Research On Traffic Sign Detection And Recognition Algorithms In Digital Images

Traffic signs contain a large amount of road condition information,which provide an important guarantee for the safe and orderly driving of drivers.With the development of technology and demand for application,traffic sign detection and recognition system plays an important role in advanced driving assistant systems and autonomous driving research.Traffic signs can be divided into many categories according to their functions,and each category can be further divided into subclasses with similar generic shape and appearance but different details.In view of the shortcomings of existing traffic sign detection and recognition algorithms,we propose a traffic sign detection algorithm based on cascade convolutional neural network and a traffic sign recognition algorithm based on multi-feature fusion.The main contributions are as follows:1.Traffic sign detection is different from general object detection because traffic signs normally account for less than 5% in the whole image.And in complex natural scenes,there are many interference factors,such as low intensity,motion blur,snow and other inclement weather conditions,occluded or damaged traffic signs,etc.,which have brought huge challenges for traffic sign detection.In this paper,the proposed algorithm realizes the coarseto-fine detection and has good detection performance for traffic signs of different sizes in the images.The detection network consists of three cascaded convolutional neural networks.Firstly,the traffic signs of different sizes can be obtained by multi-scaling.Then the multiscale images are fed into a full convolution neural networks to quickly filter out the nontraffic signs.And the remaining two neural networks are used for further filtering.Finally the traffic signs in the images are located.Moreover,the proposed detection network can be adapted to multi-scale images for training,which is more in line with the actual situation of traffic sign detection.The experimental results on several traffic sign detection datasets show that the proposed cascade neural network not only guarantees the recall rate,but also improves the detection accuracy.2.Based on the characteristic analysis of traffic sign shape and color,a recognition algorithm by integrating the edge features,texture features and color features of traffic signs and then combining with SVM is proposed in this paper.Firstly,HOG and LBP features of traffic signs are fused in a feature sparse way to reduce the dimension of feature vector and thus reduce the time complexity.Then,the color features of traffic signs are concatenated to be the final feature vectors.Finally,the feature vectors of the traffic sign images are fed into the support vector machine(SVM)for training to establish the traffic sign recognition model.The experimental results on two public traffic sign databases show that the traffic sign recognition algorithm proposed in this paper not only achieves the highest accuracy,but also has the lowest time complexity and the best system robustness.