| The purpose of this paper is to design a integrated license plate detection and recognition system that could be applied in the whole environment.The designed system of this paper could detect and recognize the license plate automatically in the extreme environment such as low light and low contrast.Compared with the traditional road equipment,it has stronger robustness,faster detection speed and higher efficiency.The recognition accuracy provides a more efficient and flexible detection and recognition method for normalized traffic infor-mation acquisition.First of all,for the license plate detection part,the traditional method is often not suitable for common life scenes(such as using handheld devices for shooting),so improved is used MSER method as a high robust method to detect the text in the picture,through the exper-imental proof and compared with the traditional detection methods such as edge detection method or color detection method,it is confirmed that the method has stronger robustness,more accurate detection results and higher efficiency.Then,the basic image processing such as binarization,closing operation,contour extraction and so on are carried out for the detec-tion results,and the contour is calculated to determine whether it meets the requirements,and then the size is unified after rotation.Then,the LBP algorithm is used to extract the texture features of the image,and the texture vector of the image is obtained,which is then input into the SVM model to train and predict the license plate detection.Secondly,on the basis of the previous license plate detection results,the license plate char-acter segmentation of the license plate pattern is processed,and the spatial Otsu threshold method is used to make the image segmentation have better binary effect when facing the uneven illumination image.Then the trained ANN model is applied to judge and recognize the segmented characters,and plates text recognition results are output in order.In the ANN model,momentum gradient descent method based on exponential weighted average is used,and double ANN model is used to recognize English numeric characters and Chinese char-acters respectively,and good recognition results are obtained. |
