Traffic Signs Detection Algorithm Under Complex Background Environment

Traffic signs are the major parts of road traffic system, which can provide important information to the driver for safe and comfortable driving. It's so very meaningful to research traffic sign recognition system, which is very important in driving assistance system and intelligent vehicle system.Traffic sign recognition system generally contains two procedures, which are detection and recognition. Detection is a prerequisite to recognition and also the most difficult procedure, which still has many unsolved problems. This thesis takes a deep research on traffic sign detection based on the color and shapes characteristics of traffic sign. Major contents are showing following:Firstly, some color segmentation methods are compared, and fast color converting method to convert color space from RGB to HSV is applied, then HSV color space is adopted to segment the image.Secondly, preprocessing is taken on the binary image after color segmentation, such as small area elimination, noise removement, holes filling, outer contour extraction et al. The preprocessing is necessary because of existence of a lot of noise and non-target areas after the color segmentation is not benefit to shape recognition carried out next step.Lastly, several shape recognition methods are also compared, and circularity and corner detection are chosen for detection of circular and rectangular traffic signs for their accurate and robust. The innovation named as distance histogram is proposed to detect circular signs that take on an incomplete edge after color segmentation. This innovation reduces missing rate of detection extremely.Experiments show that the algorithm taken in this thesis performs accurate and robust, and reaches real-time demand.