Automatic license plate recognition edge operation method and system

Abstract

The invention discloses an automatic license plate recognition edge operation method and system, the method is executed in a terminal system arranged in a vehicle, the system is mainly provided with a photographing module and an operation circuit, and the method comprises the following steps: generating a streaming film through the photographing module, namely judging a license plate in an image frame by frame by using an object detection model; characters on a license plate are recognized after a license plate image is obtained, and confidence coefficients of the characters and confidence scores of all the characters are calculated by using a confidence coefficient algorithm. Then, license plate confidence scores in continuous frames are calculated within a period of time, the frame with the highest confidence score is taken, a complete frame image is stored in a memory, and then vehicle attributes such as a vehicle jurisdiction, a color, a license plate and a vehicle type can be identified by using a classification model.

Description

Technical Field

[0001] The specification discloses an automatic license plate recognition method running on a terminal device, specifically referring to an automatic license plate recognition edge computing method and system that uses an intelligent model to perform license plate information processing. Background Art

[0002] With the development of artificial intelligence (AI) technology, image recognition has become a mature technology. For example, license plate recognition, an automatic license plate recognition (ALPR) system is commonly used in parking lots and roads for technological law enforcement. However, the image recognition model trained by neural networks often requires powerful computing power and is not suitable for terminal devices with relatively low computing power. It is also not conducive to the terminal device performing license plate recognition under various environmental changes. Summary of the invention

[0003] Given that automatic license plate recognition technology requires a large amount of computing resources and cannot be used in edge devices such as dash cams, and that automatic license plate recognition technology also has to face the needs of license plate recognition in various scenarios, such as weather changes, brightness changes, and other various environments, the specification discloses an automatic license plate recognition edge computing method and system, wherein the method is applicable to edge devices such as Internet of Things (IoT) devices and dash cams that do not have sufficient computing power to cope with real-time computing of high-resolution and high-frame rate streaming videos.

[0004] According to an embodiment of the edge computing method for automatic license plate recognition, the method is executed in an edge device, and a streaming video is obtained by shooting with a camera module in the edge device, and then the processor therein uses an object detection model to determine one or more license plates frame by frame, and obtains each license plate image in each frame. After that, the characters in each license plate can be identified from the license plate image frame by frame, and the confidence of each character is calculated by a confidence algorithm, and then a confidence score is obtained frame by frame based on the confidence of all characters in each license plate.

[0005] Then, the confidence scores of each license plate can be calculated multiple times within a period of time in consecutive frames to obtain the license plate image with the highest confidence score, thereby obtaining a recognition result, and then storing a complete frame image corresponding to the highest confidence score.

[0006] Furthermore, an intelligent model is used to identify the characters in each license plate, and an intelligent algorithm is used to calculate the confidence of each character in each license plate. Preferably, the confidence score of each license plate is a statistical value of the confidence of multiple characters in the license plate.

[0007] When calculating the confidence score of the license plate, the number of characters in the recognized license plate should exceed the character number threshold set by the system. However, if the number of characters in the license plate is lower than the character number threshold, it will be regarded as an unreliable recognition result and excluded from the license plate tracking program.

[0008] Furthermore, when the frame image with the highest confidence score for each license plate is obtained, the vehicle information is obtained from the frame image using an image processing technique and written into the metadata of the frame image.

[0009] Furthermore, the object detection model is run in the edge device to obtain the bounding box of the license plate of each vehicle in each frame, calculate the probability that the bounding box is the license plate and the probability that each license plate belongs to various vehicle categories, so as to perform character recognition of each license plate.

[0010] Furthermore, the edge device runs a classification model to obtain the probability that each license plate is from a different vehicle jurisdiction, the probability that each vehicle is of a different color, and the probability that each vehicle is of a different brand and model, so as to identify the attributes of each vehicle.

[0011] Furthermore, after identifying each license plate character in the continuous frames, a license plate tracking program is executed, wherein, after identifying one of the license plates frame by frame, when the target license plate in the first frame is selected, the distance between the target license plate in the first frame and each of the one or more license plates identified in the second frame is calculated, and one or more license plate distances are obtained accordingly. Then, the character string in the target license plate in the first frame and the one or more license plates in the second frame are identified to calculate the character string similarity between the character string in the target license plate in the first frame and the character string in each license plate in the second frame.

