Research On Traffic Sign Detection Method In Rainy Weather Environment

Traffic sign detection in autonomous driving technology is one of the current hot research directions.However,extreme weather conditions will have a certain degree of impact on the detection effect.For example,rain will block the captured image,thereby affecting the judgment of traffic signs.Therefore,the study of traffic sign detection in rainy weather environment has far-reaching significance for the development of modern transportation.The paper uses deep learning technology to conduct in-depth research on image rain removal tasks and traffic sign detection tasks,analyzes the current difficulties and challenges of the above tasks,summarizes the shortcomings of existing methods,and improves the existing problems.From the perspective of ease of testing,an experimental platform based on the Web is built.The main research content of this paper is as follows:(1)Rain can cause visual obstruction to clear background images.Existing rain removal methods have the drawbacks of insufficient rain removal and large models that are difficult to deploy to vehicle systems.To address the above issues,an adaptive lightweight image rain removal method is proposed,and an adaptive multi-scale feature extraction block is designed to alleviate the drawbacks of easy loss of feature information;A cross-channel deformable mixed attention mechanism and low parameter residual dense blocks are designed to alleviate the drawbacks of high memory requirements due to large parameter quantities during training;Considering the fixed distribution pattern of current synthetic rain streaks datasets,a rain streaks enhancement method is proposed to narrow the gap between synthetic rain images and real rain images.Compared with several other methods of the same type,the proposed rain removal method achieves better performance on four datasets while satisfying real-time requirements.(2)Aiming at the problems of low accuracy and missing detection in existing traffic sign detection methods,a traffic sign detection method based on improved YOLOv7 was proposed.A sparse shifted-window transformer block is designed to improve detection performance by introducing a sparse multi layer perceptron and a shifted-window transformer;A omni-dimensional efficient layer aggregation network is designed,which assigns convolution dynamic attributes from four dimensions to better focus on feature information;In addition,in view of the high requirements for deployment equipment in the current reasoning process of traffic sign detection tasks,a multi-path structured re-parameterized block is designed.During the training process,network features are fully learned through a multi branch structure,and it is converted into a single path structure during the reasoning process.Experimental results on the CCTSDB datasets show that the proposed traffic sign detection method achieves better performance and meets real-time requirements compared to existing methods.(3)The reasoning process of most existing deep learning tasks is implemented based on the command line method,and the operability is poor.Therefore,a web-based experimental platform has been built,which can effectively combine the rain removal model and the traffic sign detection model,and the web-based interactive page is easy to operate.