Probabilistic Mixture Model for Mapping the Underground Pipes

Abstract

Buried pipes beneath our city are blood vessels that feed human civilization through the supply of water, gas, electricity, and so on, and mapping the buried pipes has long been addressed as an issue. In this article, a suitable coordinate of the detected area is established, the noisy Ground Penetrating Radar (GPR) and Global Positioning System (GPS) data are analyzed and normalized, and the pipeline is described mathematically. Based on these, the Probabilistic Mixture Model is proposed to map the buried pipes, which takes discrete noisy GPR and GPS data as the input and the accurate pipe locations and directions as the output. The proposed model consists of the Preprocessing, the Pipe Fitting algorithm, the Classification Fitting Expectation Maximization (CFEM) algorithm, and the Angle-limited Hough (Al-Hough) transform. The direction information of the detecting point is added into the measuring of the distance from the point to nearby pipelines, to handle some areas where the pipes are intersected or difficult to classify. The Expectation Maximization (EM) algorithm is upgraded to CFEM algorithm that is able to classify detecting points into different classes, and connect and fit multiple points in each class to get accurate pipeline locations and directions, and the Al-Hough transform provides reliable initializations for CFEM, to some extent, ensuring the convergence of the proposed model. The experimental results on the simulated and real-world datasets demonstrate the effectiveness of the proposed model.