[0012] Afterwards, a comprehensive score is calculated using the license plate distance and string similarity between the target license plate in the first frame and each license plate in the second frame, so that the system can determine whether the target license plate in the first frame is any license plate appearing in the second frame based on the comprehensive score. When one of the license plates in the second frame is determined to be the target license plate selected in the first frame based on the comprehensive score, the same identifier is assigned to perform license plate tracking in consecutive frames.

[0013] Furthermore, after tracking a license plate with the same identifier in consecutive frames for a certain period of time, the system will store the complete frame image with the highest confidence score, or store multiple complete frame images with confidence scores reaching a threshold.

[0014] According to an embodiment of the automatic license plate recognition edge computing system, the system mainly includes a camera module, a memory and a processor, in which the above-mentioned automatic license plate recognition method is executed.

[0015] When the frame image with the highest confidence score among the continuous frames is obtained from the automatic license plate recognition method, the frame image and the recognized license plate number are transmitted to an external system, such as a computer device or a cloud system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1A and Figure 1B A schematic diagram showing the use of edge computing methods for automated license plate recognition;

[0017] Figure 2 A diagram showing an embodiment of the architecture of a license plate recognition edge computing system;

[0018] Figure 3 A schematic diagram showing an implementation example of a license plate recognition edge computing system;

[0019] Figure 4 A flowchart showing an embodiment of an edge computing method for automated license plate recognition;

[0020] Figure 5 A flowchart showing an embodiment of object detection in an edge computing method for automated license plate recognition;

[0021] Figure 6 Flowchart 1 of an embodiment showing the classification of objects in the edge computing method of automatic license plate recognition;

[0022] Figure 7 Demonstrating the process of classifying objects in the edge computing method for automated license plate recognition Figure 2 ;as well as

[0023] Figure 8 A flow chart showing an embodiment of license plate tracking in an automated license plate recognition edge computing method. DETAILED DESCRIPTION

[0024] The present disclosure proposes an automatic license plate recognition (ALPR) edge computing method and system. The method is executed in a terminal device, wherein the main components of the system include a camera module, a processor, and a memory, etc., and the proposed method is suitable for terminal devices with relatively low computing power. By using edge computing, artificial intelligence (AI) can be used to detect the license plate position in the image, perform license plate string recognition, and track the license plate based on the determined license plate. In the process of automatic license plate recognition, the recognition confidence of the license plate string is calculated in real time, and a screenshot can be taken for the frame with the highest recognition confidence, and then the attribute information about the vehicle in the screenshot is obtained, such as the license plate character (plus U.S. state recognition), vehicle type, brand (manufacturer), and artificial intelligence methods can also be used to identify the vehicle brand, color, etc. based on the vehicle appearance curve, and finally it can be recorded in the metadata of the image.

[0025] According to the scope of application of the edge computing method of automatic license plate recognition, you can refer to Figure 1A and Figure 1B Schematic diagram of the scenario in which the edge computing method for automated license plate recognition is applied.

[0026] Figure 1A The first frame 101 is shown as being taken by a camera module in a terminal device (such as a driving recorder). The first frame 101 shows that there are vehicles 11, 12, and 13 in front of the rear vehicle where the terminal device is installed. The object detection model running in the terminal device determines the license plate 110 of vehicle 11, the license plate 120 of vehicle 12, and the license plate 130 of vehicle 13 in the first frame 101. The vehicle 11 is set as the target vehicle, and the license plate 110 of the vehicle 11 is set as the target license plate.

[0027] then Figure 1B The second frame 102 is a schematic diagram showing an image. The second frame 102 still includes the vehicle 11 set as the target vehicle, and the license plate 110 is continuously captured. In the automatic license plate recognition edge computing method, the license plate recognition technology is used to track the vehicle 11 with the same license plate 110 in the continuous frames. In this example, according to the license plate recognition result, the second frame 102 no longer has the vehicles 12 and 13 parked on the roadside. The license plate recognition result determines that another vehicle 14 (license plate 140) parked on the roadside is present, and the vehicle 11 continues to appear.

[0028] Figure 2 The diagram shows an implementation example of a system architecture for executing an automated license plate recognition edge computing method, wherein the license plate recognition edge computing system 20 shown therein may be a circuit system disposed in a terminal device, including a photographic module 201 for capturing images in collaboration with hardware and software, a computing module 203 for executing a license plate recognition program, and a communication module 205 for transmitting data, and is provided with a memory 207 for storing images.

