The Research Of Vechile Detection And Recognition In Traffic Video Image

With the continuing progress of society and economy, transportation issuesbecome more and more serious. Meanwhile, under the rapid development of computerand other technologies, Intelligent Transportation System develops gradually andstarts attracting more attention. Vehicle Extraction and Classification is an importantbranch for Intelligent Transportation System, and has gained more and more concernand study. As a most important step, Vehicle Identification also becomes hot.The article briefly introduces the current situation and development of IntelligentTransportation System home and abroad, then it analyzes the present vehicleextraction and classification technology. And it stated its structure arrangement,content and creative points in brief, including foreground extraction, shadow detection,vehicle identification, etc.In the respect of foreground extraction, it first analyzes the current three mainmethods of foreground extraction. They are Optical flow method, frame differentialmethod and background difference method. Having analyzing all respects of theiradvantages and disadvantages, it uses background difference method to extractforeground. And then it discusses some current main background generationalgorithms and compares their experiment effects. Finally, it uses backgrounddifference method basing on mixture Gaussian distribution model to extractforeground.Regards shadow detection and elimination. It firstly elaborates the shadow’sillumination model, attribution and classification. Then it puts forward two kinds ofshadow detection methods：moving shadow detection based on local texture featuresintegrated with linear characteristics, the moving target can be obtained via thebackground subtraction method, then the description of texture can be achievedthrough its brightness and the improved LBP local texture operator, based on the firstjudgment through Hamming distance, the judgment of lineament is thus formed viatexture information,this method aims to get a better balance between real-timeperformance and accuracy,and can get better shadow detection; moving shadow detection based on Image edge detection which using Susan algorithm, the movingforeground position can be accurately obtained via the background subtractionmethod, then use the Susan algorithm to detect the image edge in interestedarea,finally，do shodow detection by analyzing the statistic characteristics of edgepixels,through the experiment,this method possesses high rate of accuracy. Finally thefirst method is chosen for shadow detection, and the object eliminated the shadows isreconstructed.In vehicle identification, it summarizes the current main algorithms for vehicleidentification, and uses geometry parameters to obtain the restructured object aftereliminated the shadow. In views of many local characteristics of the message, theinformation deficiency for the existence of the moving target, and big difference ofthe characteristic value when the same target stays in different positions in the image,a variety of geometry parameters are used to jointly determine, and mapping workunder the same basis during the parameters extraction, which enable the characteristicvalues to contrast. The experiment demonstrates that this method reduce wrongidentification rates and classify the objects more accurately, and has obtained goodeffects.