Track irregularity in high-speed maglev systems serves as a critical excitation source for the vehicles. Precisely measuring long-wavelength track irregularity is of great significance for maintaining riding safety, reliability, and comfort. To achieve accurate characterization of stator surface and guide surface irregularities across a wide speed range, this paper proposes a method that integrates sensor data from the inertial-reference and chord-reference method. To estimate the pose and error states of the inspection sensors, this study constructs a composite manifold state space and employs an iterative update strategy for optimal state estimation. This process deeply integrates multi-source sensor information to suppress noise, ultimately successfully reconstructing high-precision track geometry curves based on a global pose reference. Experimental results demonstrate that the proposed method significantly enhances the reliability and accuracy of the reconstructed irregularity geometry data in the spatial domain. It achieves superior restoration accuracy for irregularities within the wavelength range of 0–200 m, providing effective algorithmic support for the wide-speed-range inspection and precise maintenance of high-speed maglev tracks.