Design And Implementation On Vehicle Information Recognition System

With the rapid increase in the number of vehicle and the development of computer-vision based Intelligent Transportation System(ITS), vehicle information identification becomes an important subject in the research field of pattern recognition. Traffic problems, such as congestion, violation, and the use of fake license or unlicensed plates, have become increasingly serious due to the growth of motor vehicles. Despite the wide and mature application of license and unlicensed plates. To address this issue, a new approach to identify the vehicle information including the vehicle model(related to a specific brand) and the vehicle body color is presented in this thesis.The vehicle information identification mainly consists of six parts in this study: sample image preprocessing, extraction of “vehicle grille” and body color recognition area, feature extration and dimensionality reduction, vehicle model recognition as well as recognition of vehicle body color. The follow work has been carried out.1. Affter processing the vehicle image, use the location of the license plate to locate the vehicle grille bar if present, otherwise, use the improved percentage-threshold image binarization algorithm to locate the vehicle grille bar. Then the vehicle grille can be accurately located from the vehicle grille bar by applying gradient projection method to greyscale image.2. Based on the analyses and comparisons of the commonly used texture features, the combined feature of Histogram of Oriented Gradients(HOG) and Gabor wavelet transform is proposed to describe the image of vehicle grille. Moreover, Locality Preserving Projection(LPP) algorithm is used to reduce the dimension of the feature.3. The final recognition rate of vehicle model can reach 92.88% when the naive Bayes classifier with minimum probability of error or risk is chosen to identify the vehicle model.4. According to the comparison of four different color space models, HSV color space is selected as the best to identify the vehicle body color. Thus its color quantization and color determination are greatly optimized to improve the final recognition rate of vehicle body color to 88.73%.Under the environment of VC++2010, the vehicle information identification system is ultimately built up to test the obtained vehicle images. Our experiment confirms the feasibility of this system in identifying the vehicle model and body color and indicates the potential broad applications of this system in ITS.