A new statistical method to segment photogrammetry data in order to obtain geological information

Abstract

A novel method is presented that is capable of more accurately extracting rock surface features based on the geometrical characteristics of a 3D point cloud obtained from a survey. The core feature of the algorithm is a newly established method that can provide a robust estimation of point normals, while excluding statistical outliers from the calculation. A resilient multivariate mean and covariance estimator, termed Deterministic Multivariate Mean (DetMM) estimator, is used, which is capable of removing outlier effectively when calculating normals. To ensure the efficient operation of the estimator, a hybrid method is proposed where a simple and efficient method to estimate normal vectors is implemented first to filter the data before the robust DetMM method is applied for more accurate estimation of difficult cases. By using this strategy, the efficiency is significantly improved while the same level of accuracy is maintained. In addition, a novel segmentation algorithm, a region growing method, is introduced to handle complex geological features , which is capable of accurately detecting very small rock surfaces. The robustness and reliability of the developed method are compared with those of the well-known normal estimation method using principal component analysis (PCA). Finally, the method is validated on two case studies where the 3D point datasets were gathered from scanning two very different rock faces. The results have demonstrated that the proposed method outperforms significantly the conventional techniques in terms of accuracy with an acceptable increase in computation cost.