Fatigue Driving Detection Based on Eye Features

Fatigue driving is one of the main causes of traffic accidents. The detection and early warning of fatigue driving are of great significance for improving driving safety. This study proposes a fatigue driving detection algorithm based on eye features. A detector in the Dlib open-source library is employed to extract the key feature points of the driver's eyes, and the blink duration, amplitude, speed, and frequency are calculated in real time. In order to recognize the fatigue state accurately, an augmented transformer model, called Transformer-Attention model, is proposed by combining global and local attention mechanisms. In the self-attention layer of the encoder, the global attention mechanism is first introduced to capture long-term dependencies and identify the overall changes in the driver's state. Then, the local attention mechanism is applied on the last few time steps of the sequence to pay attention to the state fluctuations in the short term and the initial changes of fatigue. Through this dual attention mechanism, the model can capture the temporal changes of the driver's state effectively and classify the driver's state into three categories: “alert”, “semisleepy”, and “sleepy”. Through comparative experiments, it is found that compared with the HM-LSTM model, the blink sequence accuracy and video accuracy of the proposed model are improved by 33.2% and 25.71% respectively.