Research On Traffic Sign Recognition Algorithm Based On Deep Learning In Complex Environment

As deep learning algorithms are applied in more and more fields,research work in the field of intelligent transportation systems is also developing and innovating.Traffic signs are an indispensable and important part of intelligent transportation systems.The faster the driving speed of the self-driving vehicle,the more the system needs to recognize and respond to traffic signs as soon as possible,which means that the recognition of longdistance small target traffic signs Research is imperative.The recognition of traffic signs is also affected by light,and the poor light will reduce the efficiency of traffic sign recognition.Therefore,the research on the detection and recognition of dim small target traffic signs in complex environments has important theoretical significance and practical value.This paper studies the detection and recognition technology of traffic signs.The main research contents are as follows:This paper uses the TT100 K dataset in COCO format for experiments.Due to the impact of the size of the data volume on the performance of the deep learning model,this paper uses data enhancement technology to expand the data set.Aiming at the problem that traffic signs are affected by natural light changes and weather factors in the actual traffic environment,especially traffic signs under backlight conditions,this paper uses HDR image-based dark light enhancement technology to process them.This paper proposes an improved dim small target traffic sign recognition algorithm based on the YOLOv5 model.First,a small target detection layer with larger scale features is added to the Neck module of the YOLOv5 model.The small target detection module imitates the combined structure of the feature pyramid network and path aggregation network in the Neck module is used to retain as much shallow information as possible to obtain richer feature information,which is conducive to improving the detection efficiency of small target traffic signs.Secondly,adding the coordinate attention mechanism in the Backbone module is beneficial to solve the problem that the image features of traffic signs in complex environments are not obvious,thereby improving the efficiency of feature extraction.The experimental results show that the improved model in this paper has greatly improved the efficiency of traffic sign recognition for small targets and dim conditions.Compared with the original YOLOv5 model trained on the expanded data set,the improved YOLOv5 model in this paper has a 1.5% increase in accuracy,reaching 93.4%;a 6.8%increase in recall rate,reaching 92.3% %;the m AP value increased by 5.2%,reaching96.2%.Using the improved YOLOv5 model for dim small target traffic signs studied in this paper,a traffic sign detection and recognition system is designed.The recognition system includes functions such as uploading images or video files,performing dark-light enhancement processing on image files,turning on the camera for real-time monitoring,detecting and identifying the type of traffic signs,and displaying the recognition results.The traffic sign detection and recognition system is developed based on the Python programming language.Py Qt5 is used to create the GUI interface of the traffic sign detection and recognition system,and the traffic sign detection and recognition model trained in this study is called to upload files or open The camera images are used for detection and recognition tasks.The effectiveness of the algorithm proposed in this paper is further verified by applying it on this system.