Research On Traffic Sign Recognition Based On Deep Learning

With the popularization of automobiles,the assisted driving system has become a hot spot of increasing concern.The assisted driving system can obtain current road condition information,remind the driver of the road condition information that needs to be processed during the current driving process,and help the driver to avoid the risk of a car accident Many car accidents The cause of the occurrence is the driver’s on reasons,distraction,fatigue driving,slow response,wrong judgment of the road conditions.In order to correctly and quickly sense the road condition information,it is essential to obtain the information of the traffic sign.The accuracy of the traffic sign recognition will directly affect the driving safety.At present,many experts and scholars at home and abroad are committed to the exkploration and research in this field Improve the accuracy of traffic sign recognition.In order to accurately and quickly sense road condition information,it is essential to obtain information on traffic signs.Traditional traffic sign detection algorithms have many limitations,such as:can only classify a single category of traffic signs,are susceptible to light,weather factors,different angles of obtaining traffic sign images,and low detection rates due to target damage.In this paper,the problem of traffic sign classification is solved based on the deep learning method.Then,the deep learning target detection model based on the regional recommendation network is used to detect traffic signs under non-restrictive conditions,which is more suitable for natural scenes.The main works is as follows:(1)In view of the lack of unified traffic sign datasets in China,this paper collected traffic sign images under non-restrictive conditions,produced data sets and preprocessed and expanded them.Although there are unified traffic sign data sets in foreign countries,there are many differences between foreign traffic signs and domestic traffic signs,and most traffic signs occupy too large a area,and the background area is small and single.The datasets produced in this paper were taken under non-restrictive conditions,and the background was complex and varied,which was more in line with the actual situation.(2)Because the VGG-16 network model is complex,the network depth is deep,and the number of neurons is large,it is easy to over-fitting for small-scale digital image classification tasks.In response to one problem,this paper proposes measures such as cutting the original VGG-16 network model,reducing the complexity of the model,and reducing network parameters.The experimental results show that the improved VGG-16 network model not only can achieve classification tasks,but also has the advantages of short detection time and high accuracy.(3)Aiming at the problem of missed inspection of traffic signs with small area,this paper proposes a multi-layer feature fusion strategy.After convolution and pooling in the feature extraction network,many details of the image are ignored By combining the convolution features of the fourth and fifth layers in the feature extraction network,the ability of the network to express the overall information of the image is preserved,and the detailed features of the image are added,which further improves the accuracy of the traffic sign model.Figure[46]Table[7]Reference[53]...