Research On Kernel Recognition Method Of Traffic Sign

With the rapid pace of urbanization in China and the significantly increasing of popularity of vehicles, the number of people in traffic is increasing in a great pace. It is difficult to meet the carrying capacity of the road to the heavy traffic, thus road congestion increases, traffic accidents happen more and more, road safety and transport efficiency get more and more attention. ITS (Intelligent Transportion System, ITS) system, the Intelligent Transportation System, is considered to be an effective means to solve these problems. ITS system is an integrated system with computer science, control science, detection, technology and communication technology. TSR (Traffic Sign Recognition, TSR) system is part of ITS system as an important subsystem. TSR system is a smart system based on digital image processing and detection technology, through the vehicle of information and intelligence to address driving safety and urban traffic congestion problems. The current study for the TSR system has become a hot research at home and abroad. Currently, the existing traffic sign recognition algorithm is different, but there are some drawbacks. Therefore, this paper presents some new algorithms and ideas to make up for these shortcomings, and to make contributions to the TSR system development and practical application.Traffic sign recognition system consists of cameras mounted on vehicles to capture the traffic signs in traffic scene, and then through the computer to complete the detection and identification. Due to natural conditions's instability and the complexity of the traffic signs, there are two main difficulties in the traffic sign recognition: 1) the natural environment for the traffic sign image quality is unstable; 2) the establishment of traffic signs gallery. The thesis of the above two points, combined with China's actual conditions, traffic sign recognition in the classification algorithm to do the following three areas:First, based on the existing traffic sign recognition and classification algorithm, a multi-layer traffic sign recognition model is proposed. Combined with road traffic sign colors - the geometric model, the present model is used for the recognition part of the TSR. The model is simple easy to implement, and has real-time characteristic.Second, the current classification and recognition algorithm of traffic signs are lack of testing in the natural environment. Because of that, we took a deep study of the neural network in traffic sign recognition, and combined it with traditional classification algorithms. The final algorithm is flexible when used to identify the model. Experimental results show that intelligent classification algorithms combined with traditional classification methods are easy to achieve, and the identify success rate is high.Finally, we analyze the image feature extraction and the different images of the description and characteristics. The final solution is of the Hu invariant moments and images of other regional characteristics. The method combines geometric features, for different types of traffic signs, and the characteristics of different value. This approach optimizes the relatively simple model of the traditional feature extraction. The multi-features traffic sign recognition algorithm can improve the success rate of classification.This dissertation in the final chapter, summarizes the work of the full text, and proposed follow-up research ideas, guidance on further research has significance. Finally, the research thesis by the application in real life also discussed.