滑坡误检测形变区域定位与分析

Identification and Analysis of Error Detection Areas in Landslide Deformation

植被检测是利用近景影像进行滑坡变形监测应用的一项关键步骤,滑坡近景影像中植被信息的特殊性给自动化匹配带来困难。针对经过高精度影像配准和点云滤波等处理后仍然存在形变区域检测误差,通过叠加滑坡体形变前后的数字表面模型和植被检测结果,实现对误检测形变区域的定位和分析。实验结果表明,形变区域检测误差主要来自植被剔除残余、点云滤波或点云模型不精确和数字高程模型采样误差等。

Vegetation detection is one of the key steps in landslide deformation monitoring based on close-range images.Automatic matching has been proved difficult because of the Particularity of the existing vegetation information in landslide close-range images.In order to deal with the problem of deformation detection error after high-precision image matching and point cloud filtering,this paper identifies and analyses the deformation detection error by overlaying the DSMs before and after the deformation combining with the vegetation detection result.The results indicate that the deformation monitoring errors mainly come from the residual of vegetation removal,the error of point cloud filtering or inexact point cloud model and the DEM sampling error.