Predict Driver Fatigue Using Facial Features

Fatigue on board is amoung the causes of fatal accidents today, however drained driver means that there a metal box without control that threatens lives in our roads. Nowadays, there is no effective non-intrusive method to detect driver tiredness. This paper presents a method for early prediction signs of fatigue during driving. Luckily our face characteristics reflects our current state, therefore we will base on facial features to monitor the driver state. To localize driver's face we had chosen Viola-Jones for faces detection and for face tracking Kanade-Lucas-Tomasi is the best alternative due to their implementation simplicity using just standard phone camera equipped by IR LED. Then all extracted frame characteristics will be presented to SVM for classification that separate the normal state from the critical state. Our objective is to avoid false alerts and early fatigue detection in real-time, for this reason we will combine HOG+SVM, eyes blink rate/duration and PERCLOS. The driver state detection and fatigue alert are not the final steps in our method because a bad reaction can cause disasters that is why we include different road users (V2V and V2I) by shearing alert notification message.