Markerless Registration for Augmented-Reality Surgical Navigation System Based on Monocular Depth Estimation

In augmented reality(AR) surgical navigation system, the depth estimation using RGBD camera has limitation in obtaining the dense depth necessary to increase the registration accuracy. Recently, deep learning based monocular depth estimation has showed remarkable performance. In this study, we developed a markerless registration method using the monocular depth estimation, and applied it to AR surgical navigation system. The accuracy of our method of monocular depth estimation was 2.47 ± 1.15mm, while that of the method of the RGBD camera was 2.33 ± 1.24mm. There was no significant difference by paired T-test. Furthermore, the monocular depth estimation was able to acquire denser depth than the RGBD camera.