[0029] The streaming video captured by the camera module 201 is processed by the computing module 203 , which includes an object detection model 231 for identifying the location of the license plate and a classification model 233 for performing license plate information recognition.

[0030] The object detection model 231 uses a machine learning algorithm to deeply learn the features of specific objects in the image from a large number of training images, and implements a model for predicting the positions and categories of various objects in the image. Taking the license plate as an example, the general features of the license plate include a rectangular appearance, in which the combination of the glyphs of the characters and the English numbers has specific features, plus some additional information, such as the jurisdiction of the vehicle and others.

[0031] So, you can refer to Figure 5 The flowchart of the embodiment of object detection in the edge computing method of automatic license plate recognition is shown. The object detection model is executed by the processor in the system (step S501). That is, the object detection model 231 obtained through the above feature training can obtain the bounding box of the license plate of each vehicle from each frame image (step S503), and calculate the probability that the bounding box is the license plate (step S505). The position of the license plate can be confirmed according to the result of the calculation rate of the object detection model 231, and then the characters of each license plate are recognized according to the license plate position obtained in the above steps, as well as the position and character type of each character therein (step S507).

[0032] When the license plate character recognition is completed, because the license plate number is generally a string of multiple characters, the system can further confirm whether a sufficient number of characters are recognized, and compare it with the minimum number threshold set by the system, that is, the character number threshold, to determine whether the number of characters in the recognized license plate exceeds this character number threshold (step S509). If the number of characters in the license plate is lower than this character number threshold (no), it will be regarded as an unreliable recognition result, the character recognition result will be discarded, and the vehicle will be excluded from the subsequent license plate detection and license plate tracking procedures (step S511); if the number of license plate characters recognized is sufficient and exceeds the character number threshold (yes), the character recognition result can be output (step S513).

[0033] The classification model 233 also uses machine learning algorithms to deeply learn the information features of various license plates from a large number of training images. The trained classification model 233 runs in the license plate recognition edge computing system to obtain the probability that each license plate belongs to different vehicle jurisdictions (such as the location of the vehicle's regulatory unit and government agency), the probability that each vehicle belongs to a different color, and the probability that each vehicle belongs to a different brand and model, so as to identify the attributes of each vehicle. Taking the US license plate as an example, the classification model 233 can determine the state government to which the license plate belongs, and can identify the vehicle color, car manufacturer or manufacturer, and model from the vehicle image. The obtained vehicle information will form a record in the metadata of the vehicle image.

[0034] The license plate recognition edge computing system 20 is provided with a communication module 205 for establishing a connection with the external system 22, so that the license plate recognition edge computing system 20 can obtain a key image with the highest confidence score (such as a frame corresponding to the highest score) in the continuous frames, and transmit the relevant image and the recognized license plate number and other data to the external system 22. According to the embodiment, the external system 22 can refer to a computer device connected to the driving recorder by wire or wirelessly, or a cloud system.

[0035] Furthermore, the operating programs and various intelligent modules of the license plate recognition edge computing system can be referred to Figure 3Schematic diagram of the embodiment, which shows various software function modules implemented by using various intelligent models (such as the above-mentioned object detection model 231 and classification model 233) in the license plate recognition edge computing system and cooperating with hardware circuits. The processes running in the system can also refer to Figure 4 The flowchart of an embodiment of the edge computing method for automatic license plate recognition is shown.

[0036] The license plate recognition edge computing system can be installed in a terminal device such as a driving recorder. When a streaming video composed of continuous frames is obtained through the camera module 31 (step S401), the license plate recognition program 311 executed in the license plate recognition module 33 can identify the position of one or more license plates in the image (step S403). According to the above embodiment, the license plate recognition program 311 executes an object detection module to obtain the bounding box of the license plate of each vehicle in each frame, the probability that the bounding box is a license plate, and the probability that each license plate belongs to various vehicle categories. Then, based on the recognition result of the license plate position, the license plate image is captured (step S405) and temporarily stored in the memory. Then, the license plate string recognition program 313 that runs the character detection technology in the system performs character recognition of each license plate (step S407).

