License plate camera accuracy self-checking system and method based on multi-modal model

Abstract

The invention provides a license plate camera accuracy rate self-checking system and method based on a multi-modal model, and belongs to the technical field of image recognition. A license plate spot check module is responsible for extracting a preset number of license plate numbers and recognition pictures corresponding to the license plate numbers from recognition results of a license plate recognition camera every day; the result verification module is responsible for submitting the identification pictures to a plurality of third-party large models for identification to obtain identification results and determining whether reminding needs to be carried out or not according to comparison results of the identification results and license plate numbers, and the verification result notification module is responsible for automatically notifying related personnel to check a license plate camera when reminding needs to be carried out. And the recognition processing efficiency of the accuracy abnormality of the license plate camera is improved.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a license plate camera accuracy self-checking system and method based on a multimodal model, which is suitable for scenarios such as intelligent transportation systems and parking lot management systems that require real-time monitoring of license plate recognition accuracy. Background Art

[0002] With the popularization of intelligent transportation systems and parking management systems, license plate recognition cameras serve as core equipment to recognize and process vehicle license plates.

[0003] However, for the vehicle license plate recognition system, its recognition accuracy directly affects the system reliability and user experience. Existing license plate recognition systems usually lack an effective self-checking mechanism and cannot monitor the recognition accuracy in real time, resulting in recognition errors that are difficult to detect and correct in a timely manner.

[0004] In response to the above technical problems, the present application specifically provides a license plate camera self-inspection and accuracy monitoring method based on a multimodal model. Summary of the invention

[0005] To achieve the purpose of the present invention, the present invention adopts the following technical solutions: In a first aspect, the present application provides a license plate camera accuracy self-checking system based on a multimodal model, specifically comprising: License plate sampling module, result verification module, verification result notification module; The license plate sampling module is responsible for extracting a preset number of license plate numbers and their corresponding recognition images from the recognition results of the license plate recognition camera every day; The result verification module is responsible for submitting the recognition images to multiple third-party large models for recognition to obtain recognition results, and using the recognition results to compare with the license plate number to determine whether a reminder is needed; The verification result notification module is responsible for automatically notifying relevant personnel to conduct license plate camera inspection when a reminder is required.

[0006] The beneficial effects of the present invention are: By introducing multimodal model verification, cross-validation of license plate recognition results is achieved, thereby improving the reliability of recognition accuracy.

[0007] The majority voting mechanism can automatically trigger manual intervention when the recognition results are inconsistent, ensuring that the problem is handled in a timely manner.

[0008] Through daily sampling tests, the recognition accuracy of the license plate recognition camera can be monitored in real time to ensure the long-term stable operation of the system.

[0009] A further technical solution is that the third-party model calls the third-party large model through an API interface to perform license plate recognition.

[0010] A further technical solution is to use the comparison result between the recognition result and the license plate number to determine whether a reminder is needed, specifically including: When it is determined that there is no recognition result inconsistent with the license plate number according to the comparison between the license plate number and the recognition results of different third-party models, it is determined that no reminder is needed; When there is a recognition result inconsistent with the license plate number, the third-party model is divided into multiple model groups according to whether the recognition result is consistent. The number of third-party models consistent with the license plate number is increased by 1 as the target number. When there is a model group in which the number of third-party models is greater than the target number, it is determined that a reminder is required.

[0011] A further technical solution is that the verification result notification module is triggered by email, text message or system notification.

