Vision based Driver Drowsiness Detection using Deep Learning

Every year many people lose their lives in road accidents, according to a survey many road accidents occur due to driver fatigue condition. Feeling sleepy, tiredness due to long journeys, being on alcohol while driving has costed many lives. In order to save lives a perfect technique must be implemented to detect driver drowsiness and alarm the driver before it costs the driver life. Deep Learning is an emerging technology has given solution to many problems in different domains. Deep Learning can solve this problem using Convolutional Neural Networks and Image Processing techniques. Proposed system solves the above-mentioned problem using object detection, driver video from the embedded devices in the vehicles is taken as input and identifies if eyes of driver in video are closed or opened and eventually an alarm sent. Object detection is done using MobileNet convolutional neural network architecture along with SSD (Single Shot Detector). On the output given from MobileNet and SSD architecture an algorithm is applied to detect drowsiness of the driver. We take 6000 images from different datasets which consists of images of people driving, eyes open, eyes closed, yawning and use them to train the model and then we label the images into different classes like face, open eyes, closed eyes. The trained model is then tested using 300 images taken randomly from the same dataset. Also we calculate evaluation measures like accuracy, precision, recall and F1 score for the model which is trained. The proposed model gives highest accuracy with less cost of implementation. Since, we use video of the driver taken from the embedded devices like mobile without any expensive hardware. The proposed model is efficient in accuracy, computation and also cost. This model can be deployed on to different embedded devices present in vehicles without any difficulty. The results of proposed model has shown accuracy of 97% and average mean precision of 0.84.