Automatic localization of gas pipes from GPR imagery

In order to improve asset knowledge and avoid third part damages during road works, the localization of gas pipes in a non-destructive way has become a wide domain of research during these last years. Several devices have been developed in order to answer this problem. Acoustic, electromagnetic or RFID technologies are used to find pipes in the underground. Ground Penetrating Radar (GPR) is also used to detect buried gas pipes. However it does not directly provide a 3D position but a reflection map called B-scan that the user must interpret. In this paper, we propose a novel method to automatically get the position of gas pipes with GPR acquisitions. This method uses a dictionary of theoretical pipe signatures. The correlation between each atom from the dictionary and the B-scan is used as feature in a two part supervised learning scheme. Our method has been applied to real data acquired on a test area and in real condition. The proposed method presents satisfying qualitative and quantitative results compared to other methods.