Shoulder pain intensity recognition using Gaussian mixture models

Automatic recognition of pain intensity has an important medical application. The approach of automatic pain assessment boosts the psychological comfort of patients. It could be a direct help to children, mentally challenged people, very elderly people, patients in postoperative care, or people with transient state of consciousness. Since pain is a subjective phenomenon, it is quite difficult to have an automatic pain measuring device. The research is relatively new in this field and is constantly evolving. In this paper we propose a completely automatic shoulder pain intensity recognition system. A very small dimensional directional displacement geometric feature vector is extracted automatically from prominent facial regions. To classify the features into sixteen levels of intensities Gaussian Mixture Model (GMM) and Support Vector Machines (SVM) are used. The UNBC-McMaster Shoulder Pain Expression Archive Database is used for the experimentation. The database has various challenges associated with it including the problem of head orientation which is also addressed in this work. We achieve an average recognition accuracy of 82.1% using GMM and 87.43% using SVM classifier.