Off-Line Driver’s Eye Detection: Multi-Block Local Binary Pattern Histogram Vs. Gabor Wavelets

Eye detection is a complex issue applied in several applications, such as human gaze estimation, Human-Robot interaction and driver’s aid framework for automatic drowsiness detection. This paper presents an algorithm that detects the human eye in still gray-scale images. The proposed scheme is based on learning statistical appearance model, which implies; the extraction of features and their classification. A comprehensive comparison of two photometric feature descriptors is performed between Gabor wavelets and Multi-block Local Binary Pattern histogram (BHLBP) features. Facial images are normalized and the eye features are extracted using the precedent descriptors, then Support Vector Machine classifier (SVM) is used to distinguish eye from non-eye classes. The presented schemes are built and tested with different frames, collected from a real driving video sequences of the RobeSafe Driver Monitoring Video dataset (RS-DMV), the driving sequences are recorded under realistic scenarios and with different subjects, which are exposed to a real driving environment. Discriminative performance of the descriptors are reported. Keywords—Eye detection, eLBPh, Gabor wavelets, Driver’s drowsiness.