Design and implementation of drowsiness detection system for drivers

Detecting drowsiness in advance is very important for preventing possible traffic accidents due to fatigue which result in physical and economic losses. It is possible to predict drowsiness by applying computer vision techniques to facial video captures using a camera. In this thesis, drowsiness measurement methods, data sets and image processing techniques used in the literature were examined and an adaptive threshold value method was proposed. Using datasets, features from the eye region, which have great knowledge in the detection of drowsiness, as well as fixed and adaptive threshold values for blink and long-term eye-closure detection were evaluated separately. This is intended to enable a better distinction between short-term blinks and long-term blinks. It was verified through experiments on two different datasets that the proposed adaptive threshold approach provides much more successful blink detection results than a fixed threshold. CLOSDUR and PERCLOS methods have been used in the literature for drowsiness detection. The percentile expression of the time the eye is closed for one minute (PERCLOS) was determined with the eye-opening information obtained using the adaptive threshold, and drowsiness detection studies were performed by combining it with the long-term eye-closure information (CLOSDUR). The results obtained are in good agreement with the actual reference values of the data sets found in the literature.