6D Dental Pose Estimation for AR-Assisted Craniofacial Surgery

Abstract: 6D object pose estimation is a fundamental computer vision task for aligning virtual models with physical objects and is therefore essential for accurate augmented reality guidance. In craniofacial surgery, such alignment enables pre-operative anatomical plans to be visualized directly on the patient during the procedure. We present a markerless monocular framework for estimating the 6D pose of the maxillary dentition. Because the upper teeth are externally visible and rigidly connected to the maxilla and cranial base, their pose can be used to register the patient-specific skull model to the intra-operative camera view. The framework relies on a pre-operative intraoral scan and avoids patient-mounted markers, external tracking systems, and photorealistic synthetic RGB training data. Instead, patient-specific pose-estimation models are trained on synthetic geometric representations, with separate pathways for natural dentition and orthodontic brackets. The method was evaluated on real patient images using reprojection-based registration error and demonstrated accurate alignment across different subjects and imaging conditions. The system operates in real time, enabling continuous tracking by estimating the dental pose independently in each video frame without relying on temporal tracking alone. These results demonstrate accurate, real-time, markerless registration of patient-specific craniofacial anatomy using the maxillary dentition as a rigid anatomical landmark.