Experiments with 2 and 3-Dimensional Detection Algorithms Preliminary to Driver Drowsiness Monitoring

Drowsiness caused by fatigue is one of the main factors in motor vehicle accidents. Signs of drowsiness are the reduction of spontaneous blinks, decrease in blink amplitude, increase in blink duration, decrease in blink rate, and forced eyelid closure. One of the methods for detecting these signs is video-image processing. The development of an algorithm that adequately describes the physical properties of eye signs and images is presented in this paper. A Gaussian model with a matching process was used to approximate the peaks in one-dimensional graphs, and circular and spherical features and shapes in two-dimensional and three-dimensional space. The developed algorithm, called Semi-Adhesive Approximation (SAA) was based on the Least Squares method. Its goal was to achieve an approximation by minimizing a dynamically weighed mean-square-error cost function that self-selects a sub-range of the data over which a good local approximation can be achieved to detect eye movements that identify microsleeps and consequently drowsiness state. Results showed that when the data included increased noise, SAA performed much better than existing methods. One-, two- and three-dimensional applications are presented using the fitting algorithm for peak, circle and sphere detection. Much better results, than the baseline Least Squares method, were obtained in all cases.