Fusing Heterogeneous Information for Underground Utility Map Generation Based on Dempster-Shafer Theory: Fid-Bau Portal

Fusing Heterogeneous Information for Underground Utility Map Generation Based on Dempster-Shafer Theory

Cai, Jiannan / Jeon, JungHo / Cai, Hubo / Li, Shuai

Accurately mapping and labeling buried utilities is critical to managing massive urban underground infrastructure and preventing utility incidents. However, current spatial information regarding underground utilities is inconsistent, inaccurate, and uncertain, which is a root cause of numerous utility incidents. Information about underground utilities is usually provided by multiple sources, such as existing records and nondestructive testing technologies [e.g., ground-penetrating radar (GPR)]. A key challenge is the integration of imperfect data obtained from heterogeneous sources to create accurate underground utility maps. Aiming at accurately confirming the presence or absence of underground pipes, this study presents an information fusion method based on Dempster-Shafer (D-S) evidence theory to integrate both sensing and nonsensing data. The GPR and existing utility records serve as two independent information sources that provide evidence regarding the existence and configuration of buried utilities. The trust level of evidence provided by different sources is modeled as mass value, and practical guidance is proposed to quantitatively assign the mass value of individual data sources by considering various factors that affect the accuracy of each data source. By fusing the evidence from heterogeneous sources, the joint mass value is computed, based on which the state of pipes is inferred. Indoor experiments on three different scenarios demonstrate the efficacy of the proposed method. The results show that the proposed fusion approach leads to higher accuracy, precision, and recall in detecting pipes than using an individual data source. Furthermore, the robustness of the proposed mass value selection guidance is validated through sensitivity analysis, namely the fusion result remains stable when the mass value for an individual data source varies slightly.