Improved GoogleNet Model for Tired Driving Detection

Fatigue driving accounts for one of the main influencing factors in global traffic accidents. In this study, in order to improve the training time and accuracy requirements of fatigue driving detection model, we optimize the design of GoogleNet-v3 model, and propose to introduce the pruning algorithm and GMP idea into the recognition model of GoogleNet-v3 model, and adopt a specific training strategy, which is derived from the experimental simulation data, and the model of this paper's judgement accuracy rate for the data test set used in this paper reaches more than 90%, which significantly improves the accuracy of fatigue detection; the training time in processing images is 203ms, which reduces the time consumption by more than 50% compared with the existing techniques. Therefore, the model proposed in this paper can be suitable for real-time fatigue monitoring.