Fatigue driving behavior recognition based on attention mechanism and dual flow network

Aiming at the problems of low accuracy and poor consistency in fatigue driving behavior recognition, this paper proposes a fatigue driving behavior recognition method using attention mechanism and dual flow network. Firstly, this method processes facial images through a FaceNet network, where the channel attention module within the attention mechanism is used to accurately identify key regions for feature detection. Then, extract fatigue behavior characteristics based on the angles of eye closure and head posture changes. Finally, based on the characteristics of fatigue behavior, a fatigue recognition model combining attention mechanism and dual stream network was developed. The test results indicate that provided method successfully maintains the mean square error (MSE) below 3.5. In addition, it also achieved high Pearson correlation coefficient (PCC) and consistency correlation coefficient.