Computer Vision-Based Eyelid Closure Detection: A Comparison of MLP and SVM Classifiers

In this paper, a vision-based system to detect the eyelid closure for driver alertness monitoring is presented. Similarity measures with three eye templates (open, nearly close, and close) were calculated from many different features, such as 1-D and 2-D histograms and horizontal and vertical projections, of a big set of rectangular eyes images. Two classifiers, Multi-Layer Perceptron and Support Vector Machine, were intensively studied to select the best with the sequential forward feature selection. The system is based on the selected Multi-Layer Perceptron classifier, which is used to measure PERCLOS (percentage of time eyelids are close). The monitoring system is implemented with a consumer-grade computer and a webcam with passive illumination, runs at 55 fps, and achieved an overall accuracy of 95.75% with videos with different users, environments and illumination. The system can be used to monitor driver alertness robustly in real time.