Feature-based error processing for robust surface registration in computer assisted orthopedic surgery

Accurate registration algorithms are required in computer assisted orthopedic surgery to provide a reliable correspondence between the bone and its tridimensional model. In surface registration, the main source of errors lies in the points acquisition process. This paper presents an online point processing algorithm based on features correlated to the error. The features are organized in classification trees to detect and remove outliers. Different common classification trees are investigated. The efficiency of the approach to remove outliers is demonstrated through experiments on a pelvic bone.