Study On The Basic Algorithms For Vehicle Recognition System

Traffic vehicle recognition system based on the image detecting technology and image processing technology is the basic means to improve traffic transportation efficiency and reduce traffic vehicle crimes in Intelligent Transportation System (ITS). The key research in this dissertation is one of the core technologies in vehicle recognition system, the vehicle license plate recognition technology. The vehicle image pre-processing, vehicle license plate location and vehicle license plate characters recognition in vehicle license plate recognition technology are deeply studied.A novel image salt and pepper noise removal algorithm based on adaptive weight and nonlinear predication is proposed in this dissertation for the first time. It estimates the location of the noise in advance. Subsequently, the image boundary noise, the non image boundary noise and the non image noise are processed separately. So the salt and pepper noise removal are realized in vehicle image pre-processing.Aiming at the Gaussian noise removal in vehicle image pre-processing, a novel high-order PDE noise removal model based on the Minimal Surface in differential geometry is designed by means of the original high-order PDE noise removal model theory. It removes image Gaussian noise and protected image details. Meanwhile, it has the image salt and pepper noise inhibition.Fuzzy set theory is studied and a binarization algorithm based on the type II fuzzy set for car image is proposed. It establishes a type II fuzzy set based on the type I fuzzy membership and searches an optimal image gray level ergodicly as the image binarization threshold in the minimal fuzzy degree. So the basic task for vehicle license plate location is accomplished.Vehicle license plate texture features are analyzed and a vehicle license plate location algorithm is designed based on the texture analysis. First it utilizes the Local Binary Pattern (LBP) algorithm to analyze the image texture based on the binarization image. Then scanning method and projection method are used to locate vehicle license plate.A one-dimensional multiple Hidden Markov Model (HMM) for vehicle license plate character recognition is constructed based on the Freeman chain code. It extracts the character's horizontal projection chain code, vertical projection chain code, centroid inclination projection chain code of target pixel and distance projection chain code of target pixel as the character features. These features are used to train the character's one-dimensional multiple HMM and the vehicle license plate characters are recognized by this model.Based on the theoretical study and mathematical derivation above, compute simulation verifies the theory model's feasibility and the efficiency. The foundation for the further practical application task is established.