Research Of Fatigue Driving Detection Method Based On Multi-Scale Feature Fusion

As one of the main factors inducing traffic accidents,fatigue driving causes about20% of the total traffic accidents and more than 40% of the mega traffic accidents,which is extremely harmful.The fatigue detection method based on facial features does not require any contact with the driver,and has the advantages of low cost,small limitation and high accuracy.How to detect driver fatigue quickly and accurately by facial features has important research significance and application value.At present,the fatigue detection method based on facial features fails to fuse the multi-scale features and multi-level information of images,and it is difficult to meet the requirements of both real-time and accuracy in the complex and changing driving environment.Therefore,based on practical application scenarios,this thesis aims to improve the accuracy,robustness and generalization ability of fatigue detection methods,and conducts in-depth research from multiple perspectives such as fatigue feature selection,facial feature extraction,and fatigue judgment,and proposes a method.The fatigue detection method of multi-scale feature fusion,combined with the requirements of the intelligent digital cockpit project,designed and implemented a fatigue driving detection system based on facial features.The main research contents and achievements of this thesis are as follows:Aiming at the problem that existing methods cannot fully utilize the multi-scale features and multi-layer information of images,this thesis proposes a fatigue detection method based on multi-scale feature fusion.The method uses a fusion strategy of multi-scale feature layers and a lightweight and efficient gating module to combine deep features with strong semantics and detail-rich shallow features to fully exploit the effective feature information in the image and suppress the invalid information,so as to better detect the fatigue feature state of the face.Combined with the multi-feature integrated fatigue assessment approach,the proposed fatigue detection method can reach an average accuracy of 96% on the video dataset,demonstrating that the method can accurately detect the fatigue state of the driver.A requirement analysis was conducted for the fatigue detection system,and firstly a fatigue detection scheme was designed based on a multi-scale feature fusion method.Secondly,a fatigue driving detection system based on facial features is implemented using the Pytorch framework,Py Qt5 and auxiliary hardware devices.The system can detect the driver’s fatigue status by means of real-time video capture by camera or pulling videos from the resource library.The off-duty detection module can be used to monitor whether there is off-duty phenomenon during driving and the corresponding warning function is designed for prompting.The final test results confirm that the system can meet the requirements of real-time,usability and ease of use,and can judge the driver’s fatigue state and give corresponding feedback.Its overall performance is good and has certain practical value.