<p>In this paper, we address the challenge of achieving reliable communication and accurate vehicle sensing in vehicular networks where the direct Line-of-Sight (LoS) link is often obstructed and channel conditions vary rapidly due to mobility. In particular, we investigate the use of an Intelligent Reflecting Surface (IRS) to simultaneously enable both communication and sensing functions in a Vehicle-to-Infrastructure (V2I) scenario. An Extended Kalman filter (EKF) is employed to follow the time-varying Angle of Departure (AoD) and channel state, enabling predictive beamforming in dynamic vehicular environments. Based on the estimated state, closed-form IRS phase shifts are derived to improve the signal strength at the vehicle. In addition, the Maximum Likelihood Estimation (MLE) algorithm is utilized at the Roadside Unit (RSU) to obtain the target response matrix of the vehicle using the reflected signal. Our framework addresses real-time ISAC requirements by balancing low-latency EKF beam tracking with the computational overhead of MLE-based extended target sensing, meeting the high-performance computing demands of dynamic vehicular networks. Numerical results show that the proposed framework enables simultaneous downlink communication and vehicle sensing, while significantly improving achievable rate and tracking accuracy compared to baseline schemes such as random phase shifts and non-predictive beamforming.</p>