A markerless system for the registration of a bone’s pose is presented which reduces the setup time and the damage to the bone to a minimum. For the registration, a particle filter is implemented which is able to estimate a bone’s pose using depth images. In a phantom study, the pose of 3D-printed bones has been estimated at a rate of 90 Hz and with a precision of a few millimeters. The particle filter is stable under partial occlusions and only diverges when the bone is fully occluded. During a cadaver study, the preoperatively planned cutting edges have been projected as augmented reality (AR) templates onto the hip bones of five cadavers. By cutting manually along the AR templates, surgeons were able to extract ten transplants in the same time as with conventional osteotomy templates. Using the presented navigation system can save hours spent on the construction and production of conventional templates. In conclusion, this work represents one step towards a broader acceptance of robotic osteotomies.