GamRec: A Clustering Method Using Geometrical Background Knowledge for GPR Data Preprocessing

摘要: GPR is a nondestructive method to scan the subsurface. On resulting radargrams, originally interpreted manually in time consuming process, one can see hyperbolas corresponding buried objects. For accelerating interpretation machine shall be enabled recognize on radargrams autonomously. One possibility combination of clustering with an expectation maximization algorithm. However, there no suitable algorithm for hyperbola recognition. Hence, we propose specialized this problem. Our approach directed shape based combined sweep line In contrast other approaches our finds shaped clusters and (1) able intersecting hyperbolas, (2) noise robust (3) does not require know number beginning but it number. This important step towards goal fully automatize object detection.