[0037] Among them, after completing the license plate recognition, the image of the license plate can be obtained. Generally, the license plate image is much smaller than the original frame size, so the system only captures the license plate image to make subsequent character recognition more efficient and reduce the computing power required for image processing. Afterwards, the license plate identified in each frame of the image can be tracked by the license plate tracking module 35 according to the license plate position. For example, when the position of the target license plate in the first frame of the continuous frames is determined, the system can assign an identification information and obtain a license plate screenshot; similarly, when the target license plate with the same identification information is determined in the second frame, a license plate screenshot is also obtained, and the license plate tracking can be performed.

[0038] While tracking the license plate, the characters of the license plate can be further identified. The confidence calculation module 37 in the system executes a confidence algorithm to calculate the confidence level of each character for each character obtained by recognizing each license plate frame by frame (step S409). The confidence level of the character is about the probability of recognizing each character on the license plate by an intelligent model (a character recognition model that can recognize the letters and numbers that make up the license plate number) and calculating it as a certain letter or number by an intelligent algorithm. The system can set a probability threshold (or confidence threshold), and the character with the highest probability greater than this probability threshold is the recognition result of the character.

[0039] Next, after the confidence of all recognizable letters (all or more than a certain number of characters) in the license plate is calculated, the confidence calculation module 37 can then obtain the confidence score of the entire license plate character recognition frame by frame according to the confidence of all characters in each license plate (step S411), and the confidence score of the license plate is a statistical value of the confidence of multiple characters in the license plate, such as calculating the average value of the confidence of all characters, or adding up to obtain a total, or a comprehensive score through weight setting (for example, giving a higher weight to those with higher confidence). It is mentioned here that the statistical method of the confidence score of all or multiple characters in the license plate is not limited to the above.

[0040] The above steps S411 and S413 can be used to calculate the confidence score of each license plate in multiple consecutive frames (step S413). When the confidence scores of each license plate can be calculated multiple times within a period of time, the one with the highest confidence score during this period of time can be continuously counted (step S415), or the confidence score of the license plate can be calculated at intervals in consecutive frames, and the frame image with the highest confidence score can be obtained before the license plate disappears. At this time, when the same license plate appears in multiple consecutive frames, the confidence score of each frame is calculated, and the one with the highest confidence score can be obtained. The recognition result of the license plate with the highest confidence score can be used as the recognition result (step S417). Therefore, if multiple license plates appear in the same multiple consecutive frames, the recognition result with the highest confidence score is obtained for each license plate, and the corresponding complete frame is stored for the license plate (step S419). Figure 3 In the embodiment shown, the complete frame image corresponding to the license plate with the highest confidence score is provided to the vehicle detection program 391 in the vehicle information processing module 39 according to the result obtained by the confidence calculation module 37. On the other hand, the frame image with the highest confidence score can be obtained for the license plate, and the vehicle information can be obtained from the frame image by image processing technology and written into the uniquified metadata of the frame image.

[0041] Another application is that when a frame image with the highest confidence score is obtained in consecutive frames, the frame image and the recognized license plate number can be transmitted to an external system. For example, the external system can be a computer system on the vehicle or a remote control center. The vehicle information recognized by edge devices such as terminal devices (such as driving recorders) can be stored and marked on a display in real time for real-time monitoring by the external system.

[0042] It is worth mentioning here that Figure 5 In the process shown, when calculating the confidence score of the license plate, a sufficient number of characters must be recognized to be considered an effective recognition. If the number of characters in the license plate is too small, it will be regarded as an unreliable recognition result and excluded from the subsequent license plate tracking process.

[0043] The system identifies the contents of the license plate image captured from the continuous frames through the confidence calculation module 37, including the license plate image with the highest confidence score and the complete frame corresponding to the license plate with the highest confidence score. The complete frame and the captured license plate image are provided to the vehicle information processing module 39, wherein the complete frame is provided to the vehicle detection program 391 for vehicle detection. After the vehicle detection is completed, the vehicle image can be captured, and then the vehicle attribute determination program 393 is used to determine the vehicle attributes, such as color 332 and car manufacturer 333. On the other hand, the license plate image can be identified by the vehicle attribute determination program 393 for the vehicle jurisdiction (such as state 331). Finally, the vehicle attributes (such as state 331, color 332 and car manufacturer 333) obtained by the vehicle attribute determination program 393 can be stored in the storage module 38 along with the vehicle image.