[0012] In a second aspect, the present application provides a self-checking method for the accuracy of a license plate camera based on a multimodal model, which is applied to the above-mentioned self-checking system for the accuracy of a license plate camera based on a multimodal model, and specifically includes: S1 obtains recognition deviation data of the license plate camera within a preset time period, and uses the recognition deviation data to determine whether the recognition processing accuracy of the license plate camera meets the requirements, and then proceeds to the next step; S2 determines the recognition deviation data of the license plate camera under similar environmental data in the time period based on the environmental data of the license plate camera in different time periods, and determines the recognition deviation probability type of the time period by using the recognition deviation data under similar environmental data in the time period; S3 determines the number of license plates extracted by the license plate camera on the current date based on the recognition data of license plates in time periods of different recognition deviation probability types on the current date; S4 takes the extraction quantity as a constraint condition, and uses the recognition deviation data under similar environmental data in the time period and the recognition data of the license plate to determine the extraction target time period and the number of license plates to be extracted in the time period, and determines whether it is necessary to dynamically adjust the extraction strategy of the current date based on the comparison result of the license plate in the extraction target time period.

[0013] A further technical solution is that the recognition deviation data of the license plate camera includes the recognition deviation times of the license plate camera and environmental data corresponding to different recognition deviation times.

[0014] A further technical solution is that the environmental data includes light intensity.

[0015] A further technical solution is to determine whether the recognition processing accuracy of the license plate camera meets the requirements, specifically including: Determine the number of recognition deviations of the license plate camera using the recognition deviation data of the license plate camera, and determine the number of recognition deviations within different illumination intervals using the illumination corresponding to different recognition deviations. Based on the number of identification deviations in different light intensity intervals, the number of identification deviations in a preset light intensity interval is performed; Whether the recognition processing accuracy of the license plate camera meets the requirements is determined by the number of recognition deviations within a preset light intensity range.

[0016] A further technical solution is that the preset light intensity interval is an interval in which the light intensity is less than a preset light intensity threshold.

[0017] A further technical solution is to determine whether the recognition processing accuracy of the license plate camera meets the requirements by the number of recognition deviations within a preset illumination range, which specifically includes: When the number of recognition deviations within the preset illumination amount interval is greater than the preset number of deviations, it is determined that the recognition processing accuracy of the license plate camera does not meet the requirement.

[0018] A further technical solution is to determine whether it is necessary to dynamically adjust the extraction strategy of the current date, which specifically includes: Based on the comparison results of the license plates in the target extraction period, determining the number of recognition deviations of the license plates in different target extraction periods; Determining recognition deviation ratios for different extraction target periods according to ratios of the number of recognition deviations of license plates in different extraction target periods to the number of extracted license plates; The identification deviation ratio of different extraction target periods is used to determine whether dynamic adjustment of the extraction strategy for the current date is required.

[0019] A further technical solution is to determine whether it is necessary to dynamically adjust the extraction strategy of the current date through the recognition deviation ratio of different extraction target time periods, specifically including: When the average value of the recognition deviation ratios of different extraction target time periods is greater than the preset deviation ratio threshold, it is determined that a dynamic adjustment of the extraction strategy for the current date is required.

[0020] Other features and advantages will be described in the following description. The objects and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and drawings.

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above and other features and advantages of the present invention will become more apparent by describing in detail example embodiments thereof with reference to the accompanying drawings; Figure 1 It is a framework diagram of a license plate camera accuracy self-checking system based on a multimodal model; Figure 2 It is a flow chart of a self-checking method for the accuracy of a license plate camera based on a multimodal model; Figure 3 It is a flow chart for determining that the recognition processing accuracy of the license plate camera meets the requirements; Figure 4 is a flow chart of a method for determining a type of identification deviation probability for a time period; Figure 5 The present invention is a flow chart of a method for determining the number of license plates extracted by a license plate camera on the current date. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0024] The present invention proposes a self-checking and accuracy monitoring method for license plate recognition cameras based on multimodal model verification. Through daily sampling detection, multimodal model verification, result comparison and manual intervention, real-time monitoring and automatic correction of the recognition accuracy of license plate recognition cameras are achieved.

[0025] The technical solution of the present invention is as follows: 1. Daily sampling test: Randomly select once a day from the recognition results of the license plate recognition camera, and record the license plate number A1 and its corresponding recognition image B.

[0026] 2. Multimodal model verification: Submit image B to two third-party large models (such as DeepSeek-Version and Gpt-4V) for recognition to obtain license plate numbers A2 and A3.

