Underground Pipeline Mapping Based On Multiple Sensors Fusion

Underground pipeline detection and mapping is an important part of urban con-struction.Ground Penetrating Radar(GPR)is a widely used tool for detecting under-ground pipelines.By combining GPR with GPS,the positions and directions of un-derground pipelines could be estimated as well.When the GPS signal is blocked by high-rise buildings,Simultaneous Localization And Mapping(SLAM)algorithm could provide location assistance.when the survey points in a detection site are all detected,a pipeline mapping algorithm is proposed based on the Dirichlet Process Mixture Model(DPMM).This paper consists of three parts:GPR data processing,SLAM algorithm and pipeline mapping.GPR data processing is to process the B-scan images and esti-mate the depth and radius of the pipeline.SLAM algorithm is to estimate the location when the GPS could not work.Pipeline mapping is to map the underground pipelines by the data estimated by the two algorithms above.There are some challenges when detecting and mapping pipelines.B-scan images contain large number of data.There are accumulative errors during SLAM.There is a lack of prior knowledge of pipeline map.To solve these problems,the main contribu-tions of this paper are summarized as below:1.The B-scan images are only scanned once to detect the pipelines.The low time complexity makes it possible for the algorithm to run during GPR detecting pipelines.2.Submaps and loop closure detection are used to reduce accumulative errors.The accumulative errors are limited in each submaps and when loop closure is de-tected,the accumulative errors in each submap will be optimized.3.DPMM helps to calculate the prior probability of the survey points.To avoid using prior knowledge when simulating the mixture distribution,survey points are randomly sampled from the input dataset.Least square error and maximum likelihood are used to reduce noise data when fit the pipelines.The experimental results show that the sensor data could be processed during ex-periments.The pipeline mapping results show that the algorithm could work in various environments.