Indirect bridge health monitoring using vehicle-based measurements offers a cost-effective alternative to traditional methods that require sensors mounted directly on the bridge. However, extracting the bridge’s dynamic responses from vehicle vibrations is challenging due to road roughness and the complexity of the vehicle’s own mechanics. This paper presents a novel two-stage approach to estimate on-bridge contact point (CP) responses without relying on prior knowledge of vehicle properties. In the first stage, before the vehicle enters the bridge, a Kalman Filter (KF) combined with an optimization framework uses measured vehicle’s body accelerations to identify the vehicle’s mechanical parameters, creating a surrogate model without direct input measurements. In the second stage, as the vehicle crosses the bridge, this surrogate model and a second KF are used to estimate the CP responses. To evaluate the method’s robustness, simulations were performed using a half-car test vehicle driving on a Class B road profile that transitions onto a two-span bridge. Results show that while our method can identify the residual CP responses with high accuracy, the surrogate model itself is not unique, implying multiple configurations may yield similar CP estimates. Unlike existing techniques, this approach does not assume known vehicle properties, making it highly adaptable to real-world scenarios, where such data is often unavailable and particularly valuable for crowdsensing-based bridge monitoring, where diverse, uncalibrated vehicles are used.