Abstract
A template matching method for detection of isolated objects in GPR (Ground Penetrating Radar) images is proposed. Echoes from point-like or line-like targets such as cable ducts and utility pipes form characteristic patterns in images measured by GPR. Such characteristic patterns in the images are searched as candidates of the targets by using template matching. Then the candidate is tested whether it effectively reduce the norm of the image or not when the template is subtracted from its neighborhood. The template is prepared based on a model of the targets and response of the GPR. The template matching enables identification of the target position and elimination of transient response of GPR at the same time. Effects of errors in models on identification of the targets are evaluated by numerical simulations to show the method can identify the targets even when small errors exist in the model. Also, it is successfully tested by using field data to obtain a 3D image of a flexible duct. The method takes time comparable to diffraction stack migration to process images and produces images easier to understand than the diffraction stack migration.
