A Method of Traffic Sign Detecting Based on Color Similarity

In order to detect traffic sign effectively and quickly, firstly, this paper firstly discriminates the maximum and minimum values of the color-components R, G and B, and then ignores pixels having little color similarity with traffic sign. Secondly, it regards the points of RGB space as color eigenvector and calculates the cosine value of included angle of vectors to discriminate color similarity. Thirdly, the image is divided into core image, undetermined area and background area by setting two thresholds, and then two corresponding templates and modes are used to scan the image, realizing growth of core image by area mergence. Finally, the operator Roberts is applied to detect fringe of the traffic sign. The experimental result shows that the method has better result of detection both in sunny and in rainy day. The process of detecting avoids huge amount of calculation for color space conversion, and improves the speed of calculation, which will lay the foundation for recognition of traffic sign.