The Research On Vehicle Recognition Technique Based On Video Image Processing

This thesis mainly discusses the fundamental theories and key technologies of vehicle recognition technique based on video image processing in ITS(Intelligent Transportation System). Traffic detection and information collection has become an important subject in ITS,and video image processing,vehicle recognition are the basic parts.Based on summarizing and analyzing the existing technologies of vehicle recogniton,this thesis studies these issues,and brings forward new methods.Moreover,experiments are implemented to demonstrate the validity.The main contents of the study include such aspects as following:1. Study on video image pre-processingThis thesis includes gray of color image,vehicle area detection and location,noise elimination and determine the vehicle corner information of the video image collected.This thesis collects one group of video traffic image which includes vehicles information by digital cameras,then transfers color image to gray image to reduce dimension information for vehicle area detection and location.Vehicle recognition maybe affect by noise interference,so image noise eliminiton is the most important part.Edge detection and corner detection after noise elimination can reduce the processing of vehicle information.This thesis is procved that it can get more corner information which is used in vehicle recognition.2. Vehicle feature extractionVehicle feature extraction is one most important procedure in vehicle reconition technique.The difference between eigenvalue is the most important theoretical basis to distinguish between the type of vehicles.This thesis chooses cross ratio of the projective invariant feature based on Invariant Theory of Vision ,then calculates projective invariant value to recognize vehicles by using vehicle corner information.The experiment shows that after comparing the projective invariant value can achive the correct identification of vehicles in an ideal case.It has an good robust performance.3. Study on vehicle recognitionThe corner information collected maybe missing or shifting afterimage pre-processing because of noise.It causes huge affection on projective invariant value,then misjudges the type of vehicles.This thesis uses na?ve Bayesian classifier model to judge the posterior probability of the type of vehicles when the corner information is wrong,and then compares the posterior probability value of different type to recognize the vehicle.This study is esay for vehicle recognition system basd on video image processing.