Research On Traffic Sign Detection And Recognition Algorithm In Natural Scene

Traffic signs play an important role in traffic regulation and accident avoiding.Although traffic signs are designed to be striking in color,drivers are not always able to notice traffic signs in complex traffic scenarios.Moreover,during driving,the car needs to continuously collect and process the surrounding environment information.Therefore,the accurate detection and recognition of traffic signs is of great significance in the intelligent driving system.In recent years,traffic sign detection and recognition has become a hot research topic,and there are certain research results.The research results have positive reference for other object detection and recognition.In the image,traffic signs belong to small objects.Therefore,in the actual driving environment,it is still very difficult to accurately locate and identify traffic signs.This paper mainly focuses on the prohibition sign in traffic signs as the research object,and proposes two different methods based on vision,which uses different features to detect and recognize traffic signs.First of all,based on the unique features of traffic signs,this paper proposes the first research method of this paper,which is based on super resolution.The method first segments a red region from the image based on the specific color of the traffic sign.The red region can be segmented from the image.Then the method combines the inherent shape and the size of the traffic sign to locate the location of the traffic sign from the image.In the recognition stage,super resolution technology is introduced to improve the image quality of small objects.After image reconstruction,HOG+SVM approach is used to classify traffic signs.Finally,the feasibility of the method was verified by experiments.The results show that compared with the method without introducing super-resolution technology,the proposed method can effectively detect small targets,and has a certain effect on the detection of large targets.In addition,this paper starts from the deep learning object detection and recognition algorithm,and finds that the current general object detection and recognition network is not very good for small objects.Aiming at this problem,this paper proposes the second research method,namely the method based on attention model.The method transforms the detection and recognition problem of small objects into the detection and recognition of large objects by using the attention mechanism,which can solve the small object recognition problem well.This method mainly generates the area containing the object by using the attention model based Faster RCNN.Then this paper uses the YOLO network to detect and recognize traffic signs in the generated attention area.Finally,the feasibility of the method was verified by experiments.The results show that the proposed method based on attention mechanism proposed in this paper is better than single YOLO network in detecting small targets.