Article Preview
Top2. Literature Review
The works on traffic signs recognitions can be traced back to the 1990's. At the very beginning of the studies on traffic signs recognitions, most works were focusing on detecting the traffic signs from a images about real-scene on the street while those method left the contains of the traffic signs not recognized. For example, Blancard (1992) recognized the signs by their color and form. In order to classify the colors, he used a band-pass filter to filter out most color but the chosen red colors attached to a black and white background. Meanwhile, a Sobel filter is applied to the images in order to find the edges inside the images. Associating with the edges, some features, including perimeter, length, gravity center and compactness are calculated and sent to a neural-network to recognitions. The method is fast (about 0.7s ~1s) but quite limited since it can only recognize the red background sign “stop” or similar warning signs while leaving other signs not recognized. Similar method is proposed in (Kehtarnavaz, Griswold, & Kang, 1993; Kang, Griswold, & Kehtarnavaz, 1994; Kang, 1994), where the combination of color and shape processing are used as the feature of the traffic sign. Besides the “stop” sign, Aoyagi and Asakura (1996) present a genetic algorithm to detect speed limit signs. They only work with the bright image due to the limitation of the Hue variations used in their method. After obtaining the Laplacian of the original image, the pixels are thresholded for recognition. However, they method do not take into account different scales for the horizontal and vertical axes; thus they do a matching only with a circular pattern. However, these results still remain on recognize the “stop” or similar red sign only.