Study Of Automatic Vehicle Classification Based On Artificial Neural Network

Electronic Toll Collection (ETC) represents the developing direction of road toll in the future. Nowadays ETC springs up both in China and abroad. Automatic Vehicle Classification (AVC) plays a very important part in ETC. Its performance determines the quality of ETC. It is very significant to do some researches in AVC.This thesis brings the artificial neural network(ANN) to AVC. Taking advantage of its powerful ability to deal with classifying problem, this thesis designs preliminary AVC that can classify huge trucks, trucks, huge passenger cars, middle passenger cars and cars correctly. Therefore the research is done in several aspects as follows:1. Comprehensive and systemic research is done in the existing method to classify vehicles automatically. On the basis of analyzing the merits and demerits of diverse methods, the thesis puts up its own scheme.2. The conception of characteristic parameter is put forward in this thesis. From the relevancy analysis of the characteristic parameter of the vehicle shape, this thesis discusses the confirmation of the vehicle's characteristic parameter. After the characteristic parameter being confirmed, the thesis expatiates the measuring of characteristic parameter and the way to implement.3. This thesis adopts the method which combines theory with experiments to design the vehicle classifying BP network. As to the complex problem that how to select the number of hidden-layer, a good deal of experiments are done and the best consequence is gained finally. It is validated by sums of experiments that the different sequence to feed training samples in has a tremendous effect on the performance of the network. According to this, the proper input sequence of the samples is confirmed.4. With MATLAB, vehicle classification BP network is built and will be trained until its error satisfies the demand of the design.5. After the training, simulation tests will be given to the BP network under the circumstance of MATLAB. In order to test the network overall, some special tests are added after the ordinary tests. The results of the simulation tests indicate that the AVC designed by the thesis can classify huge trucks, trucks, huge passenger-cars, middle passenger-cars and cars accurately.