Research On Driver’s Mobile Phone Behavior Detection Algorithm Based On Yolov5

Drivers using mobile phones is a dangerous driving behavior that affects their attention and responsiveness,increasing the risk of traffic accidents.Therefore,detecting whether drivers are using mobile phones can help traffic management departments better monitor driver behavior,reduce the incidence of traffic accidents,and ensure the safety of public roads.There are many detection algorithms for detecting the behavior of drivers using mobile phones,but due to small mobile phone targets,long detection distances,and the impact of natural environmental factors,the detection accuracy is low.In this paper,the following research has been conducted on driver’s mobile phone behavior detection algorithms:(1)Aiming at the problem of small mobile phone targets,which are difficult to capture and locate due to their low proportion in the image,and are prone to low detection accuracy due to factors such as illumination,occlusion,and angle,this paper proposes an improved Yolov5 driver’s mobile phone behavior detection algorithm.By introducing an improved attention mechanism module into the Yolov5 backbone network,we can better obtain contextual information,reduce the loss of semantic information,and better improve the accuracy of small target detection.Secondly,an improved feature fusion method is used to extract features of three scales and fuse them to better extract local information.The experimental results show that compared with Yolov5,the accuracy of the detection algorithm on the self-made dataset reaches 71.9%,which is improved by 2.1%.The detection effect on small targets is significant.(2)Under foggy conditions,the contrast and clarity of the target are reduced,the edges are blurred,the scale and shape of the target are prone to change,and the overlapping phenomenon leads to the degradation of image quality,resulting in the loss of detailed information,which poses great difficulties to the detection algorithm.Therefore,this paper proposes a defogging target detection method based on improved Debr GANv2 and Yolov5.This method preprocesses the foggy images with poor quality through the improved Deblur GANv2 module to obtain clear defogging images.The processed images are used as input to the Yolov5 network.At the same time,this article also makes some improvements to the Yolov5 algorithm.The improved Deblur GANv2 module uses multiple scale convolution kernels to extract image context information,thereby better recovering image details such as edges and textures.The improved residual module can better extract network features,reduce the computational complexity of the model,and accelerate the training and convergence speed of the network.The improved loss function is more accurate in regression prediction and can quickly converge during model training.At the same time,the proposed method can maintain high robustness in the case of small data sets and data noise.The experimental results show that he accuracy of the detection algorithm on the self-made dataset reaches 83.33%.