Traffic Sign Detection And Recognition Algorithm Research Base On Machine Vision

In the study of traffic sign detection and recognition, due to its shape and color features of the traffic signs, you can utilize the logo’s color as auxiliary discriminant information and train model to realize the position detection based on shape feature during detection’s process; And for recognition, people always take advantage of the support vector machine(SVM) for classification of traffic signs, nowadays people more make use of convolution neural network this model of machine learning, use traffic signs’ color images as network input directly, and then automatically extract its characteristics, because of the characteristics of the convolution neural network, we can obtain good recognition effect with the support from the graphics processing unit,through training and recognition process in parallel computing.the main research direction is based on the traffic sign detection and recognition,starts from the analysis of the characteristics of the traffic signs’ color and shape,and makes extensive and deep research about the traffic sign detection and recognition in the natural conditions.in this article, the main research contents are as follows:(1) According to the classification of traffic signs, suffered light characteristics,for under the condition of certain natural light mark surface is too light or too dark,studied the preprocessing algorithm of traffic signs, to realize the limitation on the color image contrast adaptive histogram equalization, to improve the image quality.(2) In view of the main traffic signs in shape are rules, symmetrical features, can be used in different image resolution on describe overall shape characteristics of traffic signs and the relative relationship between internal parts and components in a variable component model describes and detection of traffic signs.(3) The calculation model for variable components of two main steps, namely the characteristics of the pyramid time-consuming problem in the process of calculation and convolution, respectively use according to the characteristics and the adjacent layer characteristic estimation specified using Fourier transformation converts convolution product calculation way to accelerate the traffic sign detection operation process.(4) In traffic sign recognition algorithm, using convolution neural network instead of the commonly used support vector machine(SVM) model, neural network has the deep structure of convolution, direct traffic sign image as network input,automatically extract the characteristic information of the traffic signs in the network of terminals, in the form of probability output traffic signs under that category.The experimental results show that the algorithm is feasible and effective, in the detection and identification of phase has made very good accuracy, good real-time performance and stability.