Research On Algorithms Of Traffic Sign Detection And Classification In Natural Environment

As an important traffic auxiliary facility,traffic sign is used to regulate traffic,warn drivers and provide useful information to help make driving safe and convenient.They play an important role in establishing well-ordered traffic.In order to convey the traffic information timely and exactly,researchers pay more attention on traffic sign recognition system(TSR).Traffic sign detection and classification are two key components of TSR.There have been some outcomes in both theoretical research and practical systems during the past decades.However,there are some challenges for traffic sign detection.First,traffic sign exposing in outdoors will suffer color degeneration or shape deformation.Second,the color and shape of traffic signs maybe changed in the process of imaging.Third,in the complicated traffic scenes,traffic signs will be covered by other objects and results in sign occlusion.Traffic sign classification is another key component of TSR.There are many kinds of traffic signs,therefore,the classification of traffic signs is an example of multi-classification.The robustness and effectiveness of classification algorithm is still a hot issue which has not been solved very well.In this paper,we pay our attention on image preprocessing,traffic sign detection and traffic sign classification to solve some problems mentioned above.(1)Two color processing methods,color distribution model and improved color contrast model are proposed to enhance the traffic sign regions in the image according to their color features.By computing the distribution probability of the predominant color of road signs in Lab space,the corresponding feature maps of the input image can be obtained in the color distribution model.The improved color contrast model enhances the red,blue and yellow regions in the image according to the color opponents existing in human vision mechanism.The two models are compared with other popular color processing methods in traffic sign images.The results show that our improved color contrast model obtains the highest detection rate with the minimum time.(2)A fast traffic sign detection method based on rotational symmetry voting is proposed after some researches on existing fast detection methods.The proposed method employs edge gradient as image feature,and chooses the points which meet a special symmetry angle to vote to their symmetry centers.Then regular polygons in the image can be detected by finding voting centers and the boundary points which vote to the centers,followed by polygon shape classification.The proposed method is sample in time complexity.Experiments show that the average processing time for an image is 55ms,which satisfies the requirement of real-time application.(3)A regular polygon detection method called link distribution(LD)is proposed after detailed research on existing methods about shape detection of traffic signs.The method can solve the problems of partly occlusion and shape rotation.The LD model considers the shape contour as a collection of links from center to boundary points.Every link is represented by three parameters,that is length of the link,inclined angle of link and edge orientation of boundary point.The model can tolerate large shape variance because the order and the adjacency relation of the links stay invariant during shape deformation.Experiments in public data set show that the method is effective in detecting traffic signs.The detection rate of prohibitive,warning and obligation signs are 98.63%,95.24%and 94.40%respectively,which are higher than most advanced methods.(4)A method based on visual saliency is proposed to detect the road signs in complex scene.The method combines the bottom-up and top-down saliency together to detect road signs.The bottom-up salient region detection method begins with cluster division and image segmentation,the saliency of a region is evaluated by computing its contrast with other regions in the image and the relative spatial compactness of cluster it belongs to.Furthermore,each type of traffic sign has special color features which can be used to construct class-specific distribution,that is the top-down saliency maps.The traffic signs are detected from the final saliency maps,which are generated by combining the bottom-up saliency maps with top-down saliency maps.Experiments on real world images show a high success rate and a low false hit rate and demonstrate that the proposed framework is applicable to prohibitive,warning and obligation signs.Additionally,the proposed method can be applied to achromatic signs without extra processing.(5)A coarse-to-fine traffic sign classification method is proposed according to features of traffic signs.The traffic signs of China can be divided into five categories according to their color and shape,which are prohibitive signs,warning signs,obligation signs,release of prohibitive signs and other signs.In the first stage,a RoI(Region of Interest)is represented with HOG and CN descriptors to describe its shape and color features.A linear SVM classifier is used to classify the Rol to corresponding category.In the second stage,the different fusion methods of color and shape features in BoW model are discussed and the color-shape early fusion method is employed to describe RoIs.The final class labels of Rols are obtained by Guass kernel SVM classifier.Experiments show that the proposed method achieves a high classification precision of 99.15%,which outperforms human performance and is the second best one among all the public results.