Article Directory
1 Introduction
In recent years, with the development requirements of traffic modernization, the automatic license plate recognition technology has been paid more and more attention by people. In the automatic license plate recognition technology, license plate positioning, character cutting, character recognition and post-processing are the key technologies. and memory size limitations, the previous license plate recognition is mostly based on the recognition technology of grayscale image processing. Among them, it is first required to detect the license plate area correctly and reliably, so many methods have been proposed, such as Hough transform to detect the straight line to extract the license plate boundary region, using gray-scale segmentation and region growth for region segmentation, or using texture feature analysis techniques, etc. The Hough transform method will greatly increase the possibility of failure when the license plate area is deformed or the image is defaced, while the gray-scale segmentation is more efficient than the straight line The detection method should be stable, but when the image has many areas that are very similar to the gray scale of the license plate, the method is powerless. When texture analysis encounters other interference similar to the texture features of the license plate, the correct rate of license plate positioning is also low. Will be affected. This paper proposes a color segmentation method based on the color information of the license plate.
2. Implementation steps
The main modules of this paper are as follows: color information extraction, license plate area positioning, recognition, extraction, detection inclination, license plate correction, license plate area binarization, erasing interference areas, text segmentation, template matching, and result output.
1) Color information extraction
This system recognizes the license plate with white letters on the blue background of small family cars. According to the RGB ratio of the color image, the candidate regions that are approximately blue are located. However, since the Euclidean distance between two points in the RGB three-primary color space is not linearly proportional to the color distance, it cannot be well controlled when setting the positioning range of the blue area. The positioning error caused by this is the most important. In this way, the recognition rate will drop when there are more blue backgrounds in the picture, and the license plate area cannot be effectively extracted. In this paper, an adaptive adjustment scheme is proposed. Identify and adjust the segmented regions. According to the aspect ratio, the blue-white ratio performs multiple positioning on the candidate area. Finally found the license plate area.
2) Tilt correction
In this paper, the rando algorithm is used to calculate the tilt angle for the tilted picture, and the tilted picture is corrected. Thus, a picture with the same horizontal direction can be obtained. It is beneficial to the later image segmentation and image recognition.
3) Character segmentation
Calculate the color segmented image of the license plate area, project the white horizontally and vertically, calculate the horizontal and vertical peaks, and detect a reasonable character aspect ratio. The same method as the area segmentation can be used to delete and merge the peaks. But in the character When cutting, the character cutting is often inaccurate due to poor acquisition of the threshold value. In this case, the prior knowledge of the license plate format can be used to conduct statistical analysis on the width of the cut characters to guide the cutting, and to correct the wrong cutting process. Wide characters are split. For the word 'gui', it often appears to separate the radicals next to the wooden character and on the right. To solve this problem, the system compares the width of the segmented font with the width of the entire license plate, and merges misoperated characters. An intelligent recognition system should reduce the system's over-reliance on thresholds.
4) Font recognition
A common practice is to use a neural network model to train the system. However, this approach increases the complexity of the system and is not suitable for occasions with high real-time requirements. A simple template matching algorithm is used here. Due to the effective processing in the early stage, the clarity and integrity of the divided fonts can be maintained at a high level. It is beneficial to improve the success rate of template matching. It has been verified that the recognition rate can reach 95% for non-slanted pictures, and more than 90% for skewed pictures. D–0, 6–8, 2–Z, A—4 are characters that are easier to recognize and make mistakes.
5) Voice broadcast
After the characters are correctly recognized, the recording of each character is played according to the corresponding character sequence. When there is an error in the recognition of the license plate area or the error in character segmentation, the program pauses and there is a voice reminder.
6) Store data
After the playback is over, the corresponding recognized characters are stored in the EXCEL file in the specified folder. And store the corresponding time at the same time.
3. Effect display
Example:
Locating license plate:
Grayscale processing:
Tilt correction:
binary processing, noise removal:
precise character positioning:
segmented characters:
recognition result prompt box:
Excel table record:
4. Summary
Advantages: Automatic area positioning, text extraction, and recognition of license plates. After the result is recognized, a voice reading reminder will be given. Voice warnings can be given when positioning errors and recognition errors occur. The recognition result is stored in a file within the specified file.
5. Complete source code download
Graduation design based on Matlab license plate recognition system + GUI interface (complete source code + project report + explanatory document + data), click the link to download: https://download.csdn.net/download/m0_62143653/87707450
