Camera Based Eye State Estimation for ICU Patients: A Pilot Clinical Study
In the Intensive Care Unit (ICU), the awakening of patients from comas is indicative of recovery. This article investigates the feasibility of using conventional and deep learning-based methods for eye state estimation based on videos recorded by a CCTV-camera installed in the ICU. For handcrafted feature-based methods, HOG and RGB features are combined as the input of the SVM classifier to classify the eye state as open and closed. For deep learning-based methods, the eye and face images were used as joint input for classification. The clinical trial involved 48 ICU patients, and the accuracy of the benchmarked methods was compared. The results show that the HOG-RGB based method achieved an accuracy of 91.39%, while the deep learning-based method achieved an accuracy of 89.35%. These findings highlight the chances of using CCTV cameras to estimate the eye state of ICU patients, which can be a useful mean to provide information regarding the consciousness for clinicians to assess the patient's recovery.