Time-delay estimation (TDE) of pavement usingground penetrating radar (GPR) is an important task in the fieldof civil engineering. However, when dealing with GPR waves atthe centimeter scale, there are challenges in real-time processingthe coherent, overlapping backscattered echoes within limitedbandwidth. To address these problems, this paper proposes aTDE method for GPR signals using co-prime sampling strategyvia atomic norm minimization (ANM). The conventional uniformfrequency sampling strategy requires lengthy data acquisitiontime and large data storage in GPR system. To address theseproblems, the co-prime sampling strategy is adopted in thispaper to lower the sampling rate and lighten the hardwareburden. The “holes” in co-prime sampling are filled by solvingthe ANM problem, where the reconstructed Hermitian Toeplitzmatrix can be used for decorrelation without loss of degreesof freedom (DoFs). By selecting multiple-measurement vectors(MMV) directly from the second statistics of co-prime samplingsignal, the scale of the semidefinite programing (SDP) of ANMis reduced. Moreover, we propose a 2-level ANM by recoveringtwo Hermitian Toeplitz matrices instead of one Toeplitz matrixand one Hermitian matrix, reducing further the complexity ofthe proposed method. Numerical and experimental results showthe superiority of the proposed method in terms of temporalresolution, estimation accuracy, and computational complexity.