Driver Drowsiness Detection with Region-of-Interest Selection Based Spatio-Temporal Deep Convolutional-LSTM

Driver fatigue and drowsiness instigate road traffic accidents while driving throughout the years. to reduce road traffic injuries and fatality cases, a real-time drowsiness detection system is needed by using artificial intelligence algorithms to detect drivers' tiredness and drowsiness at an early stage. This study proposes an automatic region-of-interest selection based stacked spatio-temporal convolution-long short-term memory (ConvLSTM) drowsiness detection neural network for an in-vehicle surveillance and security system. Haar Cascade classifiers are used to select the region-of-interest on the human face. A ConvLSTM model is implemented to extract spatio-temporal features from the selected region-of-interest and to predict the drowsiness state of the driver. The performance of the proposed model is compared with various pre-trained deep learning models such as CNN, VGG-16, VGG-19, ResNet-50 and MobileNet. The proposed model is trained on the Yawn Eye and MRL benchmarked image datasets. The proposed approach achieves an accuracy of 99.44% on the Yawn Eye dataset and 90.12% on the MRL dataset. The model is further tested and validated using a live feed camera.