| As an important part of intelligent traffic system,traffic sign detection and identification has a broad application prospect in the fields of assisted driving and unmanned driving.In recent years,with the increasing number of vehicles,the traffic environment is increasingly complex,in order to ensure the safety and smooth road travel,traffic sign detection and identification has become particularly important.However,it is difficult to predict the damage caused by external factors such as weather and human,so the traffic sign detection and recognition algorithm still has great challenges.How to detect and identify the mark accurately and efficiently is a practical problem at present.In this context,this paper uses the theoretical knowledge of deep learning to study the detection and recognition of traffic signs.Through the improvement and optimization of the algorithm,the accuracy and real-time performance of the traditional method are improved.The main research contents and results include the following two parts:In view of the traditional detection method of poor real-time,high complexity,and the existing detection framework based on convolutional neural network of small targets detection problem such as the effect not beautiful,through to the SSD(more than a single frame detection)studied the model,put forward an improved algorithm of convolution network model based on SSD,algorithm by adding the lower figure,rich semantic information,the small size targets and multi-layer fusion characteristic figure,implement different characteristics of the layer of multivariate information classification regression prediction and position,improving model of traffic signs of small target object detection effect;K-means clustering algorithm is used to analyze the selection of SSD default box values,adjust the size ratio of original feature boxes,accelerate the convergence of the model,and improve the accuracy of the algorithm.The model was verified by two different data sets of GTSDB and CCTSDB,and compared with other models.The results show that the algorithm presented in this paper shows high accuracy and meets the demand of real-time.In view of the complex network structure and time-consuming network training and recognition process of the existing traffic sign recognition algorithms,this paper proposes an improved traffic sign recognition method based on lenet-5 model on the basis of previous studies.Two convolutional layers are added on the basis of lenet-5model,and the network adopts the structure of two convolutional layers and one pooling layer to fully extract the local detail features of the image and improve the recognition ability of the network.The dropout operation was added after the pooling layer to make up for the deficiency of the classic lenet-5 network during the traffic sign recognition training.Change the sigmoid activation function to Re LU to prevent gradient dissipation.The training set in GTSRB was used to train the improved network,and the recognition rate and recognition time of the trained model were compared with other recognition methods.Experimental results show that the improved network model has certain advantages over other classical network models in terms of accuracy and real-time performance. |
