Research And Implementation Of Traffic Sign Recognition Algorithm Based On Computer Vision

Over the last decade, with social progress, economic development and the popularity of cars, traffic accidents occur frequently, most of which must be due to human factors. Traffic signs which are led on the top of lane or both sides of the road, is used to provide the maximum or minimum speed limit, warn the change of conditions in front of the car itself, give some restrictions to protect the safety of other vehicles. However, when the attention of the driver is poor, he is often unable to identify this information clearly, As a result, accidents happened. Therefore, the automatic detection and recognition of traffic signs is of great theoretical significance and researchful value.Traffic signs'detection and classification are quite difficult because on one hand there are variety of traffic signs in the world and on other hand they are often in complex outdoor environmental conditions, which result in the recognition algorithms are vulnerable to the weather, light, deformation, tilt, bleaching, similar background, the image edge fuzzy and many other factors, furthermore, the performance of the algorithms is need to consider of not only recognition rate but also real-time. This article has improved and innovated the existing recognition algorithm by summing up large various research methods at home and abroad, and has designed and realized a new traffic signs recognition algorithm, composed by three parts:the detection, classification and tracking.The detection is based on the color and shape. On the first the combination of the RGB and HSI to do a rough segment is used, then the features such as the corners of the polygon and the edge of the circular is used to do a fine detection. The algorithm can overcome the impact of light and complex conversion of model, solve the failure of original corner detection to circular, avoid the time-consuming Hough transform and improve the accuracy and reliability of the algorithm.In the classification stage, the improved Harr wavelet has been used to extract the edge inside of the signs and an algorithm based on support vector super-spere binary tree SVM multi-classification has been used. It can effectively solve the often encountered non-separable problem of multi-classification, improve the promoted capability of the algorithm, shortened the training time and accelerated the speed of classification.In the tracking phase, The Lucas-Kanade feature points tracking algorithm is used to not only avoid the difficulty of modeling the non-ground targets, but also enhance the applicability and stability of the algorithm.In this paper, A lot of road sign pictures in China and Japan have been searched to test our algorithm, the experimental results show that this method has a high recognition rate and a certain robustness to the complicated scenes. It is find a practical solution to solve the existing problems and difficulties of the traffic sign recognition.