Research on Vehicle and Pedestrian Detection Algorithm in Foggy Weather Based on Improved YOLOv5s

Aiming at the problem of the bad performance of current automatic driving perception algorithm in foggy weather, this paper proposes a vehicle and pedestrian detection algorithm in foggy weather based on improved YOLOv5s. Firstly, based on Cityscapes dataset, artificially generate foggy images of three concentrations to form Foggy Cityscapes dataset. Secondly, C3STR structure is introduced into the backbone network to better improve the extraction ability of model for global features. Thirdly, de-weighted BiFPN is introduced into the neck network, adding a jump connection from backbone directly to PAN, strengthening the feature fusion and reducing the loss of feature information. Finally, EIoU is used in the boundary box loss function, which improves the regression precision and accelerates the convergence speed. The experiment results show that the algorithm proposed in this paper achieves good results in both synthetic foggy image dataset and real foggy image dataset, and meets the real-Time requirement, which has certain practicability.