[0027] 3. Result comparison: Consistency verification: If A1, A2, and A3 are completely consistent, the recognition result is considered accurate.

[0028] Majority voting mechanism: If the three are inconsistent, the majority voting principle is adopted, that is, the majority result is considered accurate. If A1 is in the minority, manual intervention is automatically triggered to further troubleshoot the problem.

[0029] 4. Human intervention: When A1 is in the minority, the system automatically notifies relevant personnel to conduct checks to ensure the accuracy of the recognition system.

[0030] Example 1 Figure 1 As shown, the present application provides a license plate camera accuracy self-checking system based on a multimodal model, specifically comprising: License plate sampling module, result verification module, verification result notification module; The license plate sampling module is responsible for extracting a preset number of license plate numbers and their corresponding recognition images from the recognition results of the license plate recognition camera every day; The result verification module is responsible for submitting the recognition images to multiple third-party large models for recognition to obtain recognition results, and using the recognition results to compare with the license plate number to determine whether a reminder is needed; The verification result notification module is responsible for automatically notifying relevant personnel to conduct license plate camera inspection when a reminder is required.

[0031] Furthermore, the third-party model calls the third-party large model through an API interface to perform license plate recognition.

[0032] Specifically, the comparison result between the recognition result and the license plate number is used to determine whether a reminder is needed, which specifically includes: When it is determined that there is no recognition result inconsistent with the license plate number according to the comparison between the license plate number and the recognition results of different third-party models, it is determined that no reminder is needed; When there is a recognition result inconsistent with the license plate number, the third-party model is divided into multiple model groups according to whether the recognition result is consistent. The number of third-party models consistent with the license plate number is increased by 1 as the target number. When there is a model group in which the number of third-party models is greater than the target number, it is determined that a reminder is required.

[0033] It should be noted that the verification result notification module is triggered by email, text message or system notification.

[0034] In one specific embodiment, the following is the core code related to this application: import requests def fetch_third_party_recognition(image_path, model_api_url): # Call the API interface of the third-party model for license plate recognition response = requests.post(model_api_url, files={'image': open(image_path, 'rb')}) if response.status_code == 200: return response.json().get('license_plate') else: return None def daily_self_check(license_plate_a1, image_b): # Third-party model validation license_plate_a2 = fetch_third_party_recognition(image_b, 'https: / / api.deepseek.com / recognize') license_plate_a3 = fetch_third_party_recognition(image_b, 'https: / / api.openai.com / recognize') # Result comparison if license_plate_a1 == license_plate_a2 == license_plate_a3: print("The recognition results are consistent, the system is normal.") else: # Majority voting mechanism majority_vote = max([license_plate_a1, license_plate_a2, license_plate_a3], key=[license_plate_a1, license_plate_a2, license_plate_a3].count) if license_plate_a1 != majority_vote: print("The recognition results are inconsistent, triggering manual intervention.") # Notify relevant personnel notify_operator(license_plate_a1, majority_vote) def notify_operator(original_plate, majority_plate): # Send notifications to relevant personnel print(f"Notice: The original recognition result {original_plate} is inconsistent with the majority voting result {majority_plate}, please check the system.") # You can add email or SMS notification code here # Example call daily_self_check('A12345', 'path_to_image_b.jpg') Embodiment 2 The second aspect, as Figure 2 As shown, the present application provides a self-checking method for the accuracy of a license plate camera based on a multimodal model, which is applied to the above-mentioned self-checking system for the accuracy of a license plate camera based on a multimodal model, and specifically includes: S1 obtains recognition deviation data of the license plate camera within a preset time period, and uses the recognition deviation data to determine whether the recognition processing accuracy of the license plate camera meets the requirements, and then proceeds to the next step; Furthermore, the recognition deviation data of the license plate camera includes the number of recognition deviations of the license plate camera and environmental data corresponding to different numbers of recognition deviations.

