Traffic Signs Recognition Based On Two-Dimensional Principal Component Analysis And Support Vector Machine

In recent years, along with the rapid development of global economy and the general improvement of people’s living standard, the number of vehicle is becoming bigger and bigger, and the intelligent transportation system is being paid attention by more and more people. As a part of the intelligent transportation system, the traffic signs detection and recognition play a crucial role on the safety of the traffic, at the same time is conducive to ease traffic congestion phenomenon. Therefore, the further discussion and study on the traffic sign recognition will play an extremely important particular value and profound significance on people’s lives.In order to be able to quickly and effectively identify the traffic signs contained in the image, the thesis first preprocesses the traffic signs images. Histogram equalization and median filter was adopted to realize the image enhancement and denoising. Then, by using of their remarkable characteristics, such as color or shape features, the images are detected and segmented. Through analyzing and comparing several color spaces, a coarse segmentation in HSV color space is made by the determined threshold value, and further, a precise segmentation is made according to the image shape feature, so as to realize detection and segmentation of traffic signs images.Next, feature extraction is done using an improved two-dimensional principal component analysis method, which the essential characteristics of the traffic signs are obtained. These simple characters, facilitating computation, compose the characteristic vector and form the sample characteristics database. Through the experiment came up from traffic signs recognition rate and recognition time contrast research principal component analysis method, two-dimensional principal component analysis method and the improved two-dimensional principal component analysis method. Finally, a combined kernel function of support vector machine classifier is used to classify and recognize the input trafficsigns. The recognition accuracies of the traffic signs image under the polynomial kernel function, the radial basis kernel function and the combined-kernel function of the support vector machine classifier are compared, and adopt the method of cross validation parameters optimization was carried out on the kernel function.Using LIBSVM, the software package of MATLAB, some numerical simulations are carried out, and the experimental results show that the improved 2D-PCA method and combination kernel function of support vector machine classification equipment in the present thesis owns higher recognition rate and accuracy than the others and can meet the requirement of real-time.