Traffic Sign Recognition And Straight Road Detection For Intelligent Car

With the development of computer hardware, micro computers have powerful information processing ability. Computer vision has come into people's life. At the same time, due to the fast develop of economy, the number of cars on the road increase very rapid. As a new direction to solve traffic problems, Intelligent Traffic System (ITS) emerges at this moment. As the important elements of (ITS), traffic sign recognition and road detection have great significance.At the beginning of this paper, the state of development at home and abroad is introduced with analyzing of the merit and demerit. Then, the paper proposes the a new framework of intelligent car auto-driving system, including two aspects:1. A new traffic sign recognition algorithm based on multi-level Neutral Network, 2. road lane detection based on geometric constraints.To meet the real-time need of traffic sign recognition, this paper does not adapt complicate algorithms. This method takes full advantage of the features of traffic signs themselves, using simple image preprocess, RGB and HSV color coarse segmentation, image morphology process and self-defining shape coarse filter to locate traffic signs fast. And then, inner signs are separating out. This paper proposes a new kind of light-weighted feature to express the inner signs. After feature vectors are derived, Neutral Network is used to train a fast classifier. The method has both the precision and speed.The straight lane detection algorithm of this paper utilizes Sobel and Tukey to fit basic lines which used for next process. Then, with the geometric constraints of distances between roads, distances to vanishing point and line slope to fix left, right and middle lanes of roads. This method reduces affects of shadow and building edges significantly. The distances computed are based on Geometrical Calibration, which realize a transformation from 2-d image coordinate to 3-d word coordinate, with a constant height.