Research On Visual Detection Technology Of Urban Road Traffic Information In Mountainous Areas Under Adverse Weather Condition

The visual perception technology,as a necessary means for vehicles to achieve precise target detection,tracking,and traffic information recognition,has become a research hotspot in the field of autonomous driving.In complex and changeable mountainous urban scenes,adverse weather conditions such as rain,fog,and low-light will decrease the accuracy of visual environment perception technology in obtaining surrounding road traffic information,increasing the driving danger of intelligent connected vehicles,and even causing traffic accidents.Therefore,researching how to perform stable traffic information visual detection technology in mountainous urban road scenes under adverse weather conditions has important theoretical and practical application value.This paper is mainly focuses on the research of road traffic information visual detection technology in rainy weather scenes in mountainous cities.The specific research work and characteristics of this paper are as follows:(1)To solve the problem of lack of comprehensive images data sets of traffic information in mountainous urban areas,an image collection platform will be established,and video of complex scenes including rain,fog,night,and tunnels in Guizhou Province are systematically collected to establish the comprehensive data set.The images are classified based on the scene,and all visible lane lines in the image are calibrated and preprocessed.This study has enriched the public data set of diversified scene image datasets and has provided data support for subsequent research on traffic information detection.(2)A raindrop image enhancement algorithm that integrates multiscale feature information is proposed to address the adverse effects of raindrops on image background,such as distortion and blurring.The Raindrop Shape-Driven Attention Module is used to capture raindrops,while the Spatial and Channel-Coordinated Attention Mechanism is introduced to enhance the weight of important spatial and channel features.Additionally,a Novel Dilation Convolution Pooling Pyramid Module is designed to capture multiscale features in the image.Quantitative experiments show that this algorithm achieves the PSNR value of 30.75 and the SSIM value of 0.9254 on the public dataset,and also performs well in removing raindrops and restoring background image details in the self-built rainy image dataset,providing reliable image data for subsequent lane detection studies.(3)Based on the raindrop image enhancement algorithm,a road traffic information detection method is studied for rainy conditions in mountainous urban areas.Taking lane detection as an example,a lane detection algorithm based on attention residual mechanism is proposed to solve the problem of inadequate detection accuracy in rainy scenes.The SENet network and ResNet structure are combined to improve the basic unit of the lane detection network,and the raindrop image enhancement algorithm is used to preprocess image data as the training set of the network.Experimental results show that the detection accuracy of this algorithm on the self-built rainy image data set and the sunny public data set is 86.59% and 96.67%,respectively,proving that this algorithm has significant effects.(4)A lightweight lane detection network is proposed to solve problems such as complex structure and large number of parameters in the lane detection algorithm.First,the number of hourglass modules in the detection network is reduced and the network structure is simplified.Then,Depthwise Separable Convolution is used to replace standard convolution in the size adjustment module to reduce the number of parameters.Finally,the module unit is improved by using the ECANet network and ResNet network structures.The experimental results show that the detection network with only 1.64 MB of parameters can achieve a detection accuracy of 93.86% and 96.62% on the self-built multi-scene image data set and the sunny public data set,and the algorithm processing speed reaches 38 fps,demonstrating the effectiveness of the lightweight lane detection algorithm.By establishing a mountainous urban road traffic image dataset under adverse weather conditions,we can enrich the diverse scene image data in the field of traffic information detection.The design of raindrop image enhancement algorithms and lane detection algorithms provides theoretical support for solving the problem of lane detection on mountainous urban roads under rainy weather conditions.