Research And Implementation On Traffic Signs Detection Based On Intersection Scene

With the increasing popularity of transportation,people are paying more and more attention to the issue of traffic safety.Intelligent traffic assistance systems have gradually entered people's horizons and played a significant role.As an important part of the intelligent transportation system,the intersection traffic sign detection technology has been given full attention and extensive research by the academic and industrial circles.At present,the difficulty of traffic sign detection technology is that the accuracy and real-time of detection are difficult to support complex and fast changing traffic scene at intersections.The traditional methods mostly use the method of extracting features such as color,texture and edge and then using classifier to predict.This method is difficult to meet the requirements of the project in the detection accuracy.In order to achieve good real-time performance while ensuring good detection accuracy,this paper aims at the actual traffic intersection scene,using the popular convolutional neural network detection research of traffic signs,and design a practical application of embedded intelligent auxiliary system.The main contents of this paper are as follows:First,this paper studies a fast intersection recognition algorithm based on multi-resolution convolutional neural network.On the basis of the network that extracts the global characteristics of the intersection,the feature network is added to the feature of the local target area under the intersection scene separately.The local target information fully contributes to the scene classification,and the richness of the feature is increased,making the classification more accurate.Second,this paper studies a multi-level traffic sign real-time detection algorithm.CNN is used to detect traffic signs at different scales of receptive field.A more rapid feature extraction and fusion network is designed for the detection targets,and the method of splitting and convolution is used to accelerate traffic detection so that a good real-time detection effect can be achieved.Third,this paper studies a minitype traffic sign detection algorithm based on candidate regions.Aiming at the problem that the regression model has a low target recall rate,this paper introduces a miniature traffic sign detection algorithm based on candidate regions,generates candidate regions using superpixel segmentation in the lower eigenvalues,and uses the grid corresponding to the candidate regions to predict the category information.Making the mini-traffic sign recall rate greatly improved.In order to validate the effectiveness of the proposed algorithm,this paper constructs a traffic sign detection database for urban intersections in China and designs an intelligent support system based on embedded terminals.The experimental results show that the proposed algorithm has good performance in traffic sign detection at intersections.