Globally, the road safety is a major concern in which driver drowsiness and fatigue being significant contributors to traffic and accidents. According to previous studies, it is evident that the drowsy driving affects abilities of decision making, reaction time, attention etc. leads to fatal consequences. The early detection of driver drowsiness and fatigue aim to detect tiredness in drivers result in prevention of accidents. In this work, a deep learning framework used night vision imaging for driver drowsiness detection in real-time. Furthermore, the facial landmark is used to preprocess the night vision images and the three different models such as Gated Recurrent Unit (GRU), Recurrent Neural Network (RNN) and Convolutional neural network (CNN) are utilized to classify two different driver states namely drowsy and non-drowsy. Also, the models performance are analyzed using various performance metrics. All the models are coded using Python programming and the night vision input images are given for the models to train-validate-test of the utilized models. The CNN model exhibits the results to the maximum accuracy of 98.7% which outperforms both RNN and GRU in detecting driver drowsiness under night conditions. The sensitivity of CNN is 98.5% which ensures reliable detection of drowsy states with minimal false negatives. Additionally, the industrial feasibility is validated through embedded hardware testing using Raspberry Pi and low computational latency.