Eye blink based fatigue detection

V magistrskem delu smo preučili možne rešitve zaznavanja utrujenosti očesa na podlagi mežikanja. Ugotovili smo, da obstaja več tehničnih rešitev, s katerimi je možno zaznati mežike. Rešitev, ki zaznava mežike z obdelavo zaporednih sličic s kamere, je za uporabnika najmanj invazivna in ne zahteva dodatne strojne opreme, zato smo preučili več metod tovrstne rešitve. Metoda, ki nam je dala najboljše rezultate, analizira premike na dveh zaporednih sličicah. Druga metoda, ki daje prav tako zadovoljive rezultate, zaznava mežike na podlagi ujemanja med predlogo odprtega očesa. Tretja metoda, ki smo jo preizkusili, zaznava količino črne barve na območju oči. Pri tej metodi se je izkazalo, da se količina črne barve ne zmanjša dovolj, da bi lahko na podlagi tega zaznavali mežike. Ob mežikih se površina trepalnic poveča in s tem vpliva na količino črne barve na območju oči. Za tovrstne metode smo izdelali ogrodje, ki v eni niti zajema sličice z maksimalno hitrostjo kamere in jih shranjuje na koncu povezanega seznama. V drugi niti pobira sličice z začetka povezanega seznama in jih posreduje metodi za procesiranje. To omogoča zaznavanje kratkih mežikov in uporabo metode, ki ne teče ves čas v realnem času. Ko metoda zazna majhno število mežikov, obvesti uporabnika z zvočnim signalom. Rešitev smo izdelali za osebni računalnik z operacijskim sistemom Linux in za pametni telefon z operacijskim sistemom Android.This master thesis analyses possible approaches for eye fatigue detection based on eye blinking. We have found multiple approaches for eye blink detection. The approach based on consequent images acquired from a video camera is the least invasive for the user and does not require additional hardware, hence we analysed multiple methods of this approach. The method which gave best results detects eye blinks by analysing movements on two consequent images. The second method which also gives satisfactory results detects eye blinks using open eye template matching. The third method detects eye blinks by analysing the amount of black color in the eye region. This method showed that the amount of black color does not decrease enough to reliably detect eye blinks, since the surface of eyelashes increases the amount of black color in the eye area when the eye is closed. For these methods we have built a framework which in one thread captures images at maximum speed of the camera and saves them at the end of a linked list. In the second thread, the framework takes the images from the beginning of the linked list and transmits them to the method for processing. That makes it possible to detect very short eye blinks and to use a method which does not run in real time all the time. When the method detects low eye blink rate, it informs the user with an audio signal. We developed a solution for personal computers with the Linux operating system and smartphones with the Android operating system