Ensuring the safety and longevity of aging bridges has become a pressing concern due to factors such as increased traffic loads, environmental degradation, and extreme events. In this paper, we propose a novel approach for identifying curvature influence lines (ILs) of beam-type bridges using time-history acceleration data from low-cost accelerometers. Specifically, we focus on isolating the quasi-static component of the measured acceleration signal, which captures the curvature of the structure under a moving vehicle. To achieve this, an autoregressive (AR) model-based method is introduced. By training the AR model on ambient vibrations—recorded when no vehicle is present—we reconstruct the dynamic portion of the signal, and subsequently remove it from the total response obtained during a vehicle passage. This process yields a residual that accurately reflects the quasi-static curvature IL. A numerical case study on a simply supported concrete bridge highlights the effectiveness of the proposed approach. We also investigated the effects of road roughness, for which repeated tests, which can be carried out with different vehicles, improve the fidelity of the identified curvature IL. Results confirm that the AR model-based approach outperforms state-of-the-art threshold-based low-pass filters.