Abstract: Cerebral blood flow velocity is an important parameter for assessing whether a patient has cerebrovascular spasm. Physicians measure cerebral blood flow velocity at specific locations using transcranial Doppler ultrasound (TCD). However, traditional TCD cannot directly visualize cerebral vessels, leaving physicians to rely on experience to position the ultrasound probe. This results in low measurement efficiency and inconsistent accuracy. To address this issue, this paper proposes an autonomous positioning and real-time navigation algorithm for TCD based on cerebral vascular magnetic resonance imaging (MRI). The algorithm first employs an incremental structure-from-motion (SfM) method to obtain a dense reconstruction of the patient’s face. Then, it achieves an initial registration between the dense reconstruction and the facial model through 3D-to-3D matching of facial biometric landmarks. Finally, real-time pose tracking is achieved using localization markers marked on the patient's face. This paper proposes a mismatch elimination method based on the RANSAC algorithm and fully connected conditional random fields (CRF), which effectively reduces the impact of feature point mismatches on pose estimation. Additionally, based on localization landmarks, this paper employs an extended Kalman filter (EKF) algorithm to achieve real-time tracking of the patient's facial pose. Experiments demonstrate that the proposed algorithm, based on cerebral vascular MRI, achieves accurate initial pose estimation even under outlier interference, with significantly better performance than existing PnP algorithms.