The Fine Classification Research Of Vehicle Type Based On Vehicle-logo Recognition

The system of Automatic recognition of vehicle type is an important component of ITS (Intellegent Transportation System), it has been widely used in the field of intelligent transportation, so the Research of technology related classification of vehicle is of catholic concern. At present, domestic and foreign researchers who focus on the recognition research of shape, size, color of vehicle mainly reasearch coarse classification of vehicle. In this paper, we pay attention to the research of location and recognition of vehicle-logo in order to achieve the fine classification of vehicle as the standard by vehicle-logo type(vehicle brands). In the location of vehicle, deffering from traditional segmentation method of image processing segmentation, we train a cascade classifier based on Adaboost learning algorithm to locate vehicle-logo in vehicle-logo candidate region which is roughly located adopting idea of pattern recognition. In order to achieve the classification of vehicle-logo, we train a classifier based on Adaboost. MH algorithm which extend a multi-class problem to be more issues of two classes. The methods to deal with above problems are testified in the experiments.In this paper, we mainly introduce, analyze and study 4 ports following:(1) Wavelet MomentTo better apply wavelt moment to the application of this article, we introduce and analyze the theory of wavelet moment in detail, and analyze and study the construction method of wavelet moment in depth in the port.(2)Research of Adaboost Learning AlgorithmAdaboost algorithm which fit together a large number of simple(weak) classifiers which have only general ability of classification to be a strong classifier that have great ability of classification is a classifier algorithm. It has proved by theory that the error rate of strong classifier will trend to zero while the number of simple classifier trend to infinity as long as the capacity of each simple classifier is better than random guessing. We analyze and study the background and theory of Adaboost learning algorithm in depth in the port.(3)The Study of Location Method of Vehicle-logoThis port is the most important in the paper. Deffering from traditional method of target detection based on image processing, we accomplish location of vehicle-logo adopting idea of pattern recognition.We improve Adaboost algorithm according to the shortcomings of Adaboost algorithm and the the distribution characteristics of wavelet moment of vehicle-logo, and then train a cascade classifier for a kind of vechile based on improved Adaboost algorithm to make a real-time and precise vehicle-logo location in multi-scale mechanism on the candidate region of vehicle-logo According to prior knowledge of the vehicle license. (4) The Study of verifying recognition Method of Vehicle-logoIt should automatically classify it after locating vehicle-logo. we use Adaboos. MH which extend a multi-class problem to be more issues of two classes to classify Vehicle-logo in this paper. We alse study recognition methon of vehicle-logo using Template matching and feature matching of wavelet moment do not match the vehicle identification methods and compare the performance of 3 methods by experiment.