基于灾后激光雷达点云数据的地震灾害建筑物损毁评估

The buildings damage assessment is the key content of disaster physical quantity assessment. Remote sensing has played an important role in building damage assessment during Wenchuan Earthquake, Yushu Earthquake, Yingjiang Earthquake. As a new airborne remote sensing thechnology, LIDAR has been applied in Disaster condition assessment.It is practical to use LIDAR points cloud data to carry out post-earthquake building damage assessment, because this method isn't affected by lack of pre-disaster points cloud data. Existing methods lack the means of quantitative assessment of damage condition, and can be further improved by improving the recognition accuracy and level of automation. The project intends basic object-oriented supervised classification process to extract damage building area in LIDAR data. Mean Shift algorithm is introduced to segment points cloud data. Spatial distribution characteristics and topological relationship characteristics of roof object are calculated, and optimal combination of characteristics is selected. The Sparse Expression classification algorithm is also introduced. Finally, according to the incompleteness of the geometric characteristics, damage condition is quantitatively assessed. A building damage assessment process after earthquake, using post-earthquake LIDAR data, is then constructed. The research results will be directly applied to the national building damage assessment work after earthquake disasters. To some extent, the automation level and accuracy of national building damage assessment operation after earthquake disaster will be improved.

建筑物损毁评估是灾害实物量损毁评估的重要内容，在汶川地震、玉树地震和盈江地震，遥感都在建筑物损毁评估工作中发挥了重要作用。激光雷达作为一种新兴的航空遥感技术已成功应用于灾情评估工作中。利用震后激光雷达数据开展震害建筑物损毁评估的方法可以不受灾前点云数据缺少的影响，具有实用价值。现有方法缺少损毁程度定量评估的方法，并且可以通过改进进一步提高识别精度和自动化程度。本项目拟以面向对象监督分类的基本流程提取激光雷达数据倒损建筑物区域，引入均值漂移算法进行分割，计算面片对象空间分布特征、拓扑关系特征等并进行组合优选，引入稀疏表达算法进行分类，最后按照几何特征的不完备性进行损毁程度定量评估，构建一套应用灾后激光雷达数据进行震害倒损建筑物损毁评估的方法流程。研究成果将直接应用于国家地震灾害建筑物损毁评估业务中，在一定程度上提高我国地震灾害建筑物损毁评估业务的自动化程度和精度。

本研究针对灾后激光雷达点云数据建筑物损毁评估问题，对单体建筑物分割、建筑物损毁评估指标构建进行了深入研究，并制备了一套灾后激光雷达验证数据，取得了一些有益的成果。具体包括以下3个方面内容：.（1）灾后激光雷达验证数据制备.赴四川省北川地震遗址开展地基激光雷达数据扫描，获取点云数近6000万个，最小点间距达2mm，点云覆盖范围超过3.1km*2.0km。对灾区数据进行了2倍降采样，并且对一些包含建筑物的大区域进行了分割，提取出6个有代表性的区域，对建筑物损毁情况进行人工标注，包含完好和损毁建筑物。.（2）点云数据单体建筑物分割算法.建筑物单体分割是建筑物损毁评估的基础。研究对逐点求取的最近邻协方差矩阵进行SVD分解构建局部平面特征，作为面片提取的判别信息；然后，利用马尔可夫随机场进行平面区域与非平面区域的分割；利用约束k-means算法获取平面区域法向量，删除地面区域，得到建筑物区域；对每个面片进行区域生长分割得到单体建筑物，平均检测精度优于95%。.（3）建筑物损毁评估指标构建.考虑到建筑物损毁三维特征，提出了平均倾斜角、平面置信度和建筑物完整性3个指标，从建筑物倾斜程度、轮廓高程分布状态、墙体完整程度3个方面对建筑物损毁进行评价。平均倾斜角是单体建筑物所有平面与地面法向量交角的平均；平面置信度主要反映建筑物点云分布的起伏程度；建筑物完整度是所有墙体平面和空洞的比值的平均值。3个指标均能够较为精确地反映建筑物的损毁情况。