[0035] Specifically, the environmental data includes light intensity.

[0036] Specifically, Figure 3 As shown, determining that the recognition processing accuracy of the license plate camera meets the requirements specifically includes: Determine the number of recognition deviations of the license plate camera using the recognition deviation data of the license plate camera, and determine the number of recognition deviations within different illumination intervals using the illumination corresponding to different recognition deviations. Based on the number of identification deviations in different light intensity intervals, the number of identification deviations in a preset light intensity interval is performed; Whether the recognition processing accuracy of the license plate camera meets the requirements is determined by the number of recognition deviations within a preset light intensity range.

[0037] Optionally, the preset light intensity interval is an interval in which the light intensity is less than a preset light intensity threshold.

[0038] It can be understood that determining whether the recognition processing accuracy of the license plate camera meets the requirements by the number of recognition deviations within the preset illumination interval specifically includes: When the number of recognition deviations within the preset illumination amount interval is greater than the preset number of deviations, it is determined that the recognition processing accuracy of the license plate camera does not meet the requirement.

[0039] Specifically, when the recognition processing accuracy of the license plate camera does not meet the requirement, the set number is used to determine the number of license plates extracted by the license plate camera on the current date.

[0040] In another possible embodiment, determining whether the recognition processing accuracy of the license plate camera meets the requirement specifically includes: Determine the number of recognition deviations of the license plate camera using the recognition deviation data of the license plate camera; Using the number of recognition deviations of the license plate camera on different dates, determining the recognition deviation date in the date; Based on the number of the recognition deviation dates, it is determined whether the recognition processing accuracy of the license plate camera meets the requirements.

[0041] It should be noted that when the proportion of the number of recognition deviation dates within the preset time period is greater than the proportion of the number of preset dates, it is determined that the recognition processing accuracy of the license plate camera does not meet the requirements.

[0042] In another possible embodiment, determining whether the recognition processing accuracy of the license plate camera meets the requirement specifically includes: Determine the number of recognition deviations of the license plate camera using the recognition deviation data of the license plate camera, and when the number of recognition deviations of the license plate camera does not meet the requirement, determine that the recognition processing accuracy of the license plate camera does not meet the requirement; When the number of recognition deviations of the license plate camera meets the requirements: Obtaining the number of recognition deviations of the license plate camera, using the illumination of the license plate camera on different dates and the illumination corresponding to different recognition deviations, determining the number of recognition deviations in different illumination intervals, and when the number of recognition deviations in a preset illumination interval does not meet the requirement, determining that the recognition processing accuracy of the license plate camera does not meet the requirement; When the number of identification deviations within the preset light intensity range meets the requirements: Using the recognition data of the license plate camera on different dates, determining the number of recognition deviations of the license plate camera on different dates, and when determining that the license plate camera has a recognition deviation date using the number of recognition deviations of the license plate camera on different dates, determining that the recognition processing accuracy of the license plate camera does not meet the requirements; When the license plate camera does not have a recognition deviation date: Obtaining the number of recognition deviations on different dates, and combining the light amounts corresponding to the different recognition deviation numbers, determining that the recognition processing deviation rate of the license plate camera does not meet the requirements when the number of recognition deviations within a preset light amount interval does not meet the requirements on the date; When there is no date on which the number of identified deviations within the preset light intensity range does not meet the requirements: Determine the recognition deviation coefficients of different dates based on the number of recognition deviations on different dates and the amounts of light corresponding to the different number of recognition deviations. When there is a date where the recognition deviation coefficient does not meet the requirement, it is determined that the recognition processing deviation rate of the license plate camera does not meet the requirement. When there is no date for which the identification coefficient of variation does not meet the requirements: The recognition deviation amount of the license plate camera is determined based on the recognition deviation coefficients of different dates within a preset time period, and the recognition deviation amount is used to determine whether the recognition processing accuracy of the license plate camera meets the requirements.

