This paper addresses the challenges of trajectory tracking accuracy and adaptive control for vessels navigating curved waterways by proposing an integrated approach combining a digital twin environment derived from Electronic Navigational Charts (ENCs) and a curvature-adaptive Model Predictive Control. First, a high-precision digital twin model of the waterway is constructed through the extraction, filtering, and fusion of multi-source static elements (e.g., shorelines, depth contours, and navigation marks) from S-57 ENCs, establishing an environmental perception foundation for autonomous navigation. Subsequently, to overcome the limitations of conventional Line-of-Sight (LOS) guidance—such as fixed look-ahead distances and neglect of vessel dynamics in curved waterways—a curvature-adaptive dynamic LOS algorithm is developed. This algorithm dynamically adjusts the look-ahead distance by coupling real-time vessel speed with local waterway curvature. Furthermore, an adaptive PID control strategy is integrated to update the model and optimise heading control based on real-time vessel response data, thereby enhancing adaptability to time-varying dynamics and unknown disturbances. Experimental results demonstrate that the proposed method achieves superior tracking accuracy and robustness in curved waterway scenarios.