Answers are grounded in this paper and cite their sources.
Marker-based AR image registration is a widely used technique for overlaying patient-specific anatomical models as virtual content using augmented reality head-mounted displays (AR HMDs), providing visual guidance in the real-world environment. The accuracy of AR registration is critical, particularly in surgical applications where sub-millimeter to millimeter precision is optimal. However, there is a lack of standardized evaluation methodologies and metrics for assessing the accuracy and stability of such registrations. To address this gap, we propose a novel evaluation approach for marker-based medical image registration using optical see-through AR HMDs. This method involves direct capture of AR images through the HMD display, enabling objective assessment of both the accuracy and stability of the registered images. Using a precise motion controller, we evaluate AR registrations using image-based and infra-red reflective markers under varying conditions, including different speeds and movements of the AR HMD and the tracked objects, as well as rendering computation loads. The root-mean-square error (RMSE) of the dynamic time warping path matching between the ground truth and the registered image, and latency derived from the cross-correlation of two time-series data are employed to quantify spatial and temporal accuracy. On the HoloLens 2 platform, we demonstrate a method to quantify increased spatial inaccuracy and latency due to the rendering loads and translational speed of the target marker, indicating the deterioration in the accuracy of AR registration. In contrast, during the translational and rotational movement of the AR HMD, the RMSE and the latency remained stable, which is attributed to onboard temporal warping adjustment in rendering.
No takes yet. Share an insight, caveat, or question.
Eom et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: