A hybrid approach to designing an autonomous driving alert system using geometrical features and gray level information of face images

An autonomous driving alert system has been designed in the paper to aware people on steering, feel tired and sleepy. Even at variable lighting condition the system can detect the state of the eye (open or not) by processing frontal or side views of the face image taken by a single camera mounted in the car. The paper tries to develop the system by simultaneously extracting geometrical features and gray-level information of face image unlike previous researchers who concentrate either of these techniques. The system consists of two major parts. In the first step, the centroid of the eye has been detected invoking fuzzy C-means clustering algorithm. With the help of the centroid, geometrical feature points of the eye are marked while tip of the nose and tip of the chin on the face image are also detected to obtain the reference point and rotational axis. A sub-image area around the centroid has been selected using the feature points of the eye and the pixels intensity of the sub-image area are mapped as input of a back propagation neural network. Passing a scanning window over the segmented eye, the program searches the highest density scan window and the pixels intensity of the window are mapped as the target pattern of the network. Thus, input-output patterns have been constructed from the persons of different age groups, sex and races to train the neural network. The second part, an online stage processes the frontal or side views of the face image of a person on steering, taking roughly 3 frames/sec and generates the input vector. The trained neural network now computes the target vector. The system alerts the person in case the average intensity of the pixels of the target vector becomes less than a predefined threshold.