6D Pose Estimation Based on Edge-Enhanced Point Pair Features for Surgical Navigation

This paper proposes an efficient 6D pose estimation method for the automatic spinal surgical navigation system. Visual-based pose estimation is challenging in realistic surgical scenes due to the human spine's geometric complexity and symmetric ambiguity. Instead of considering the whole scene for pose estimation, we introduce a targeted down-sampling strategy that focuses more on edge area for efficient feature extraction of complex geometry. A pose hypothesis validation approach is then presented to resolve the symmetric ambiguity. Note that our method can also work on incomplete scenes since our sampling and validation do not require a whole model. We perform evaluations on two challenging datasets, which demonstrate the superiority of our method on pose estimation of geometrically complex, occluded, symmetrical objects. We also implemented a prototype of a surgical navigation system based on the proposed method, which may provide a viable option for improving the spinal clinical surgery system.