GPR Antenna localization based on A-Scans

Automated subsurface mapping with data obtained from Ground Penetrating Radar (GPR) is essential for the construction services. So far, significant progress has been achieved in this domain by integrating such sensors with robotic platforms allowing large scale autonomous subsurface mapping. The paper at hand tackles the challenging issue of self-localization of a GPR antenna in a known subsurface map by utilizing solely GPR measurements. This is achieved by isolating spatiotemporal salient regions on consecutive GPR traces. These regions are represented by utilizing the coefficients of the Discrete Wavelet Transform (DWT) decomposition. Matched representations indicate meaningful tracked regions on the GPR traces that correspond to a fixed time window of data recording. Tracked regions are encoded in the form a vector that is treated as an observation within a particle filtering framework and is further processed to estimate the GPR sensor pose, given (i) a known subsurface map (ii) a simulated GPR model and (iii) priors in the GPR motion model. The GPR antenna self-localization approach has been assessed with real data and exhibited promising results, proving the ability of the proposed method to perform subsurface localization, exploiting only GPR sensor measurements.