[0044] Furthermore, in the vehicle attribute determination program 393, the following is performed: Figure 6 and Figure 7 The flowchart of an embodiment of classifying objects in the edge computing method of automatic license plate recognition is shown.

[0045] The first is the processing of license plate images, such as Figure 6 As shown, when the confidence calculation module 37 obtains the confidence score of the character recognition in each license plate in the continuous frames, the one with the highest confidence score is used as the license plate recognition result. In addition to storing the corresponding frame with the highest confidence score for each license plate, the classification model is also executed (step S601).

[0046] According to an embodiment of the automated license plate recognition edge computing system, a processor in an edge device performs image processing to obtain vehicle information. According to one embodiment, a classification model derived from the image features of all information in a license plate can be used to learn deeply, wherein the probability of each license plate being in a different vehicle jurisdiction can be calculated by an intelligent model (step S603). Taking the US license plate as an example, an intelligent model derived from the features of each state license plate in a large amount of license plate image data is used to learn deeply, and the state of the vehicle with the license plate having the highest confidence score is identified from the calculated probability of the license plate for different states 331.

[0047] Similarly, Figure 7 As shown, a classification model is executed for vehicle images (step S701), including calculating the probability of each vehicle being of a different color using the classification model (step S703). According to an embodiment, an intelligent model is obtained by deep learning the characteristics of vehicle colors in a large amount of vehicle image data, and the color 332 of the vehicle with the license plate having the highest confidence score is identified from the calculated probabilities of the vehicle being of various colors.

[0048] The classification model is then used to calculate the probability that each vehicle belongs to a different brand (such as brand 333) (step S705). Similarly, an intelligent model derived from deep learning of the features of each state license plate in a large amount of license plate image data can be used to obtain the outline of the vehicle from the vehicle image and calculate the probability of the vehicle belonging to different brands and models, thereby identifying the vehicle's brand 333 and its model.

[0049] According to the above classification process, the system can obtain the vehicle attributes based on the probability of the vehicle's jurisdiction, color, brand and model. Figure 3 As shown, the vehicle information processing module 39 obtains the vehicle information through the classification model and stores it in the storage module 38 together with the vehicle image and the license plate image.

[0050] The system identifies the content of the license plate image obtained by screenshots from continuous frames through the confidence calculation module 37, and provides it to the vehicle information processing module 39, including providing it to the vehicle detection program 391 to obtain the vehicle corresponding to the identified license plate in the image, complete the vehicle detection, and can screenshot the vehicle image to provide the purpose of subsequent vehicle attribute judgment.

[0051] When the automatic license plate recognition edge computing method running on the terminal device recognizes each license plate character in consecutive frames, license plate tracking can be performed. For details, please refer to Figure 8 The flowchart of an embodiment of license plate tracking in the automatic license plate recognition edge computing method is shown.

[0052] according to Figure 8 The license plate tracking program (step S801) shown in the figure uses the intelligent model of license plate recognition from the memory (such as Figure 2 After obtaining continuous frames from the memory 207 shown in the figure, one or more license plates in the image are identified frame by frame according to the license plate features, and each license plate is assigned identification information. First, a license plate in one of the frames (set as the first frame at present) is set as the target license plate (step S803), and then the image of one or more license plates in a frame (set as the second frame, representing one or more frames before and after the first frame) in which the license plate has been identified is obtained from the memory. The distance between the target license plate in the first frame and each of the one or more license plates identified in the second frame can be calculated to obtain one or more license plate distances (step S805).

[0053] According to the embodiment, in order to track the same license plate, a distance relationship is calculated between the target license plate of the first frame and each license plate identified in the second frame in this process, and whether the target license plate of the first frame also appears in the second frame is determined based on this distance relationship. Specifically, for dynamic continuous frames, even if it is the same object (such as a license plate), in the continuously changing continuous frames, the position of an object in the first frame (such as the plane coordinates in the first frame) and the same object in the second frame will have a distance change, so the distance between objects in different frames can be used as one of the bases for determining whether the target license plate of the first frame also appears in the second frame.