[0043] Further, the recognition deviation is used to determine whether the recognition processing accuracy of the license plate camera meets the requirements, specifically including: When the recognition deviation of the license plate camera is within a preset deviation range, it is determined that the recognition processing accuracy of the license plate camera does not meet the requirement.

[0044] S2 determines the recognition deviation data of the license plate camera under similar environmental data in the time period based on the environmental data of the license plate camera in different time periods, and determines the recognition deviation probability type of the time period by using the recognition deviation data under similar environmental data in the time period; Specifically, the similar environmental data is environmental data whose deviation from the light intensity in the time period is within a preset light intensity deviation range.

[0045] It is understandable that if Figure 4 As shown, the method for determining the identification deviation probability type of the time period is: Determine the number of identification deviations under the similar environmental data using the identification deviation data under the similar environmental data in the time period; Determining a recognition deviation ratio based on a ratio of the number of recognition deviations under the similar environment data to the number of recognition processing under the similar environment data; The recognition deviation number and the recognition deviation ratio are used to determine the recognition deviation probability of the time period, and the recognition deviation probability type of the time period is determined according to the recognition deviation probability.

[0046] Furthermore, the recognition deviation probability of the time period is determined according to the product of a preset weight coefficient corresponding to the number of recognition deviations and the recognition deviation ratio.

[0047] It should be noted that determining the identification deviation probability type of the time period according to the identification deviation probability specifically includes: When the recognition deviation probability of the time period is within the preset deviation probability interval, determining the recognition deviation probability type of the time period as a first-class deviation probability type; When the recognition deviation probability of the time period is not within the preset deviation probability interval, it is determined that the recognition deviation probability type of the time period is a second-class deviation probability type.

[0048] S3 determines the number of license plates extracted by the license plate camera on the current date based on the recognition data of license plates in time periods of different recognition deviation probability types on the current date; It is understandable that if Figure 5 As shown, the method for determining the number of license plates extracted by the license plate camera on the current date is: Based on the license plate recognition data of the license plate camera in different recognition deviation probability types in the current date, determining the recognition quantity ratio of the license plates corresponding to the different recognition deviation probability types on the current date; Determine the bias weight coefficients of different identification bias probability types by using the preset bias weight coefficients corresponding to different identification bias probability types; The extraction requirement coefficient of the license plate camera is determined according to the sum of the product of the proportion of the number of license plates corresponding to different recognition deviation probability types on the current date and the deviation weight coefficient, and the extraction requirement coefficient is used to determine the number of license plates extracted by the license plate camera on the current date.

[0049] Furthermore, the extraction requirement coefficient is used to determine the number of license plates to be extracted by the license plate camera on the current date, specifically including: The product of the extraction requirement coefficient and the number of license plates recognized by the license plate camera on the current date is used as the number of license plates extracted by the license plate camera on the current date.

[0050] In another embodiment, the method for determining the number of license plates extracted by the license plate camera on the current date is: Determining the number of license plates recognized by the license plate camera on the current date based on recognition data of the license plate camera on the current date; Based on the license plate recognition data of the license plate camera in different recognition deviation probability types, determine the recognition quantity ratio of the license plates corresponding to different recognition deviation probability types on the current date, and determine the recognition quantity ratio of the license plates corresponding to a type of deviation probability type on the current date; The number of license plates extracted by the license plate camera on the current date is determined by multiplying the number of license plates recognized by the license plate camera on the current date by the proportion of the number of license plates corresponding to a type of deviation probability on the current date.

