Vision-Based Driver Authentication and Alertness Detection Using HOG Feature Descriptor

Driver’s drowsiness is one of the main causes of the car collision, which raise fatality and wound count worldwide. This makes it necessary to develop a system that ensures the safety of the driver as well as co-passengers. Further, driver authentication plays an important role in preventing car robberies and fraudulent switching of designated drivers, thus ensuring the security of the vehicle and passengers. In this work, we propose an algorithm for real-time and continuous driver authentication and drowsiness detection. Driver authentication is implemented using face recognition with dlib’s face detector which is found to be robust compared to openCV face detector. The SMS regarding authentication of the driver is sent to the vehicle owner with details so that he can keep track of drivers. Further, if the driver is authenticated, then he is monitored continuously using the webcam to detect the early signs of drowsiness. Behavioral measures of the driver-like eyelid movements and yawning are considered for the detection. First, for face detection, we will apply facial landmark detection made using Histogram of Oriented Gradients (HOG) feature and then extract the eye and mouth regions using shape predictor with 68 salient points next to the eye aspect ratio and mouth aspect ratio, i.e., EAR and MAR, respectively, are computed in order to determine if these ratios indicate that the driver is drowsy, if so then he is alerted using speech message.