Integrated Recognition Of Vehicle's Shape And Marking Based On Image

With the rapid development of socio-economic and the constantly improvement ofpeople's living standards, in order to meet its development needs, the automobile industrydevelopves vigorously and the number of cars grows rapidly, cars have played an importantrole in transportation and daily life. The development of all aspects of society is[ aussian[ y promoted by the vehicle, But at the same time, The problem of trafficcongestion and the Transportation Sector's management of vehicles, which caused by alarge number of vehicles application, has increasingly attracted the concern of thecommunity. First of all, the smooth path is essential to energy conservation, environmentalprotection and transport efficiency in transportation, Moreover, the stagnation of traffic hasbeen smooth the impact of traffic bottlenecks at freeway toll station and in large parking lot,the fundamental cause of all lies in whether we can be effective models of vehicles anddetailed vehicle classification information. In addition, when the motoring organizationpunish fake plate vehicles and public security departments in pursuit of fugitives who aredriving, vehicle models as well as detailed information related to the identification ofvehicles can provide the evidence for them. As well, in a intelligent residential suburb,Models such as the vehicle identification information can be effective in preventingunauthorized access to foreign vehicles. To sum up, the study of identification of thevehicle models has an important significance.At present, models of vehicles have a lot of identification methods, such as the methodof radar, ultrasound, circular coils, infrared, electronic tags cards and so on, and they aremainly used in highway toll booths. Judging for the vehicle models are just small mediumand large car or sedan, vans, trucks, such as the identification of a rough. In this paper,through the appearance of vehicles based on image characteristics, vehicles are finelydivided and car model, series, brand information are also identified. Models correspondingto each category of its own attribute information, such as the number of seats of vehicles,wheelbase, Tread, dead weight, body's length, width, high and so on can be extractedaccurately. Relatively traditional methods, Image-based vehicle identification are simpleequipment, information-rich images, easy installation and small damage to the environment, vehicle pattern recognizing and a lot of information about vehicles are able to be wellcombined. In this paper, Mainly through the following four areas to explore.1,Selection and extraction of vhicle model featuresVehicle characteristics is the basis for model identification, a good or bad choices ofVehicle characteristics will directly affect the effectiveness of vehicle identification. Thebasic task of feature selection is to examine how to derive from a number of characteristicsthat identify the most effective of the classification characteristics. From the photographedimage of the vehicle, although we can get rich information of the characteristics, such asthe outline characteristics of the overall and local, characteristics of closed area, colorcharacteristics, lines characteristic and so on. But, when taking into account the vehicleidentification, vehicles which were shooted, are random changes at the distance androtation angle, the characteristics of the selection should not affect the sensitivity of these,In addition, if the characteristics of the direct selection exist the sensitivity of the above,whether they have invariance and the desirability after properly transformation of thecharacteristics. Based on the classification criteria of vehicle model and all thecharacteristics of vehicle models, this paper selects some transformation characteristics ofthe absolute distance which can be easily measured and have the features of classificationas the basis for model identification, They are some relative than absolute distance of thehorizontal and vertical.Characteristics of the vehicle models can be used by vehicle identification system onlyafter Extracted accurately. Some relative than absolute distance of the characteristic whichare choosed by this paper, attributed to some of the accurate extraction of feature points,they are the best value for point, center point, corner point and so on, if they are notaffected by External conditions, through the appropriate algorithm, these features are ofrectifiable. Finally, the characteristics of the vehicle can be extracted and calculated.2,Set up feature space of vehicles models sampleThis paper adopt template matching method to identify the vehicle models, Therefore,in this vehicles models recognition system, we should set up a sample database to identifyvehicles models. When this paper setting up feature space of vehicles models sample, itinvolve substantial and disorderly characteristics of data. For these feature data, How todeal with and how to use reasonably is focal point and difficult point. Sample feature spaceis standard model of model identification, so the appropriate method of data organization isvery important. In this paper, according to characteristics of a large quantity of data,disorderly and its classification, this paper adopted Fuzzy c-means clustering method to make reasonable use of 75 types of models and each of the nine characteristics of datamodels and analyed the special characteristics of the obtained data.In order to be able to roughly classify these large amount of data, this paper discussesthe number of cluster center selection problem and finally select four cluster centers. Forthe determination of membership function, in accordance with the characteristics of data ofall samples, this paper made a general analysis of the distribution of triangle membershipfunction and aussian membership function, this paper chose aussian membership functionwhich closer to the actual and adopt cluster center of each characteristics, which obtainedfrom Fuzzy c-means clustering as a distribution center of aussian membership function.While Fuzzy c-means clustering are dealing with the feature data, each of thecharacteristics of each model relative to degree of membership of the cluster center hasbeen also obtained. Organizing nine characteristics of each vehicles models obtain 75matrix which are 9×4, sample feature space models.3,Fuzzy recognition of vehicle modelsFeature space of samples provides a standard template for the classification of vehiclemodels, then this paper should identify a suitable classification m-ethod. By the impact ofexternal factors or other factors, the characteristics data of question to identify vehiclemodels may not precisely equivalent to data of the sample template and change through acertain range of error scope. In order to enhance the robustness of the vehicle modelsclassification, this paper selects fuzzy identification, the relation of models to beidentification and standard model isn't an either-or question, it is certain affiliation, the bigaffiliation belong to or close to its template model and the small affiliation don't affiliate itstemplate model. Vehicles to be identified set up the characteristics space of vehicles inaccordance with the model established in the form of feature space of the vehicles, it takethe idea of template matching to classify the vehicles. In specific classification algorithm,This paper discusses the classification algorithm based on the close degree of the fuzzymembership function and distance and improved close-weighted classification algorithmaccording to the stability of eigenvalue extraction size. this paper puts forward in theapplication of hierarchical classification of fuzzy feedback method.which based on thespecial characteristics data, fuzzy neartude of membership function and fuzzy neartude ofdistance. This paper can make use of closest principle or maximum membership principleto make the final classification.4,Identification experiments and results analysis of VehiclesIn order to test the recognize result of Vehicle Recognition's method which is identifie ed in this paper. In C++ Builder6.0 environment, this paper carried out Vehicle identification experiment of software in different Shooting distances, rotation angles and under normalcircumstances. The experimental results show that the vehicle can be correctly identified inthe not too far shooting distance, not too great rotation angle and the external environmentunder conditions of non-complex, It verifiys the effectiveness of vehicle identificationmethods and achieves the desired effect of classification. This paper identifies the directionof improvement and optimize, on the issue of eigenvalue extraction of recognition modelsand membership function.