characteristic, and promote the application of three-dimensional laser scanning in the field of deformation monitoring.

变形序列特征准确提取是进行变形监测的重要内容，本课题从计算几何、数值分析及误差熵理论入手，探讨点云变形序列特征提取理论及方法。首先，针对不同类别点云噪声对变形序列特征提取的影响，研究基于张量投票算法及渐进有限元理论的噪声分类剔除融合模型，实现典型和非典型噪声的剔除。其次，研究误差熵与变形可监测性的关系，根据扫描阵列及不同点位误差熵的权值状况，建立基于误差熵理论的变形可监测性指标模型。最后，针对不同目标物变形特点，提出基于投票策略的自适应点云变形序列特征提取融合算法。研究基于豪斯多夫距离和余弦相似度模型的点位变形特征提取算法；研究基于模式分类算法的水平集阈值确定方法，建立基于快速水平集的线性变形特征提取模型；研究基于高斯马尔科夫模型的整体最小二乘正形变换算法，建立整个点云表面变形六参数提取模型。本项目的研究，将提高变形序列特征提取的真实性和效率，促进三维激光扫描在变形监测领域的应用。

三维激光扫描目前已经得到了广泛的应用，特别是在变形监测领域，而变形序列的提取是进行目标物变形监测的关键，对变形序列提取的准确评价是保证变形监测监测的可靠性。针对此，本项目首先对点云噪声进行了研究，并在分析不同类别点云噪声的基础上，给出了基于有限元理论的噪声剔除方法，实现了噪声的精细剔除。其次，研究了变形可监测指标模型，通过分析点云误差模型，构建了点云误差椭球，并给出了平均误差椭球，实现了在点云误差服从正态分布情况下，进行基于误差椭球的点云精度的评价，再分析了点云误差服从广义正态分布情况下，构建了广义P-范分布的点云误差模型，并利用误差椭球和误差熵的关系，建立了误差熵模型，并构建了点云平均误差熵模型，从而实现了广义误差分布下的点云误差阈值的确定，在考虑误差空间不同方向极限值的情况下，根据不同期点云比对差值和误差熵的关系，确定了点云变形可监测指标。研究了边界领域点法向量投影夹角特征，并根据该特征实现了目标物的边界特征提取；分析了目标物折边点法向量聚类特征，给出了基于领域法向量k-均值聚类的折边提取算法；利用目标物配准过程中的旋转和平移参数特点，构建了六参数面状目标变形的提取，在结合线状及点状目标点云比对分析情况下，最终实现了不同特征的提取，利用不同特征的比较结合误差熵变形可监测指标，确定了变形监测的可靠性。本项目的研究成果已经应用在桥梁的振幅、建筑物裂缝及其他构造物的变形领域，对采集的点云数据进行了系统的分析和评价，确定了点云数据应用的有效性，并对基于点云比对的研究结果进行了评价，得出了可靠的变形监测结果。本项目的研究结果促进了三维激光扫描技术应用在变形监测及损伤检测中的可靠性，特别是利用本想项目的研究成果开发了基于三维激光扫描技术的滑坡监测及隧道监测软件。