Automatic detection of buried pipes from ground penetrating radar data

In previous works [1][2] the authors presented a method for the detection of hyperbolic patterns in radar B-scans by using the gradient-sigmoid filter. The processing of these patterns by a modified Hough Transform allows to estimate lateral position (y), depth (z) and propagation velocity (v) of targets. In this work we will present a further step towards the automatic detection of buried pipes/cables exploiting the 3d spatial tracking of sequences of hyperbolic patterns positions as obtained by the modified Hough Transform method on a number of adjacent B-scans. The capability of the full automatic process has been tested on experimental data acquired with a 200 MHz IDS-RIS radar on a test field containing 18 distinguishable pipes/cables (plastic or metal) arranged in different positions (deep and/or shallow, isolated and/or clustered). Obtained results are the correct detection of 17 out 18 pipes/cables with only 2 false alarms; the missed target being a single large deep pipe. The 3d spatial tracking is based on the assumption that a pipe is nearly orthogonal (within an assigned solid angle) to the B-scan plane and that a pipe is defined by a series of hyperbolic patterns positions belonging to different B-scans. In order to declare a pipe, the feasible sequences are checked to satisfy the following 5 properties: serialization, directionality, continuity, straightness and minimum energy (Dijkstra algorithm). The robustness of this pipe detection method has been tested on simulated data where false hyperbolic patterns and some degree of true target misplacement were introduced. The computational complexity of this type of algorithms is generally a problem for practical applications; in this case it has been limited to a linear factor working on small sequences of five apexes. A Matlab program has been executed in 43 s on a 1.5 GHz laptop PC on 18 B-scans of 20 meter length each and corresponding to an area of 140m2.