The Research Of Vehicle Recognition Based On Image Processing

With the increasing problems in traffic management, the development of computer technology has made the intelligent transportation system more and more applied. ITS have been widely applied in advanced country such as Europe and Ameriea. But in our country, the application of it is in research phase. Automatic identification and tracking of vehicles is an important part of ITS. This thesis analyzes various types of vehicle identification technology, and uses Image processing and pattern recognition technology to complete vehicle identification.《Vehicle Classification of the Toll for Highway》is adopted in this dissertation as the vehicle classifieation standard.First, the vehicle image was extracted from camera image after removing the background. Then this dissertation describes image pretreatment, feature extraction and pattern recognition algorithm, finally the type of the vehicle was obtained.The main works and creative point in this dissertation are summarized as follows:(1) In the process of vehicle image pretreatment, this dissertation analyzed various types of noise, and utilized mean filter, median filter and wavelet filter to carry on the filter pretreatment. The experimental results show that wavelet filter can remove noise of vehicle image and reserve contour features. According to experimental results, this dissertation choosed the edge detection method suitable to vehicle recognition. Using Hough transform to detect circular laid a good foundation to feature extraction.(2) Feature extraction.Seven car image invariant moments are extracted from the car outline based on Freeman chain code except for the normal features such as the vehicle length-height ratio, the ratio of distance between axles and vehicle length, dispersity(the ratio of vehicle outline perimeter's square and area surrounded by car outline), A high performance was achieved to identify the vehicle at different positions in the image because of the utilization of these vehicle image invariant moments features.(3) This dissertation studied basic theories and main method of pattern recognition, and designed SVM classifier based on MATLAB, and presented a vehicle recognition System based on Support Vector Machine(SVM) using decision tree on the basis of contrast and analysis. In the last part of the paper, simulation experiment shows that system has a relatively high recognition rate.