The Traffic Sign Tracking And Recognition System Research Based On The Autonomous Vehicle

The research of intelligent vehicles based on computer vision is one of thekey problem for the realization of ITS. Traffic sign automatic recognitionsystem is an important part of autonomous vehicle vision and one of the keytechniques for the design of intelligent vehicles. Through the research on thedetection and recognition methods, this dissertation designed a traffic signtracking and recognition system based on the autonomous vehicle.The main contents of this dissertation are concentrated on the colorsegmentation, shape detection, traffic sign tracking and recognition. The colorof traffic sign was influenced by the variations of the light conditions. A methodthat combined color filters and fast two dimensional OTSU algorithm wasproposed for the fast and efficient detection in different light conditions. Thismethod improved the segmentation robustness to the light variation.In order to differentiate the traffic signs and the objects which have thesame color features as traffic signs, shape detection was implemented after colorsegmentation. This dissertation proposed a method that combined optimalcorners and symmetric detection which can overcome the influence of trafficsign rotation and tilt. The tests show that this method has better robustness tothe traffic sign rotation and tilt and can detect the abrupt slope traffic sign whichhas only part yellow color information. Because there is not obvious corner incircular traffic, a method that combined edge fitting and symmetric center wasproposed. Through this method, the center of the circular traffic signs can belocalized and the radius of the circle can be computed.When the velocity of the autonomous vehicle was changed, the Kalmanfilter can not track the traffic sign accurately. In order to solve the problem thatan improved Kalman tracking method was proposed by using the edgeinformation of the traffic signs. This method can track the traffic sign accuratelywhen the velocity was changed and avoids predicting the size of traffic sign inevery frame.This dissertation designed a multi-level recognition system based on theSVM to improve the accuracy of traffic sign recognition. This method was based on the stepwise refinement strategy and by using the color information, shapeinformation and CVOG feature, the multi-level recognition part can implementlayered recognition.The tests on the autonomous vehicle and by using the actual video showthat the traffic sign tracking and recognition system can obtain both highaccuracy and better robustness to the light variation and traffic sign at a tilt.