Drowsiness Recognition Using the Least Correlated LBPH

In recent years, the drowsiness recognition is widely applied to the driver alerting or distance learning. The drowsiness recognition system is constructed on the basis of the recognition of eye states. The conventional methods for recognizing the eye states are often influenced by the illumination variations or hair/glasses occlusion. In this paper, we propose a new image feature called "least correlated LBP histogram (LC-LBPH)" to generate a high discriminate image features for recognizing the eye states robustly. Then, the method of independent component analysis (ICA) is applied to derive the low-dimensional and statistical independent feature vectors. Finally, support vector machines (SVM) are trained to recognize the eye states. Furthermore, we design four rules to classify three eye transition patterns which define the normal (consciousness), drowsiness, and sleeping situations. Experimental results show that the eye-state recognition rate is about 0.08 seconds per frame and the drowsiness recognition accuracy approaches 98%.