Research On Recognition And Retrieval Methods For Fake Plate Vehicles

Recently, with the development of city construction and automobile industry, the number of vehicles grows, and the illegal activities caused by fake plate vehicles increases. Fake plate vehicles violate the interests of the legitimate vehicle owners, provide tools for lawbreakers, bring difficulties to police to catch the hit-and-run vehicles, cause losses to state taxes and highway maintenance fees.However, the investigation of fake plate vehicles mainly relies on the traffic police verifications and public tip-offs. It is inefficient and produces very little effect. Based on above analysis, a method of recognition and retrieval for fake plate vehicles is proposed in this paper which to investigate the fake plate vehicles effectively, Increase the intensity of the fight against crime.Vehicle face images are detected and trained, SVM model is adopted to predict the fake plate vehicle type. In order to retrieve fake plate vehicles, foreground of vehicle images are extracted to compute the similarity. The extraction of vehicle face and foreground can filter lots of background information, reduce the calculate complication and enhance the precision and efficiency.For the vehicle face and foreground set in this paper, HOG features are extracted and trained by the Adaboost method to generate cascade detector to detect the face and foreground area. MSRCR method is adopted to preprocess the high noon and night images to increase detection ratio. The experimental results show that using the method in this paper can detect the vehicle face and foreground images of unlicensed vehicles, license plate covered vehicles, license plate defaced and forged plate vehicles.On the vehicle recognition aspect, vehicle faces can be detected by cascade detector, and the type of the input vehicle face can be predicted by the SVM classifier. This paper proposes the method of second recognition using experience threshold based on SVM to enhance the fake plate vehicle recognition ratio, and the ratio can achieve 92%.On the vehicle retrieval aspect, the foreground of input image is extracted to compute the similarity by ACS method. The experimental results have showed that the technology of recognition and retrieval for fake plate vehicles using in this paper has high accuracy and good performance. The retrieval precision and recall are 86% and 77% respectively.Finally, exercise result presents that methods in this paper can detect fake plate vehicles efficiently, and the detection precision is 84%. The research of this paper solved problems of foreground extraction and fake plate vehicle recognition and retrieval, lays a strong foundation for the future development of fake plate vehicle recognition and retrieval system.