Key Technology Research For Traffic Sign Recognition Based On Machine Vision

As an important part of Intelligent Transportation System, Traffic Sign Recognition based on machine vision is of great significance for auxiliary driving or self-driving technology and has been widely concerned.Scholars at home and abroad is put forward many methods for traffic sign recognition, but because it’s under a lot of interference conditions, there are still some shortcomings. This paper is about the key technology s of image detection and recognition which is based on machine vision,and realized the traffic sign image of natural scene preprocessing, segmentation, extraction and recognition.Experimental results show that the proposed method is of high recognition rate and can well adapt to some adverse conditions.Because of natural scene is affected by factors such as light, weather,it’s not ideal to make segmentation on the image directly. At present, the common preprocessing method solved the interference from light changes, but the weather especially fog’s influence on the image segmentation is seldom considered.This paper adopted the image enhancement method to reduce the influence of illumination on the image segmentation and dark channel prior to dispel fog in traffic sign image preprocessing. The experimental results show that through the traffic sign image preprocessing, the color characteristics of traffic sign image were more outstanding.It’s the precondition of recognition to segment traffic signs from the background in the image.This paper used Lab color space which has obvious color characteristics to segment traffic signs and at the same time achieved coarse classification of traffic signs.Because of occlusion and interference to some traffic signs in natural scene, this paper adopted the bi-directional projection method to extract the traffic signs, the experiment proved that the method can be adapt to night, side, shade and other complex situations.In addition,this paper put forward the improved bi-directional projection method according to the problems of the same color group traffic signs.Experiments show that this method is simple to separately extract signs in the group and the extraction accuracy was higher,at the same time lay a foundation to efficient recognition on traffic sign.This paper adopted the SIFT algorithm which is based on local characteristics to do recognition on traffic sign. Experiments show that SIFT algorithm has good robustness for occlusion,rotate,dirty conditions and so on. This paper analysed the deficiency of the SIFT algorithm based on Euclidean distance and for the first time used RootSIFT algorithms in traffic sign recognition which based on the Hellinger distance. Experiments showed that RootSIFT algorithm improved the matching accuracy of traffic sign feature point, so as to improve the recognition accuracy.