ABSTRACT
ABSTRACT: Traffic lights are very important for traffic management and control. If there are errors in traffic lights, managing the traffic becomes difficult. Therefore, it is highly essential to recognize the state of traffic lights on line. This paper proposes a detection method of errors in countdown traffic lights based on color union spaces and fuzzy Principal Component Analysis (PCA). In this method, first, pixels are classified into color pixels and achromatic pixels in RGB color space. Then, the principal color of the color pixels is extracted by histograms in HVS space, and the color feature is recognized based on the H component of the principal color. The color countdown characters are converted into gray according to the principal color, and then the training samples of the countdown characters are divided into subsets in terms of fuzzy degree; then, their corresponding PCA subspaces are constructed. Lastly, the character is recognized in the specific subspaces, which are chosen by the character’s fuzzy degree. The experiment results show that our method has a more favorable comprehensive performance than other algorithms, and that it can better meet the demands of precision and real-time processing simultaneously.
