Research And Application Of Detection Algorithm For Fatigue Driving Based On Improved YOLOv3

Aiming at the problems of large network model and low detection accuracy in the current fatigue driving detection model,this paper proposes a fatigue driving detection algorithm based on improved YOLOv3,and its feasibility is proved by experiments,At last Driving warning system and fatigue driving supervision system.is implemented based on the algorithm model.The main research contents and work of this paper are as follows:1.So as to speed up the convergence of the model and improve the detection accuracy,this paper improves the Anchor Boxes of the YOLOv3 algorithm and the regression loss function Io U.Firstly,the K-Means++ clustering algorithm is used in combination with the fatigue driving data set used in this paper to re-cluster to generate new Anchor Boxes to optimize the convergence speed and accuracy loss caused by different detection datasets.Through the experimental results we can know that the reclustered prior box can not only speed up the convergence of the model,but also improve the accuracy of the model detection.Afterwards,use the CIo U boundary regression loss function to replace the original Io U boundary regression loss function to optimize the gradient disappearance problem of the Io U regression loss function.Through the experimental results we can know that the improved loss function not only solves the above problems,but also improves the detection accuracy of the algorithm.2.A new fatigue driving detection model ESA-YOLOv3 based on the improved YOLOv3 algorithm is proposed.In order to solve the problems of large number of parameters and large amount of calculation in Darknet53,this paper uses EfficientNet to replace the original Darknet53 as the new backbone feature extraction network of ESA-YOLOv3 algorithm.Through the experimental results we can know that the detection model based on EfficientNet effectively reduces the number of parameters and the size of the model reduced by 69.7%;meantime,so as to further improve the detection accuracy of the network for fatigue driving data,this paper adds the backbone feature extraction network after improving the network.DBL×3 and SPP modules are added,and ECANet efficient channel attention mechanism is added after multi-scale feature fusion.The experimental results show that the fatigue driving detection algorithm of ESA-YOLOv3 based on SPP and attention mechanism effectively improves the network’s ability to capture key features,thus improving the detection accuracy by 1.85%.3.Based on the ESA-YOLOv3 fatigue driving detection algorithm and combined with Python,Pyqt5,and Django frameworks,this paper fully implements the driver fatigue driving supervision and warning system and the traffic control department’s fatigue driving supervision system,and finally passes the functional test to verify that it meets the basic requirements for use.