To address the increasing demand for high-precision map modeling in operational environments involving unmanned rollers, this study proposes a robust map matching method that integrates RTK-GNSS positioning data with a sliding window-based polynomial fitting algorithm. Static experiments were conducted to characterize the error properties of RTK-GNSS measurements, upon which a Gaussian error distribution model and corresponding confidence intervals were developed. Furthermore, a composite smoothing strategy, integrating sliding window filtering with polynomial fitting, was developed to reduce noise disturbances along linear and curved road boundaries, enhancing coordinate accuracy and enabling precise boundary estimation. Experimental and simulation results demonstrate that the proposed approach reduces the mean boundary fitting errors to 0.138 cm for straight sections and 0.178 cm for curved sections, effectively suppressing positioning noise and fully satisfying the requirements for high-precision map matching in engineering applications. The findings provide robust technical support for high-precision mapping and trajectory planning in autonomous roller operations.