Research Of Traffic Sign Detection And Recognition Based On Saliency And Convolution Neural Network

With the rapid development of science and technology,the vehicle has become an essential means of transport in people’s daily life,of course.A series of traffic problems followed come because of the increasing number of vehicles,such as traffic safety,traffic congestion,traffic pollution and so on.In this different background,Advanced Driver Assistance System(ADAS)comes into sight.As a basic branch of ADAS,Traffic sign detection and recognition is an important means to improve traffic safety and efficiency.Therefore,the paper carries some related research on traffic sign detection and the recognition.Firstly,by analyzing the principle of saliency detection and combined with the feature of traffic sign,we propose a traffic sign detection method based on salient features,and verify the method through experiments;Secondly,through the analysis and research of convolutional neural network,the paper put forward a traffic sign recognition method based on improved Alex Net convolution neural network.The main contents and achievements of this paper include the following aspects:(1)Propose a traffic sign detection method based on saliency.First,establishing a saliency model with three visual features of traffic signs,three features are color,border and location information;Secondly,establishing a rule to fusing salient feature,and combining with the minimization of a cost function,we obtain the final optimal saliency map by fusing the features of input image;Finally,through binarization of the optimal saliency map,we obtaine and mark the connected regions in the binary Image.Then we map the connected regions to the original RGB image and using sliding window method to extract the region of interest,achieving the detection of traffic signs.The experimental results show that the proposed algorithm is suitable for the detection of traffic signs in complex environment.By comparing and analyzing with 4 commonly used saliency detection algorithms,It is proved that the proposed algorithm in this paper has higher detection performance than other algorithms.(2)Propose a traffic sign recognition based on improved AlexNet convolution neural network.The method firstly analyze the structure of Alex Net network model,then a new Alex Net model is obtained through adjusting and optimizing the structure and parameters of Alex Net model;then we can identify traffic signs using this new network.Using AlexNet model to realize traffic sign recognition mainly includes two parts,one is the training of Alex Net model;the other is achieving the classification of the input by use the trained Alex Net model.(3)Propose a method of expanding the training data set.The proposed Alex Net model is trained and tested by German traffic sign recognition data set(GTSRB).Due to the imbalance of GTSRB training samples,we propose two sample expansion method to improve the data set.The experimental using the extended data set and the original data set to train and test the proposed AlexNet model.Results show that,using the extended training sample set to train the Alex Net classification model for traffic sign recognition,most categories of the traffic signs in the test set can reach more than 95% of the traffic signs recognition accuracy,which is higher than the original training set’s 93%.By comparing the proposed AlexNet with LeNet convolution neural network and the classical“Hog + SVM”classifier,It is proved that the proposed method is superior to the other two methods in both identification accuracy and time complexity.