Recognition Of Traffic Sign Based On Multi-feature Fusion

Recent year with the increasing of the demand of car in domestic car market,car ownership continues to show a trend of rapid growth,the structure of transportation has changed fundamentally too.And it brought traffic problems,such as driving environment become more complex,road information is increasingly saturated and the pressure on the drivers is getting bigger and bigger.So advanced driver assistance system(ADAS)came into being.With the development of computer vision technology,as an important part of the advanced driver assistance system,traffic signs recognition system(TSR)has attracted much attention in recent years.Traffic signs as a necessary element of road construction plays a very important role,it is the basis for the driver to understand the road environment,it can prompt the driver current road information and potential danger while assist in traffic control.Thus,the driving safety is improved.However,if we provided all signs identified in the road environment to the driver,it will be a information overload to distract the driver and become a hidden danger of traffic accidents.The main content of this paper is divided into two parts: traffic sign detection,identification and traffic sign information filtering.This paper studies the problem of excessive information supply when traffic sign recognition system is used in actual use.Through the visual recognition evaluation of traffic sign,a friendly traffic sign recognition system is proposed.Using the visual evaluation value to filter unnecessary sign information to lighten the burden caused by the excessive information reminding.First of all,we pretreat the image collected by the camera,and establish HSI space model,then get precise thresholds by color characteristics of signs to segment the image.According to the shape characteristics of signs,determine the exact area of the sign by improved Hough transform.Secondly,using the improved Zernike moments to obtain the shape moment invariant characteristics on the basis of locating the exact area of the sign.Then,though the slightest distance classifier to classify the signs and the ultimate recognition result is achieved;Then,according to the influence of different factors in the road environment on traffic sign recognition,top-down and bottom-up visual models are used to build a multi feature fusion model for traffic signs recognition.The model is used to evaluate the recognition of the identified signs,and only feedback the information of the signs with low recognition,so it plays the role of information screening.Through studying the technology of the existing image processing method,to summed up the defects and deficiencies of various types of method,we improved it and put forward a traffic sign recognition system based on visual evaluation.According to the experimental data,proved that the characteristics of visual recognition model is available,and the evaluation results with high accuracy.In addition,the system has better performance in real-time and anti-interference,can provide more useful information to the driver.