Research Of Traffic Signal Lights Recognition Technology

In recent years, with the general increase and popularity of the motor vehicle, traffic safety has become the focus of global world. It is not an easy task for drivers to keep constant attention on the traffic signs, because traffic scenes are complex and there are so much information to be noticed.In some cases, for instance, traffic roads are dramatically complex or there are too much visual information, even the drivers’ stress or visual fatigue, therefore some traffic data can be missed, even worse lead to traffic accidents. As the drivers’ great assistant, driver assistant system and unmanned system have got more and more favor of the drivers, and help the drivers to get the complex traffic information. As the important part of the driver assistant system and unmanned system, traffic signal lights recognition technology research can help the drivers get the information of traffic signal lights, and speed up the development of intelligent traffic management. Around the recognition of traffic signal lights, this paper made the following work:Firstly, since the traffic signal lights emit light, this paper proposes the algorithm that extracts light points, pre-processing the input image, getting rid of the areas which don’t conform to the morphological criteria, picking up the light pixels. Secondly, since the traffic signal lights has constant colors(red, yellow and green), this paper proposes the algorithm that extracts the traffic lights’color candidate points, transforming the RGB image to HSL image, setting threshold for H(Hue), S(Saturation), L (Light) components respectively dispending on the color(red, yellow and green)of traffic signal lights, removing the points which don’t satisfy the threshold, extracting the candidate points and labeling the color. Once again, since there are so many misrecognition in color segmentation, this paper proposes the algorithm that fuses the result of light points and color candidate points, dispending on the coordinates of light pixels, collecting respectively the numbers of red, yellow and green points in color segmentation image with template and expended template, saving the points who satisfy the thresholds setting for template and expended template as traffic lights’ candidate points, this process is the fusion of light pixels and color candidate points extraction. Finally, after the above steps, misrecognition still exists, so this paper proposes the algorithm of shape recognition basing on SVM(Supported Vector Machine) and Haar wavelet features, carrying out shape recognition on traffic lights candidate areas which produced by traffic lights candidate points, getting the final correct recognition result.The traffic signal lights recognition technology proposed this paper, is not rigidly adhere to the requirements of the environment and specific equipments, at the same time, has simple algorithm, the testing results show that this technology can recognize the traffic lights efficiently, and the result is perfect, the method is easy to succeed.