Study On Traffic Signs Detection And Recognition In Natural Environments

With the rapid development of auto industry in our country and the putting forward of wisdom city concept,on the one hand,the number of private cars present explosive growth and then have brought serious challenges for city traffic and safe driving,on the other hand,the putting forward of wisdom city concept,one of the goals is to use the advanced information technology to solve the problems of the safety of public transportation.Therefore,the road traffic sign detection and recognition system,as one of the most important modules of future intelligent transportation,is becoming more and more important,and need to be further in-depth study.The main research object of this thesis is traffic signs in the natural environments,in this thesis,the research work can be divided into two parts,namely,traffic signs detection and traffic signs recognition.In the part of traffic signs detection,due to the different natural environments and the diversity of the collected traffic signs under driving condition,such as rain and snow,fog,haze weather and traffic signs of distortion,deformation,shade,etc.,which have brought serious challenges for traffic sign detection.Therefore,in this thesis,we employ a traffic signs detection algorithm which is based on dual characteristics of traffic sign color and shape.First of all,we segment traffic signs based on the HSV color space segmentation algorithm,then the segmentation results are morphological processed.Finally,we use the algorithm based on modified Hough transform for the detection of traffic signs.This algorithm can overcome the adverse effects of weather and illumination changes.In the part of traffic sign recognition,on the basis of detection,this thesis presents a novel and efficient traffic sign recognition algorithm based on fusion optimization of modified weighted ELM and AdaBoost.Firstly,the algorithm determines the modified weighted ELM input bias weight matrix and the hidden layer neurons by setting up a simple and effective approach,then employs the modified weighted ELM as the weak classifier by updating the training weight of original ELM iteratively.Finally,an optimal strong classifier is constructed by the weighted majority vote of all the modified weighted ELMs.Conducted on the famous German traffic sign recognition benchmark(GTSRB),the final experimental results show that the proposed algorithm can achieve a totaltraffic sign recognition accuracy of 99.12%,and only need 7.1ms to recognize a single traffic sign which can meet the demand of real-time recognition applications.Therefore,the proposed algorithm efficiently improves the recognition performance of traffic signs.