Algorithm Research On Segmentation And Recognition Of Circular And Triangular Traffic Signs

With the development of economy, the cars in our daily life play a role more and more important. At the same time, the problems of traffic jams and traffic accidents are becoming more frequently than before. Scholars have proposed many solutions about them, including the concept of intelligent driving. The intelligent recognition of traffic signs is an important part of intelligent driving. It has important practical significance about future automotive electronics development direction.Traffic signs segmentation and recognition system has both theoretical significance and practical value. It mainly solves two problems, that is traffic sign segmentation and recognition. The meaning of segmentation is to separate traffic signs from pictures of natural scene. The meaning of recognition is to classify the traffic signs. This paper is mainly doing research for circular and triangular traffic signs, such as the algorithm of their segmentation, the algorithm of image feature extraction, the algorithm of recognition. In-depth research has been done about key technology of the issue, including:1. Analyzing the framework of traffic sign segmentation and recognition system. Doing some research about pre-preprocessing algorithm, such as image filtering, color segmentation, edge chain storing. Analyzing the segmentation and recognition algorithm proposed by related scholars.2. According to the shape characteristics of the circular traffic signs, a new segmentation method is proposed, which is based on the two layers of the Hough transform. The first layer is to determine approximate circular point and radius. In the second layer, this paper improves the algorithm of random Hough transform, which is to determine the precise area of the circular traffic signs.3. According to the shape characteristics of the triangle traffic signs, a new segmentation method is proposed, that is Multi-feature fusion triangle segmentation. The method is based on using Hough transform to detect straight line and the vertex decider using Extreme Learning Machine. At the same time, the segmentation method integrates the triangular geometric features into the algorithm to filter triangle traffic sign edges. So the precise area of the circular traffic signs is determined.4. Combining the advantages of wavelet moment invariants and affine moment invariants, this paper proposes a new moment invariants. This new moment invariants have panning, rotating, scaling invariance. And it is used in the feature extraction of traffic signs. Corresponding to traditional simple extraction of Hu Moments, it has greater distinguish reliability. The new moment invariants improve the recognition accuracy.5. Selecting20groups of traffic signs, and doing some research on Extreme Learning Machine algorithm (ELM). The ELM algorithm is used for recognition of traffic signs, and achieved good recognition results. While comparing the recognition results with BP neural network, it improves the recognition speed.