The amount of cars in China is largely growing and traditional human management inefficiencies have been unable to keep up with such a high density of modern traffic. Video trafficmonitoring technology, which is mainly based on vehicle license plate recognition, is the important part of the modern Intelligent Transportation System(ITS) and has played a decisive role in illegal vehicle monitoring and maintaining good traffic order. However, illegal vehicles with fake license can escape monitoring of the system, and a lot of trouble will be brought to the true license plate owner and vehicle management bureau. Fake License Plate Detection System(FLPDS) is one of the ways to solve this problem. In this paper, some researches on detection of fake license plate of vehicle based on image matching are done and the architecture of FLPDS is proposed. These researches consist of three parts: fast localization of vehicle logo based on visual saliency, vehicle logo detection and architecture of FLPDS. In the first part, we use a simple visual saliency model to distinguish the small part including vehicle logo against the irrelevant background, thus achieving the purpose of rapid extraction of regions of interest(ROI) and significant reduction of processing time. Then in the second part, a universal high-precision matching algorithm, which is the core part of this paper, is introduced. We extract the feature points of both templates and vehicle images first by a proposed edge-feature extraction method, then we use them to compute a matching score which indicates the matching degree by our metric proposed in this paper. Finally the score is compared with a threshold(which is obtained by large amount of experiments) to decide whether the target license plate is fake. In practical scenes, challenges for vehicle logo matching are mainly from the fact that vehicle logo differs a lot due to changes in viewpoint, deformation and light condition. Leak recognition and mistake recognition are inevitable but we should reduce them to a very low level for two reasons. On the one hand, the behavior of using fake license plates is illegal and the judgement should be prudent, human judgement module should be introduced in the system. On the other hand, due to the huge traffic, there is not enough human resource to deal with suspected vehicles with fake license plate if leak recognition rate is now low enough. Thus Multi-template matching module is recommended in this paper. The sy...
修改评论