Recognition Based On Wavelet Transform And Support Vector Machine Models

The vehicle recognition system is the more import part of intelligent transport system, it can recognize and categorize the vehicle and we can provide the evidences of management, scheduler, statistics data and the charges of the transport system according to the results.Although the automatic vehicle categorization systems have been mature in some developed country , because of many reasons , the rate of accurate categorization is not able to satisfy with our requirement in our country based on those systems. Therefore, for the sake of automatic charging on the highway and scientific and normative management in our country, the study of the vehicle recognition technology must be done.This paper is studying the technology of the vehicle recognition based on the algebraic character of the vehicle images. Beginning with the application of the wavelet analysis in the image processing field and the SVM in the recognition and categorization field, and then combining the characteristics of both algorithms, this paper put forward the algorithm and the characterization of the vehicle recognition for improving the correct rate of recognition. And this paper also introduces some background of the vehicle recognition and the application, and discusses the preprocessing technology of the image, characterization technology of the vehicle categorization.Firstly, in this paper, we adopt a kind of modified algorithm of moment invariants to abstract invariants of object and insure the invariants of movement, expansion and rounding.Secondly, in this paper, basing on the characteristic of multi-resolution analysis of wavelet, we research the wavelet transform of inter-scale dependency and abstract the characteristics of the vehicle image in every layer.Thirdly,This paper proposes a vehicle automatic-classification method based on SVM theory trained Support Vector Machine(SVM) off-line,and the tested vehicle characteristic data as tested samples then the vehicle type can be judged by means of the trained SVM according to the method proposed in this paper.