Research On License Plate Recognition System Based On OpenCV Under Complex Background

License plate recognition technology is an important part of intelligent transportation system, it has been widely used in highway toll management, road monitoring and vehicle intelligent parking lot management, etc.Therefore,the research of the plate recognition system not only have great theory significance,but also have import practical economic value.This paper analyzes and summarizes the domestic and foreign scholars on the license plate recognition technology research, based on the latest computer vision library OpenCV, designed and developed a vehicle license plate recognition system under complex background.The license plate recognition system designed in this paper includes three important components:license plate detection, license plate location, license plate character recognition.The license plate detection includes the license plate correction algorithm, and the character segmentation algorithm is included in the character recognition.The main contents of this paper are as follow.(1) Research on the license plate detection algorithm.It is designed to detect and segment the candidate license plate region from the input image.In this paper, the license plate detection algorithm is designed according to the license plate color and the texture characteristics.Detection algorithm based on color information converts color images to HSV color space.It detects the range of color components and gets different binary image based on license plate color.In the process of detecting the license plate.first the image is pre processed by image denoising and image grayscale,and then the vertical edge detection algorithm and threshold method are used to get the binary image.The binary images obtained from the license plate detection algorithms are processed by morphological methods with different parameters.By using the method of contour detection, rectangle search and fracture plate reconstruction to process the results, we obtain the candidate plate regions.Based on the characteristic information of vehicle license plate, this paper designs the method of size verification and canny verification to remove partial interference candidate area.Finally, using the Hof transform and affine transform algorithm to tilt the candidate license plate region,then get the final license plate candidate region.(2) Research on license plate location algorithm.It is designed to accurately locate the license plate in the candidate regions obtained from the plate detection algorithm.In this paper, the RBF kernel function support vector machine is used as the classifier.Considering that the license plate region contains more edge features than the non license plate region, the multi scale segmentation HOG method is used to extract features.The real license plate area is determined by training the license plate location model to classify the license plate candidate regions.(3) Research on license plate character recognition algorithm.It is designed to recognize the license plate characters from the plate regions, which are obtained from the license plate location algorithm. Character recognition algorithm includes plate character segmentation and character recognition.In this paper, a preprocessing method is proposed, which can be used to deal with the threshold according to different colors of characters.At the same time, using improved line scan method to remove plate border and rivets.At last, this paper proposes a method to achieve character segmentation based on the second character position and the character of license plate.In this paper, four types of character recognition models are designed based on the elastic BP neural network, which is used to identify the different position characters of the license plate.For Chinese characters and similar characters feature extraction, this paper proposed a method of using combination of binary image features and corresponding gray image features, improved the traditional single image feature extraction method.At last, this paper designs the task pool and the thread pool model to realize the character recognition parallelization, which greatly improves the processing speed of the algorithm.