Vehicleand Pedestrian Detection Algorithm Based On Improved AOD-net And YOLOv5 In Hazy Eenvironment

With the rapid development of science,technology and society,the number of vehicles with automatic driving function on the road is increasing.The reason why these advanced vehicles can realize driverless is that they are equipped with a large number of sensors and cameras for sensing the surrounding environment,and the images and road condition information collected by these devices,The vehicle detection system can accurately identify the targets on the road.However,in practical application scenarios,the camera as the "eye" of the car often faces more complex situations than expected,such as low light intensity at night and unclear camera imaging,resulting in serious loss of image details,or reduced visibility and blurred target contour in the field of vision in smog weather,As a result,the image acquisition quality is greatly reduced.These problems are likely to lead to the decline of the accuracy of the detection algorithm,so that the driving system can not effectively obtain the road condition information.Therefore,it is very necessary to overcome the influence of bad weather environment on image acquisition quality and ensure the accuracy of detection system in automatic driving task.Aiming at the problem that the haze in the automatic driving scene affects the imaging quality of the imaging equipment,resulting in the reduction of the accuracy of the detection algorithm,this paper proposes a lightweight dehazing algorithm based on AOD-Net.By used the MS-SSIM loss function,it not only improves the defogging effect and image quality,but also alleviates the color distortion.Aiming at the problem of low illumination encountered in traffic image acquisition,this paper adopts the method of integrating AOD-Net and Zero-DCE image enhancement network to enhance the low illumination image at night.Then,this paper makes a comparative experiment on two datasets,and uses multiple evaluation indexes to evaluate the algorithm.Aiming at the problems of traditional detection methods,such as unsatisfactory detection effect of vehicles and pedestrians,low detection rate,long reasoning time and insufficient accuracy,this paper proposes an improved algorithm based on YOLOv5.This method uses YOLOv5 s as the basic framework and refers to VOV-Net feature extraction network.By changing the backbone network structure,C3 module in the original algorithm is replaced by OSA module.the feature extraction ability of the backbone network is improved without increasing the amount of calculation.Aiming at the problem that small-scale target detection is not accurate enough,this paper replaces the PANet responsible for feature fusion in the original algorithm with a more efficient Bi FPN,which further improves the ability of the algorithm in multi-scale feature fusion.Then,this paper designed many comparative experiments to prove the effectiveness of the change on multiple datasets.Finally,this paper combines the improved AOD-Net with YOLOv5 method and carries out experimental verification on the real fog dataset.The results show that,compared with the original method,this method improves the accuracy of vehicle pedestrian detection in foggy weather by 4.2%,and has better performance in detection and recognition.