Traffic Signs Detection Under Complex Environment

In recent years, intelligent transportation system (Intelligent Transportation System, ITS) has aroused more and more interest and attention. The system can reduce people’s driving pressure, make people travel more freedom, security, and reliability. As a necessary part of ITS, Traffic Signs Detection and Recognition System (Traffic Signs Detection and Recognition System) is a reliable guarantee for intelligent vehicle or driver get traffic information outside. Traffic signs detection is the key to identify, real-time detection of region of interest accurately will lay a good foundation for identification. For the image vehicle captured under the running state in the complex external environment, it requires preprocessing stage to ensure the accuracy of detection. This paper first aims at the motion deblurring and fog dehazing and then the detection of signsThe problem of image deblurring divided into two classes, uniform linear motion blur restoration and non uniform linear motion blur restoration. For the former, introduces the common method to estimate the direction and scale of the blur kernel and the basic algorithm of image restoration. For the latter we improve the blur kernel estimation using strong edge algorithm. Denoising with guided filters and and operation in R, G, B three channel, The algorithm can deal directly with color images At the cost of increased less computation, The restoration result keeps the image color information.For the problem of image dehazing this paper adopts the latest research results of image dehazing area, dehazing using Dark Channel Prior Priori. In the aspects of optimizing transmission function put forward the method of combining guided filtering and Linear interpolation.After natural images preprocessing, links into the traffic sign detection, this paper is divided into two steps:(1) For the red, blue, yellow three colors establish two dimensional normal distribution model for the I component and Q component in YIQ color space, calculate the conformity of a pixel for the three models to complete the segmentation of color, Then the morphology processing, get the two value image as the initial interest regions(2) Select Hu moment invariants features which are robustness to shape size, rotation and angle. Get the Hu moment invariants features of initial interest regions, put them into Support Vector Machine (Support Vector Machine, SVM) and training the classifiers. Get rid of irregular categories, and keep traffic sign characteristics of round shape, rectangle and triangle categories, as the final detection results.The main work of this paper is to explore and improve image deblurring and dehazing algorithms, to make the effect is better and real-time, make it more in line with the practical requirements of traffic signs pretreatment. In the detection stage, the algorithm in this paper are better, the false detection rate and false negative rate is low, good real-time performance and also have the good performance to the changes in illumination and a small amount of occlusion problems...