[0051] In another possible embodiment, the method for determining the number of license plates extracted by the license plate camera on the current date is: Based on the recognition data of the license plate camera on the current date, determining the number of license plates recognized by the license plate camera on the current date, when the number of license plates recognized by the license plate camera on the current date is greater than a preset number of license plate recognitions, using the set number to determine the number of license plates extracted by the license plate camera on the current date; When the number of license plates recognized by the license plate camera on the current date is not greater than the preset number of license plates recognized: When the number of license plates recognized by the license plate camera on the current date is less than a preset number threshold, the number of license plates recognized by the license plate camera on the current date is used to determine the number of license plates extracted by the license plate camera on the current date; When the number of license plates recognized by the license plate camera on the current date is not less than a preset number threshold: Based on the recognition data of the license plate in different recognition deviation probability types of time periods of the license plate camera on the current date, the number of recognitions of the license plate corresponding to a type of deviation probability type on the current date is determined; when the number of recognitions of the license plate corresponding to a type of deviation probability type on the current date does not meet the requirement, the number of license plates extracted by the license plate camera on the current date is determined by using a set number; When the number of license plates corresponding to a type of deviation probability meets the requirements on the current date: Determine the proportion of the number of license plates corresponding to different recognition deviation probability types on the current date, and determine the corresponding recognition weight value of a type of deviation type in combination with the number of license plates corresponding to a type of deviation probability type on the current date; when the recognition weight value is greater than a preset weight threshold, use the set number to determine the number of license plates extracted by the license plate camera on the current date; When the recognition weight value is not greater than the preset weight threshold: Determine the bias weight coefficients of different identification bias probability types by using the preset bias weight coefficients corresponding to different identification bias probability types; The extraction requirement coefficient of the license plate camera is determined according to the sum of the product of the proportion of the number of license plates corresponding to different recognition deviation probability types on the current date and the deviation weight coefficient, and the extraction requirement coefficient is used to determine the number of license plates extracted by the license plate camera on the current date.

[0052] S4 takes the extraction quantity as a constraint condition, and uses the recognition deviation data under similar environmental data in the time period and the recognition data of the license plate to determine the extraction target time period and the number of license plates to be extracted in the time period, and determines whether it is necessary to dynamically adjust the extraction strategy of the current date based on the comparison result of the license plate in the extraction target time period.

[0053] Furthermore, the method for determining the extraction target time period in the time period is: Based on the recognition deviation data under the similar environment data in the time period, determining the number of recognition deviations under the similar environment data in the time period, and using it as the number of matching deviations; Determining the number of license plates recognized in the time period using the recognition data of the license plates in the time period; The number of matching deviations and the number of license plate recognitions are used to determine whether the time period is an extraction target time period.

[0054] Specifically, using the number of matching deviations and the number of license plate recognitions, determining whether the time period is a target time period for extraction specifically includes: When any one of the number of matching deviations and the number of license plate recognitions in the time period is not within a preset range, the time period is determined to be an extraction target time period.

[0055] It should be noted that the method for determining the number of license plates to be drawn during the target drawing period is as follows: Determining the sum of weight coefficients for different extraction target time periods based on the sum of preset weight coefficients corresponding to the number of matching deviations in the extraction target time period and the preset weight coefficients corresponding to the number of recognized license plates; After normalizing the weight coefficients and the weight coefficients of different extraction target time periods, extraction proportional factors of different extraction target time periods are obtained; The number of license plates extracted during the extraction target period is determined by multiplying the extraction ratio factor by the extraction number.

[0056] Optionally, the method for determining the target time period in the time period is: S41 uses the recognition data of the license plates in the time period to determine the number of recognitions of the license plates in the time period, and determines the recognition processing busyness value of the time period in combination with the proportion of the number of recognitions of the license plates in the time period to the number of the current date; Optionally, the above step S41 includes the following contents: S411 uses the recognition data of the license plates in the time period to determine the number of license plates recognized in the time period. When the number of license plates recognized in the time period is greater than the preset number of license plates, the time period is determined to be the extraction target time period. When the number of license plates recognized in the time period is not greater than the preset number of license plates, the process proceeds to step S412. S412 obtains the proportion of the number of license plates recognized in the time period on the current date. When the proportion of the number of license plates recognized in the time period on the current date is greater than the preset proportion of recognition, proceed to step S414. When the proportion of the number of license plates recognized in the time period on the current date is not greater than the preset proportion of recognition, proceed to step S413. S413: when the number of license plate recognitions in the time period is within the preset recognition number interval, it is determined that the time period does not belong to the extraction target time period; when the number of license plate recognitions in the time period is not within the preset recognition number interval, the process proceeds to step S414; S414 determines the recognition processing busy value of the time period based on the number of license plates recognized in the time period and the proportion of the number of license plates recognized in the time period to the number on the current date. When the recognition processing busy value of the time period is greater than the preset busy threshold, the time period is determined to be the extraction target time period. When the recognition processing busy value of the time period is not greater than the preset busy threshold, proceed to step S42.

