Probabilistic Situation Detection for Human-Robot Interaction in an OP Lab Environment
Luzie Schreiter, Elmar Berghöfer, Tim Beyl, Lisa Gutzeit, Joerg Raczkowsky, Frank Kirchner, Heinz Woern
In Tagungsband der 14. Jahrestagung der Deutschen Gesellschaft f{{\"u}}r Computer- und Roboterassistierte Chirurgie e.V, (CURAC-15), 17.9.-19.9.2015, Bremen, Fraunhofer MEVIS, pages 99-104, 2015.
Abstract
:
Robotic assistance systems are becoming increasingly established in operating rooms worldwide. To give a suit- able assistance at a proper time it is essential to identify and interpret the current step in an ongoing surgical intervention. We propose an approach which enables to observe and interpret an operation by using dierent sensors and probabilistic models. The core of our approach is based on machine learning methods, e.g., Hid- den Markov Models. These models are trained and optimized on recorded data containing actions of a surgical work ow. The actions are later detected by using the trained model. To evaluate the trained model we perform a leave one out cross validation and achive a high performance in detection of the actions. Furthermore, the trained model is integrated in the surgical setup and evaluated in an online manner. The results show that the online use of the trained model is successful in detection of dierent surgical work ows.