[0054] However, the distance cannot be the only basis for uniquely identifying the same license plate in the front and back frames, so the similarity between the string in the target license plate of the first frame and the string in one or more license plates of the second frame (referring to the other front and back frames other than the first frame) is also identified (step S807). According to the embodiment, the string of the target license plate of the first frame and the string of other license plates and the string of all license plates in the second frame are first identified. A method for calculating the similarity between the string of the target license plate in the first frame and the string of the license plate in the second frame can adopt a Levenshtein edit distance calculation method. The technical concept of judging the similarity based on the Levenshtein edit distance is based on the minimum number of edits required between the two strings to convert the first string into the second string. The fewer the number of edits, the higher the similarity. If more edits are required to complete the conversion, the string similarity is low. When the edit distance is normalized to between 0 and 1, when the two strings are exactly the same, the edit distance is 0; when the two strings are completely different, the edit distance is 1.

[0055] After the license plate distance and string similarity between the target license plate in the first frame and each license plate in the second frame are obtained, a comprehensive score can be calculated based on the license plate and string similarity (step S809), and the comprehensive score can be used to determine whether the target license plate in the first frame is any license plate that appears in the second frame. In the process, the system sets a comprehensive score threshold, and can compare the comprehensive score to determine whether the target license plate in the first frame appears in other frames (step S811).

[0056] If not, it means that the target license plate in the first frame does not appear in the second frame (No), indicating that the license plate appearing in the image is a new license plate, that is, it is assigned new identification information (step S813). In this way, the above process is repeated, and another target license plate can be selected in the first frame again, and then it is determined whether the target license plate continues to appear in the second frame. If it is determined based on the comprehensive score that the target license plate in the first frame appears in the second frame (Yes), the same identification information can be assigned to the same vehicle in different frames (step S815), indicating that it is the same vehicle, and the same identification information is marked for the same license plate in subsequent consecutive frames, completing the goal of license plate tracking (step S817).

[0057] Specifically, when tracking a license plate with the same identifier in consecutive frames, after a certain period of time, the system can store the complete frame image with the highest confidence score, or store multiple complete frame images in which the confidence scores of the character strings reach a certain threshold.

[0058] In summary, according to the automated license plate recognition edge computing method and system embodiments, the intelligent models used in the method can be executed by edge devices, wherein the license plate position is first identified to obtain the license plate image, and then the license plate string recognition is performed. Because only the frame image with the best recognition result is used for subsequent processing, a large amount of computing power requirements are reduced, making the method suitable for terminal devices with lower computing power.

[0059] The contents disclosed above are only preferred feasible embodiments of the present invention, and are not intended to limit the scope of the present invention. Therefore, all equivalent technical changes made using the contents of the present invention's specification and drawings are included in the scope of the present invention.

Claims

1. An edge computing method for automatic license plate recognition, executed in an edge device, characterized in that: The method comprises: Get a stream of videos; Determine one or more license plates frame by frame, and obtain the image of each license plate in each frame; Recognize the characters in each license plate frame by frame, calculate a confidence level for each character, and then obtain a confidence score based on the confidence levels of all characters in each license plate frame by frame; In continuous frames, the confidence scores of each license plate are calculated multiple times within a period of time, and a recognition result is obtained according to the license plate image with the highest confidence score; and A complete frame image corresponding to the highest confidence score is stored.

2. The edge computing method for automatic license plate recognition according to claim 1, characterized in that: An intelligent model is used to identify the characters in each license plate, and an intelligent algorithm is used to calculate the confidence of each character in each license plate.

3. The edge computing method for automatic license plate recognition according to claim 2, characterized in that: The confidence score of each license plate is a statistical value of the calculated confidence of multiple characters in the license plate.

4. The automatic license plate recognition edge computing method according to claim 3, characterized in that: When calculating the confidence score of the license plate, the number of characters of the recognized license plate should exceed a character number threshold. If the number of characters of the license plate is lower than the character number threshold, the recognition fails.

5. The edge computing method for automatic license plate recognition according to claim 4, characterized in that: The confidence score of the license plate is calculated at every time interval in continuous frames, and the frame image with the highest confidence score is obtained before the license plate disappears.

6. The automatic license plate recognition edge computing method according to claim 5, characterized in that: After obtaining the frame image of each license plate having the highest confidence score, a vehicle information is obtained from the frame image by using an image processing technology and written into a metadata of the frame image.

7. The automatic license plate recognition edge computing method according to claim 1, characterized in that: The edge device runs an object detection model to obtain a bounding box of the license plate of each vehicle in each frame, calculate the probability that the bounding box is the license plate and the probability that each license plate belongs to various vehicle categories, so as to perform character recognition on each license plate.