[0057] S42 determines the number of recognition deviations under the similar environment data in the period based on the recognition deviation data under the similar environment data in the period, and uses it as the number of matching deviations, and determines the recognition deviation value of the period based on the number of matching deviations in the historical periods corresponding to different similar environment data and the number of recognitions of the license plates; Optionally, the above step S42 includes the following contents: S421 determines the number of identification deviations under similar environmental data in the time period based on the identification deviation data under similar environmental data in the time period, and uses it as the number of matching deviations. When the number of matching deviations is greater than a preset number of deviations threshold, it is determined that the time period belongs to the extraction target time period. When the number of matching deviations is not greater than the preset number of deviations threshold, it proceeds to step S422. S422: if it is determined that there is a historical period in which the number of matching deviations does not meet the requirement based on the number of matching deviations in the historical periods corresponding to different similar environment data, then the process proceeds to step S423; if there is no historical period in which the number of matching deviations does not meet the requirement, then the process proceeds to step S424; S423: When the number of historical time periods whose matching deviation times do not meet the requirements is greater than the preset time period number threshold, it is determined that the time period belongs to the extraction target time period; when the number of historical time periods whose matching deviation times do not meet the requirements is not greater than the preset time period number threshold, the process proceeds to step S424; S424 determines the recognition deviation value of the time period based on the number of matching deviations and the number of license plate recognitions in the historical time periods corresponding to different similar environmental data. When the recognition deviation value of the time period is greater than the preset deviation threshold, it is determined that the time period belongs to the extraction target time period. When the recognition deviation value of the time period is not greater than the preset deviation threshold, it proceeds to step S43.

[0058] S43 determines a target matching value of the time period by using an average value of the recognition processing busy value and the recognition deviation value of the time period, and determines whether the time period is an extraction target time period according to the target matching value.

[0059] It should be noted that determining whether the current date extraction strategy needs to be dynamically adjusted specifically includes: Based on the comparison results of the license plates in the target extraction period, determining the number of recognition deviations of the license plates in different target extraction periods; Determining recognition deviation ratios for different extraction target periods according to ratios of the number of recognition deviations of license plates in different extraction target periods to the number of extracted license plates; The identification deviation ratio of different extraction target periods is used to determine whether dynamic adjustment of the extraction strategy for the current date is required.

[0060] Specifically, the identification deviation ratio of different extraction target periods is used to determine whether the extraction strategy of the current date needs to be dynamically adjusted, including: When the average value of the recognition deviation ratios of different extraction target time periods is greater than the preset deviation ratio threshold, it is determined that a dynamic adjustment of the extraction strategy for the current date is required.

[0061] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0062] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0063] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.

Claims

1. A license plate camera accuracy self-checking system based on a multimodal model, characterized in that: Specifically include: License plate sampling module, result verification module, verification result notification module; The license plate sampling module is responsible for extracting a preset number of license plate numbers and their corresponding recognition images from the recognition results of the license plate recognition camera every day; The result verification module is responsible for submitting the recognition images to multiple third-party large models for recognition to obtain recognition results, and using the recognition results to compare with the license plate number to determine whether a reminder is needed; The verification result notification module is responsible for automatically notifying relevant personnel to conduct license plate camera inspection when a reminder is required.

2. The license plate camera accuracy self-checking system based on a multimodal model as claimed in claim 1, characterized in that: The third-party model calls the third-party large model through an API interface to perform license plate recognition.

