Research On Traffic Sign Detection Method Based On Single-stage Multi-anchor Bo

The detection and recognition of traffic signs is a key link to realize autonomous driving and driver assistance,which is of great significance to the safe driving of cars and the construction of smart cities.In autonomous driving and driverless systems,the traffic sign objects reflected in the driver’s line of sight are usually small,so there is still a possibility for intelligent traffic systems to be optimized in detecting them.In complex weather conditions,the weather environment in which the vehicle is located can have a great impact on the decisions made by the driving system.The recognition system can detect the correct road signs in time and assist the driver and the system in making judgments,thus reducing the incidence of violations and major accidents on the road.Based on this,the research of this thesis is as follows:(1)This thesis presents a traffic sign detection method based on GR-CNN network to solve the problems of single Shot Multi Box Detector(SSD)algorithm,such as low accuracy,high miss rate and large model complexity.First,the GDW-Res Block module is introduced in the SSD network to enhance the feature extraction ability of small targets and reduce the model complexity.Then,an efficient and disordered attention ESA module is added at the end of each GDW-Res Block module to eliminate the loss of features caused by invalid information and pool operations in the image.Finally,LI-FPN,a feature fusion module enhanced by local information,is added to aggregate feature information at different scales.This method has a recognition accuracy and real-time detection speed of 9.54% and 1.64% higher than the benchmark model for small traffic signs,respectively,meeting the demand for real-time accurate detection.(2)Aiming at the problems of low detection accuracy,complex background,noise and model complexity in small traffic sign images taken in complex weather by existing algorithms,this thesis presents a complex traffic sign detection method based on MABF-CNN network.In this thesis,attention-guided ADBlock image noise reduction module is used to remove noise from the input complex weather-style image,then the inverse residual structure of Mobile Net V2 network is used to extract feature information efficiently with the attention mechanism MA module based on mobile network,while maintaining the lightweight of the model.Then,the bidirectional feature enhancement module BF-FPN is used to complement the geometric and semantic information of deep and shallow features.Finally,the loss function of classification and location is optimized.The recognition accuracy and real-time detection speed of this method for small traffic signs in complex weather conditions are 9.35% and 1.88%higher than the benchmark model,respectively,which helps to achieve real-time and accurate recognition of traffic sign indications ahead.