Localization Method for Ear Surgery Robot Based on Point Cloud Registration of the Auricle

During robot-assisted ear surgery, precise movement of the surgical instrument to the planned position is essential for successful outcomes. To meet this positioning requirement, we propose a point cloud-based registration scheme. Our approach employs pass-through and statistical filtering to remove outliers, fol-lowed by fast uniform sampling and an enhanced voxel filtering technique for data downsampling. Features are extracted using the Fast Point Feature Histograms (FPFH) algorithm, and coarse registration is achieved through an optimized Random Sample Consensus (RANSAC) algorithm. Fine registration is accomplished using the Normal Distributions Transform (NDT) algorithm improved by the Levenberg-Marquardt (LM) method. Experimental results indicate that the pro-posed registration method can complete the task within 6 seconds, achieve a root mean square error (RMSE) of less than 0.3mm for the registered point cloud, and produce the final rigid transformation matrix.