Research On Object Detection Technology For Rainy Traffic Scene Images Based On Deep Learning

During the vehicle driving in rainy days,the presence of rain patterns will make the target objects in the traffic scene blurred and invisible,which seriously affects the perception and judgment of the driver and the vision sensor of the assisted driving system on the traffic scene.In addition,the complex background information in the rainy weather traffic scene also increases the potential risk of traffic accidents.Therefore,it is extremely important to carry out the research of real-time target detection technology for rainy weather traffic scenes,and the continuous development of deep learning technology provides technical support for the implementation of target detection for rainy weather traffic scenes.Based on this,this paper conducts research on the deep learning-based target detection technology for rainy weather traffic scene images.The main research is as follows:(1)The rain removal method for images of rainy traffic scenes is studied.A detail feature restoration branch-based image rain removal algorithm is proposed to restore the image details while the rain pattern removal operation generates loss of image details using the detail feature restoration branch.The dense attention structure in the rain removal sub-network and the context-dilated convolutional fusion module in the detail feature restoration branch network allow the algorithm to combine both rain removal and image detail preservation restoration,so the algorithm achieves better results in image rain removal compared with other algorithms.(2)A method for target detection of rainfall traffic scene images is studied.A multi-scale feature fusion-based target detection algorithm for traffic scene images is proposed,and the extracted multi-scale feature information is reassigned channel weights and fused by using compressed incentive path aggregation network,so that the accuracy,precision,recall,F1 value,AP,and mAP of each target detection and localization of traffic scene images are improved,and the accurate detection and localization of each target are realized.(3)The software system of rain traffic scene image target detection was developed and verified by example.The functions and interactive interface of the software are designed and the integration of the methods and algorithms studied in this paper is realized.Using the rainy day traffic scene of a road section as a background example,real-time target detection and target ranking are performed to verify the reliability and practicality of the proposed algorithm.