Traffic Sign Detection Method Based On Improved YOLOv5

With the continuous development of artificial intelligence technology,computer vision as an important branch has been widely used in intelligent traffic scene.In the real scene,in order to detect traffic signs accurately and in real time,the vehicle equipment needs to collect traffic sign images at a relatively long distance,which leads to the vast majority of the images collected at a long distance are small target traffic signs.However,the existing traffic sign detection methods have such problems as low detection accuracy,poor real-time performance and complex model in the detection of small targets.In order to solve the above problems,this thesis carries out research from two aspects of improving the performance of traffic sign detection and reducing the model complexity.The specific research content is as follows:(1)A multi-scale detection method based on feature enhancement is proposed for the problem of low accuracy of small target traffic sign detection,which is prone to miss detection and false detection.Firstly,the multi-scale prediction layer is improved,and a new prediction feature layer is constructed in the shallow layer of the backbone network.The feature layer preserves the detailed location information of small targets,which is more conducive to the detection of small target traffic signs.Secondly,more efficient SimSPPF module is used to replace the SPPF module of the backbone to improve the detection speed of the network.Finally,in order to further enhance the feature representation ability of shallow feature maps,a feature enhancement module is designed,which uses convolution kernels of different sizes to expand receptive fields to obtain contextual information,and transmits information through subsequent feature fusion paths to improve the detection ability of small targets.Through experimental comparison,it is found that m AP@0.5 of this method is up to 85.5%,8% higher than that of the original model YOLOv5,FPS is up to 86.2,achieving a good balance in detection accuracy and speed.(2)A lightweight detection method based on an attention mechanism is proposed to address the problems of large number of parameters and computation of current traffic sign detection methods.This method is improved from lightweight backbone network,introduction of attention mechanism,optimization of loss function and so on.First,a lightweight module in Ghost Net network is used to replace a common convolution module and a bottleneck structure in the previous network,which tend to reduce the number of network parameters and the amount of computation.Secondly,in order to alleviate the precision loss caused by the lightweight backbone feature extraction network,a lightweight attention mechanism is introduced to make the network pay more attention to important information and improve the detection performance of the network.Finally,the localization loss function of the network is optimized to accelerate the convergence of the model and improve the detection accuracy of the model.Through experimental comparison,it is found that compared with the original model,the number of parameters and calculation amount of this method are reduced by 26% and 33%,m AP@0.5 is up to 79.3%,FPS is up to 85.5,meeting the requirements of real-time detection.