Research On Traffic Sign Recognition Algorithm Based On Neural Network

Image recognition is one of the most important research issues of pattern recognition. In this paper, we take traffic signs recognition as background, By analyzing the recognition process,we divide this subject into image segmentation, image preprocessing, feature extraction and classification.First of all, based on RGB model, a segmentation method of traffic signs is proposed, which to extract the R,G,B color-components. Considering that the target background gray level diagram of the three components has not much difference, we use R,G,B component images subtract by each other to extract images of traffic signs and the Otsu's threshold algorithm to output the segmented image. The results show that this algorithm not only has less computations as well as satisfactory segmentation effects, but also can improve the system in real-time.Subsequently, based on the research results of both domestic and abroad, we propose an improved invariant moment approach which on the basis of Hu invariant moment, and derive a more general measure of the invariant moments, of which the rotation, translation, and the proportion of invariance are included.In the thesis, we apply the improved invariant moment and Hu moment to extract traffic signs'features, separately, then, use the BP neural network as a classifier to recognize the signs. The experimental results demonstrate that compared with traditional Hu moments, the improved invariant moment has a higher recognition rate and better classification ability. The results also indicate that by improving the extracting features, we not only can achieve the purpose of improving recognition accuracy, but also meet the requirement of traffic sign recognition in autonomous navigation system. Obviously, the presented method has more advantages.