8. The edge computing method for automatic license plate recognition according to claim 1, characterized in that: The edge device runs a classification model to obtain the probability of each license plate being of a different vehicle jurisdiction, the probability of each vehicle being of a different color, and the probability of each vehicle being of a different brand and model, so as to identify the attributes of each vehicle.

9. The edge computing method for automatic license plate recognition according to any one of claims 1 to 8, characterized in that: After identifying each license plate character in the continuous frames, a license plate tracking procedure is executed, including: After identifying one of the license plates frame by frame, a target license plate in the first frame is selected; Calculating the distance between the target license plate in the first frame and each of the one or more license plates identified in a second frame to obtain one or more license plate distances; Identify character strings in the target license plate of the first frame and the one or more license plates of the second frame; Calculating a string similarity between the string in the target license plate of the first frame and the string in each license plate of the second frame; Calculating a comprehensive score using the license plate distance between the target license plate in the first frame and each license plate in the second frame and the character string similarity; and Determine, according to the comprehensive score, whether the target license plate in the first frame is any license plate appearing in the second frame; When it is determined according to the comprehensive score that one of the license plates in the second frame is the target license plate selected in the first frame, the same identifier is assigned to perform license plate tracking in the continuous frames.

10. The edge computing method for automatic license plate recognition according to claim 9, characterized in that: After tracking the license plate with the same identifier in the continuous frames for a period of time, the complete frame image with the highest confidence score is stored, or multiple complete frame images with the confidence scores reaching a threshold are stored.

11. An automated license plate recognition edge computing system, characterized in that: The system comprises: A photography module; 1. Memory; A processor is electrically connected to the camera module and the memory, and executes an automatic license plate recognition method, including: Producing a stream of video by the camera module; Determine one or more license plates frame by frame, obtain the image of each license plate in each frame and temporarily store it in the memory; Recognize the characters in each license plate frame by frame, calculate a confidence level for each character, and then obtain a confidence score based on the confidence levels of all characters in each license plate frame by frame; and In continuous frames, the confidence score of each license plate is calculated multiple times within a period of time, and a complete frame image corresponding to the highest confidence score is stored in the memory.

12. The automatic license plate recognition edge computing system according to claim 11, characterized in that: The processor runs an object detection model and a classification model obtained through intelligent algorithm training data.

13. The automatic license plate recognition edge computing system according to claim 12, characterized in that: The object detection model is run to obtain a bounding box of the license plate of each vehicle in each frame, and the probability of the bounding box being the license plate and the probability of each license plate belonging to various vehicle categories are calculated to perform character recognition on each license plate.

14. The automatic license plate recognition edge computing system according to claim 12, characterized in that: Running the classification model can obtain the probability of each license plate being of a different vehicle jurisdiction, the probability of each vehicle being of a different color, and the probability of each vehicle being of a different brand and model, so as to identify the attributes of each vehicle.

15. The automatic license plate recognition edge computing system according to claim 11, characterized in that: An intelligent model is used to identify the characters in each license plate, and an intelligent algorithm is used to calculate the confidence of each character in each license plate. The confidence score of each license plate is a statistical value of the calculated confidence of multiple characters in the license plate.

16. The automatic license plate recognition edge computing system according to claim 15, characterized in that: When calculating the confidence score of the license plate, the number of characters of the recognized license plate should exceed a character number threshold. If the number of characters of the license plate is lower than the character number threshold, the recognition fails.

17. The automatic license plate recognition edge computing system according to claim 16, characterized in that: The confidence score of the license plate is calculated at every time interval in continuous frames, and the frame image with the highest confidence score is obtained before the license plate disappears.

18. The automatic license plate recognition edge computing system according to claim 17, characterized in that: After obtaining the frame image of each license plate having the highest confidence score, a vehicle information is obtained from the frame image by using an image processing technology and written into a metadata of the frame image.

19. The automatic license plate recognition edge computing system according to any one of claims 11 to 18, characterized in that: The frame image with the highest confidence score is obtained from the continuous frames, and the frame image and the recognized license plate number are transmitted to an external system.

20. The automatic license plate recognition edge computing system according to claim 19, characterized in that: The external system is a computer device or a cloud system.

Images