Pain Assessment Using Facial Action Units and Bayesian Network

Automatic pain assessment systems based on facial videos are consistently studied due to the demand of robust and cost-effective pain management. In order to improve the assessment accuracy under the dynamic and uncertain pain assessment environment, we propose a novel pain assessment method, AUBN, based on facial action units (AUs) and Bayesian network (BN). Firstly, the key feature points of pain expression are extracted through constrained local neural field (CLNF) model, and then AUs with a large amount of pain information are identified. The AUs are labeled to form the sample data set, and the constraints on BN conditional probabilities are constructed from the qualitative expert knowledge. Then, the sample data set is fused with the constraint extended parameter set using the variable weight method. Finally, BN reasoning method is applied to achieve the recognition of facial pain expression. Experimental results show that the proposed method outperforms SLPM+MCSVM, TDSBP+SVM and LBV+CNN methods on recognition accuracy and can significantly improve the recognition performance of facial pain expression.