Fig. 1. YOLOv5s model structure

Fig. 2. The visualization results of each image enhancement method. (a) Original image; (b) LIME; (c) EnlightenGAN; (d) SCI

Fig. 3. StemBlock module

Fig. 4. ShuffleNetV2 module. (a) Basic module S_block1; (b) subsampling module S_block2

Fig. 5. Bottleneck module and inverted bottleneck module. (a) Bottleneck module; (b) inverted bottleneck module

Fig. 6. The structure of CBAM

Fig. 7. CIB module

Fig. 8. CIBC3 module

Fig. 9. Loss function image with different parameters

Fig. 10. Improved YOLOv5s model

Fig. 11. Result instance

Fig. 12. Schematic diagram of eye area

Fig. 13. Schematic diagrams of an image capture of the eye and mouth. (a) Cropped image of the eye region; (b) cropped image of the mouth region

Fig. 14. The structure of FSR-Net

Fig. 15. Fatigue driving detection model

Fig. 16. Examples of dateset

Fig. 17. Examples of fatigue driving dataset. (a) Examples of the YawDD public dataset; (b) examples of the self-built dataset

Fig. 18. Examples of image enhanced self-built dataset

Fig. 19. Examples of fatigue detection results. (a)‒(d) YawDD public dataset detection results; (e)‒(h) self-built dataset detection results