Research On Object Detection Of Foggy Road Environment

With the continuous advancement of the modernization process,the demand for intelligence in daily life is increasing.Object detection in the road environment is crucial for the application of intelligent transportation system and autonomous driving technology.Because the absorption and scattering of particles in foggy environment reduces the visibility,greatly affects the clarity of the image,leads to a sharp decline in the object detection ability,and affects the practical application of intelligent products,it is very important to study the object detection method in the foggy road environment.A fog simulation method based on depth estimation was proposed,aiming at the lack of large foggy datasets in the foggy object detection method.The brightness and saturation were adjusted adaptively to preprocess the clear original image,self-supervised monocular depth mining network was used to generate the depth map and which was optimized by guided filtering.Transmittance map was obtained with setting the visibility of the simulated image,the dark channel map was used to distinguish sky area to estimate the atmospheric light value,and simulated foggy image with visibility was generated through the atmospheric scattering model.Based on the fog simulation technology research,a foggy object detection model based on YOLOX was constructed.The data set was generated according to the fog simulation technology,the data enhancement method of Mosaic and MixUp was adopted,and the label allocation of SimOTA and loss function was uesd in the YOLOX-based fog object detection model,so as to detect the object in the foggy scenario.The experimental data show that the fog simulation method effectively solves the problem of unreal image and object fog edge sharpening,and the effect is stable in the simulated visibility range.The average error rate of the simulated image and real fog image is 6.28%,indicating that the method is feasible.After the training based on the simulated data set,the detection result of the foggy object detection model is 11.61%higher than the original YOLOX network in the foggy detection accuracy,testing of different visibility shows a more significant improvement in object detection accuracy in dense and fog,and also improve the object detection effect in the real foggy scene,which is consistent with the test experiment..The comparison with the existing two algorithms further highlights the advantages of the foggy object detection model.