Method Of Car Body Color Recognition Based On Support Vector Machine

Car Body color recognition is an important direction of development of the field ofintelligent transportation, It makes up the technology gaps of the color characteristicsinformation of the intelligent vehicle identification system, and it also riches thecharacteristics information of the vehicle identification, moreover,it has a veryimportant significance to fight against the crime of the car deck behavior. However,theactual color of car body is easily effected by fog, rain and other weather factors, whichbrings much difficulties for the vehicle color recognition. In this contest, this thesis putsforward its own reseach on car body recognition.The main contents of this thesis are as follows:1.research the method of car body color positioning, the regional location of the carbody color is a key step for feature extraction, firstly, locating the position of the licenseplate by use of edge jump and projector positioning algorithm, and then locating the carbody color recognition area by use of the geometric relationships of license plateposition and bonnet position and method of statistics and experiences.2.research the theory and method of the fog remove of the color image,establishing the transmittance model and the atmospheric optical model using darkchannel priori theory, this method has a good effect on the fog remove of image andimage enhancement. but it also has its flaws, it is not available for some image whichcontains some white object block or highlight block area, so, this paper bring up athreshold factor to rebuild the transmittance model, and then get the new imagerestoration function model, the new algorithm makes the image becomes clear withoutcolor cast, and it also improves the car color recognition rate.3.reseach the car body color feature extraction method, the color histogram hashigh robustness to the image rotation and image scaling, which indicates the colorhistogram only has interest in the color proportion of the color space, but it has norelationship with the color distribution in color space. because of having thischaracteristic, color histogram can be used to describe the characteristics of the globalcolor similarity, so,this thesis can use color histogram to extract car body color feature information.4. Two proposals are brought up for car color recognition on the support vectormachine platform, and two proposals each has their own advantages, which achievegood body color color classification recognition accuracy.