License plate recognition is a pattern recognition technology that uses the dynamic video or static image of the vehicle to automatically recognize the license plate number and the color of the license plate. The core of the technology includes license plate location algorithm, license plate character segmentation algorithm and optical character recognition algorithm.
The working principle of license plate recognition technology
Vehicle detection: Buried coil detection, infrared detection, radar detection technology, video detection and other methods can be used to sense the passing of the vehicle and trigger image capture.
Image collection: Real-time and uninterrupted recording and collection of passing vehicles through the high-definition camera capture host.
Preprocessing: noise filtering, automatic white balance, automatic exposure and gamma correction, edge enhancement, contrast adjustment, etc.
License plate location: Scan the rows and columns on the grayscale image after image preprocessing to determine the license plate area.
Character segmentation: After locating the license plate area in the image, through grayscale, binarization, etc., the character area is accurately located, and then the character segmentation is performed according to the character size characteristics.
Character Recognition: The segmented characters are scaled, feature extraction, and the standard character expression form in the character database template is matched and judged.
Result output: output the result of license plate recognition in text format.
License plate recognition technology workflow
The license plate recognition system adopts a highly modular design, which treats each link of the license plate recognition process as an independent module.
One, vehicle detection and tracking module
The vehicle detection and tracking module mainly analyzes the video stream, judges the position of the vehicle in it, tracks the vehicle in the image, and records a close-up picture of the vehicle at the best time of the vehicle position. Thanks to the addition of the tracking module, the system can be very good To overcome various external interferences, make more reasonable recognition results, detect unlicensed vehicles and output the results.
2. License plate positioning module
The license plate location module is a very important link and the basis of the follow-up link. Its accuracy has a huge impact on the overall system performance. The license plate system completely abandons the previous algorithmic ideas, and realizes a new algorithm for license plate positioning based on the fusion of multiple features based on learning, which is suitable for various complex background environments and different camera angles.
3. License plate correction and fine positioning module
Due to the limitation of shooting conditions, the license plate in the image is always inevitably tilted. A correction and fine positioning link is required to further improve the quality of the license plate image and prepare for the segmentation and recognition module. Using a well-designed fast image processing filter, not only the calculation is fast, but also the overall information of the license plate is used to avoid the influence of local noise. Another advantage of using this algorithm is that through the analysis of multiple intermediate results, the license plate can also be precisely located, which further reduces the impact of non-license plate areas.
Four, license plate segmentation module
The license plate segmentation module of the license plate system utilizes the grayscale, color, edge distribution and other characteristics of the license plate text, which can better suppress the influence of other noise around the license plate, and can tolerate the license plate with a certain tilt angle. This algorithm is conducive to applications such as mobile inspections where the license plate image is noisy.
Five, license plate recognition module
In the license plate recognition system, a combination of multiple recognition models is usually used for license plate recognition, and a hierarchical character recognition process is constructed, which can effectively improve the accuracy of character recognition. On the other hand, before character recognition, the use of computer intelligent algorithms for pre-processing of character images can not only preserve image information as much as possible, but also improve image quality, improve the distinguishability of similar characters, and ensure the reliability of character recognition.
Six, license plate recognition result decision-making module
Recognition result decision-making module, specifically, the decision-making module uses the historical record left by the process of a license plate through the field of vision to make intelligent decision-making on the recognition result. It obtains the comprehensive credibility evaluation of the license plate by calculating the number of observation frames, the stability of the recognition result, trajectory stability, speed stability, average credibility, and similarity, so as to determine whether to continue tracking the license plate or output Identify the result, or reject the result. This method comprehensively utilizes the information of all frames, reduces the accidental errors caused by the previous recognition algorithms based on a single image, and greatly improves the recognition rate of the system and the correctness and reliability of the recognition results.
Seven, license plate tracking module
The license plate tracking module records various historical information such as the position, appearance, recognition result, credibility, etc. of the license plate in each frame of the vehicle’s driving process. Because the license plate tracking module uses a motion model with a certain fault tolerance and an update model, those license plates that are blocked or blurred for a short time can still be tracked and predicted correctly, and only one recognition result will be output in the end.
