Research On License Plate Location Technology In Low Quality Surveillance Video

As one of the important components of Intelligent Transportation Management System(ITS),Vehicle License Plate Recognition System has been widely used in daily life.However,the application scene is limited to the better quality scenes such as parking lots and community entrances and exits.For more complex environments,especially low-quality surveillance video environments,it does not apply.This paper analyses the difficulties of license plate location in low quality surveillance video,as well as the shortage of the existing license plate location technologies,and then do research into license plate location in low quality surveillance video.The main research contents are as follows:(1)A vehicle license plate location method based on feature fusion is proposed.Taking the idea of combining feature extraction and machine learning as the starting point,this method analyses license plate from the feature of edge,color and texture and then fuse the feature.Then a pre-trained SVM classifier is used to classify the proposals.Experimental results show that the proposed method using multiple features simultaneously has a higher accuracy rate and a lower undetected rate compared with other methods using a single feature or two features,but the applicability to the actual low quality surveillance video still needs to be improved.(2)A license plate location method based on convolution neural network(CNN)is realized.In order to improve the accuracy of license plate location in low-quality surveillance video,the method improved the feature extraction network and regional generation network(RPN),based on the Faster-RCNN.Besides a license plate database for CNN training is constructed.The experimental results show that the license plate location method based on CNN can be applied to multi low quality surveillance scenarios.Compared with the license plate location method based on feature fusion,the accuracy of license plate location has been significantly improved and more robust to low quality surveillance video.(3)A license plate location system in low quality video surveillance is realized.In order to reduce the false location rate and correct the tilted vehicle license plate,a method to filter out the false proposals and a correction method of tilted vehicle license plate are put forward and then integrated with vehicle license plate location module to form a vehicle license plate location system.Experimental results show that the system can further improve the performance of the algorithm and is more conducive to the subsequent license plate segmentation and license plate recognition.