Head Movement Dynamics based Pain Detection using Spatio-Temporal Network

Pain detection is one of the crucial factors for pain management in the healthcare domain. Current techniques mainly focus on facial expressions for pain detection and assessment. However, automated pain detection using head posture remains largely unexplored. The head pose dynamics from videos over time provides crucial information related to pain behaviour. Therefore, in this study, automated pain detection is proposed by utilizing only head movement information. Here, it has been demonstrated that the performance of pain detection can be improved by learning jointly from both spatial and temporal features. To learn spatial and temporal dependencies between head pose series, CNN- BiLSTM network has been utilized. To validate our proposed method extensive experiments have been carried out on the UNBC-MeMaster pain dataset. The evaluation metrics demonstrate that the proposed method achieves encouraging results as compared to facial expression-based pain detection methods. Further, the challenges and feasibility of pain detection using head pose dynamics information have been discussed.