Traffic Sign Recognition Based On Machine Learning

With the development of technology and society, more and more researchers into theresearch work for the traffic sign recognition system. Because of the complexity of the naturalscene of road traffic, however, want to develop a set of mature traffic road trafficidentification system still needs further research. In an increasingly developed traffic system,a perfect traffic sign recognition system will provide great help for people’s daily safety, sothe traffic sign recognition system research has very important significance. This paperresearches on the traffic sign recognition system under natural scene work, focus on thesystem accuracy and real-time performance. The main contents of this paper are as follows:1) For the problem of lighting condition change will influence the road traffic signdetection in natural scenes, this paper proposes algorithm that based on adaptive threshold bymultiple threshold to detect traffic signs candidate region. Firstly, according to the traffic signcolor information of the input image, this paper use red/blue standardized pretreatment, andthen perform multiple thresholding between a certain threshold range. Secondly, this paperuse contour detection to detect the thresholding image, and add geometric constraint to screenthe contours, Hu invariant moments matching to determine the shape. Finally, this papermerge the final result to get the candidate region of traffic signs.2) For the problem of Chinese road traffic signs sample library shortage, this paper basedon the standard Chinese road traffic signs use synthetic methods, such as affine transformation,blur, add random background, to establish the sample library of Chinese road traffic signs.The main idea is synthetic method not only saves a lot of manpower and resources that spendon building the traffic signs sample library, but also solves the problem of traffic signs imagecaused by the size, angle and rotation changes circumstance, etc.3) For the problem of recognition for hundreds of road traffic signs, this paper based onHistogram of Oriented Gradient (HOG) features combine Support Vector Machine (SVM)classifier to recognition hundreds of traffic signs Quickly and efficiently. The main idea is thattraffic sign is a rigid object, and the performance for HOG feature on the rigid objectrecognition is perfect. Meanwhile, the HOG feature combined with the SVM classifier, in thecase of a certain optimization, can be classified hundreds of traffic signs quickly, thus wellrealize the real-time classification of hundreds of traffic signs.This paper implements the traffic sign detection and traffic sign recognition algorithm in the Windows system with Visual Studio2008. The experimental results of natural scenesvideos show that the proposed algorithm has better stability, accuracy and real-timeperformance.