3. The license plate camera accuracy self-checking system based on a multimodal model as claimed in claim 1, characterized in that: The comparison result between the recognition result and the license plate number is used to determine whether a reminder is needed, specifically including: When it is determined that there is no recognition result inconsistent with the license plate number according to the comparison between the license plate number and the recognition results of different third-party models, it is determined that no reminder is needed; When there is a recognition result inconsistent with the license plate number, the third-party model is divided into multiple model groups according to whether the recognition result is consistent. The number of third-party models consistent with the license plate number is increased by 1 as the target number. When there is a model group in which the number of third-party models is greater than the target number, it is determined that a reminder is required.

4. The license plate camera accuracy self-checking system based on a multimodal model as claimed in claim 1, characterized in that: The verification result notification module is triggered by email, text message or system notification.

5. A method for self-checking the accuracy of a license plate camera based on a multimodal model, applied to a system for self-checking the accuracy of a license plate camera based on a multimodal model as claimed in any one of claims 1 to 3, characterized in that: Specifically include: Acquire recognition deviation data of the license plate camera within a preset time period, and use the recognition deviation data to determine that the recognition processing accuracy of the license plate camera meets the requirements, and then proceed to the next step; Based on the environmental data of the license plate camera in different time periods, determining the recognition deviation data of the license plate camera under similar environmental data in the time period, and using the recognition deviation data under similar environmental data in the time period to determine the recognition deviation probability type of the time period; Determine the number of license plates extracted by the license plate camera on the current date based on the recognition data of license plates in time periods of different recognition deviation probability types on the current date; Taking the extraction quantity as a constraint condition, the target extraction period and the number of license plates to be extracted in the period are determined by utilizing the recognition deviation data under similar environmental data in the period and the recognition data of the license plate, and based on the comparison result of the license plate in the target extraction period, it is determined whether it is necessary to dynamically adjust the extraction strategy for the current date.

6. The method for self-checking the accuracy of a license plate camera based on a multimodal model as claimed in claim 5, characterized in that: The recognition deviation data of the license plate camera includes the recognition deviation times of the license plate camera and environmental data corresponding to different recognition deviation times.

7. The method for self-checking the accuracy of a license plate camera based on a multimodal model as claimed in claim 5, characterized in that: The environmental data includes light intensity.

8. The method for self-checking the accuracy of a license plate camera based on a multimodal model as claimed in claim 5, characterized in that: Determining that the recognition processing accuracy of the license plate camera meets the requirements, specifically includes: Determine the number of recognition deviations of the license plate camera using the recognition deviation data of the license plate camera, and determine the number of recognition deviations within different illumination intervals using the illumination corresponding to different recognition deviations. Based on the number of identification deviations in different light intensity intervals, the number of identification deviations in a preset light intensity interval is performed; Whether the recognition processing accuracy of the license plate camera meets the requirements is determined by the number of recognition deviations within a preset light intensity range.

9. The method for self-checking the accuracy of a license plate camera based on a multimodal model as claimed in claim 8, characterized in that: Determining whether the recognition processing accuracy of the license plate camera meets the requirements by the number of recognition deviations within a preset illumination range specifically includes: When the number of recognition deviations within the preset illumination amount interval is greater than the preset number of deviations, it is determined that the recognition processing accuracy of the license plate camera does not meet the requirement.

10. The method for self-checking the accuracy of a license plate camera based on a multimodal model as claimed in claim 5, characterized in that: Determine whether it is necessary to dynamically adjust the extraction strategy of the current date, including: Based on the comparison results of the license plates in the target extraction period, determining the number of recognition deviations of the license plates in different target extraction periods; Determining recognition deviation ratios for different extraction target periods according to ratios of the number of recognition deviations of license plates in different extraction target periods to the number of extracted license plates; The identification deviation ratio of different extraction target periods is used to determine whether dynamic adjustment of the extraction strategy for the current date is required.

Images