Road collapses have occurred frequently in recent years. They are caused by subsurface cavities that develop when the surrounding soil is drawn into aging buried pipes through damaged sections. It is therefore necessary to identify the locations of buried pipes where cavities are likely to form. In this context, analysis of radar images obtained from vehicle-mounted ground penetrating radar (GPR) is a promising way to survey the vast subsurface of road networks quickly and accurately. Many deep learning–based methods have been proposed for radar image analysis, but they are largely limited to estimating the approximate locations of buried pipes, and high-resolution localization remains difficult. In this study, we propose a training-free method for detecting buried pipes. Theoretical hyperbolas are fitted to the radar images to extract the apexes of the reflection hyperbolas, and principal component analysis (PCA) is applied to the resulting apex point cloud to detect the linear structures corresponding to buried pipes. At a test yard, data were acquired with two systems, an air-coupled vehicle-mounted GPR and a ground-coupled push-cart GPR, and the detection results were compared. For the vehicle-mounted system, the measurement conditions were varied within the range that allows surveys at approximately 60 km/h. The hyperbolic reflections from buried pipes were weaker with the air-coupled system than with the ground-coupled system because of attenuation over the air gap between the antenna and the ground surface. Even so, the air-coupled vehicle-mounted GPR detected a polyvinyl chloride (PVC) pipe with a diameter of 100 mm buried at a depth of 1 m. Furthermore, when the method was applied to a public road, buried pipes extending across multiple traffic lanes were detected. These results demonstrate the potential of training-free pipe mapping with vehicle-mounted GPR.