AI-based vehicle feature detection method

Abstract

The present invention discloses a vehicle feature detection method based on AI, which relates to the technical field of vehicle feature detection. The method installs a network camera to perform comprehensive image acquisition of vehicles entering and leaving the factory, and transmits these image data to a central processing center through a network for preprocessing, which can significantly improve the acquisition efficiency and accuracy of image data. The network camera can obtain the front image of the vehicle, while the depth camera can accurately capture the image of the driver's window area. During the preprocessing process, the image data undergoes steps such as denoising, size normalization, and image set packaging to generate a standard entry picture set and a standard exit picture set, and store them in a MongoDB database. Through this automated data acquisition and management method, the system reduces time delays and human errors in traditional manual operations, and significantly improves the efficiency and reliability of the logistics weighing system.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle feature detection, and in particular to a vehicle feature detection method based on AI. Background Art

[0002] With the rapid development of artificial intelligence (AI) technology, AI is increasingly used in the field of industrial automation, especially in intelligent monitoring and logistics weighing systems. AI-based vehicle feature detection methods belong to the category of computer vision and image recognition, combining deep learning technology and image processing algorithms to identify and analyze multi-dimensional features of vehicles in real time. In this application, a network camera is used to capture the front image of the vehicle and relevant information about the driver, and the convolutional neural network (CNN) is used to extract key features such as the vehicle's color, license plate, logo, and headlight brightness. These feature data can be used for automated management of vehicles entering and leaving the factory to ensure the consistency and safety of vehicle identity.

[0003] At present, due to the limitations of traditional methods, logistics weighing systems often have difficulty coping with complex environments. For example, when vehicles enter and exit a factory under strong light, at night, or under changeable weather conditions, traditional image matching methods are prone to recognition errors, resulting in the vehicle's features not being accurately matched with records in the database. This situation may cause a series of problems, such as unauthorized vehicles being mistakenly identified as legal vehicles and released, or legal vehicles being refused to leave the factory due to misjudgment, which seriously affects the factory's operational efficiency and safety. These shortcomings make traditional vehicle management methods incapable of coping with large-scale vehicle in and out, and prone to misjudgment and security loopholes. Summary of the invention

[0004] In view of the deficiencies of the prior art, the present invention provides an AI-based vehicle feature detection method to solve the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A vehicle feature detection method based on AI comprises the following steps:

[0006] S1. Use a network camera to capture front images of vehicles entering the factory and vehicles leaving the factory, and use a depth camera to collect driver window images of vehicles entering the factory and vehicles leaving the factory, and send the collected front images and driver window images to a central processing center through a network for preprocessing, obtain a standard factory entry image set and a standard factory exit image set, and store the preprocessed standard factory entry image set and standard factory exit image set in a database;

[0007] S2, extracting the standard factory entry picture set and the standard factory exit picture set stored in the database, applying the convolutional neural network CNN to perform feature extraction on the standard factory entry picture set and the standard factory exit picture set, and generating a vehicle factory entry feature set and a vehicle factory exit feature set;

[0008] S3, constructing a vehicle feature similarity scoring formula and a vehicle optical feature formula, then inputting the extracted vehicle entry feature set and vehicle exit feature set into the vehicle feature similarity scoring formula and the vehicle optical feature formula, and outputting a vehicle similarity score VFSS and a vehicle optical feature consistency index VOFCI;

[0009] S4, based on the obtained vehicle similarity score VFSS and vehicle optical feature consistency index VOFCI, a comprehensive calculation is performed to output a comprehensive similarity index CSEF, and then a preliminary comparative evaluation is performed between the preset similarity threshold T and the comprehensive similarity index CSEF;

[0010] S5. Based on the preliminary comparative evaluation results, a dynamic threshold adjustment formula is further constructed, the similarity threshold T is input into the dynamic threshold adjustment formula, the dynamic adjustment threshold Tnew is output, and then a secondary comparative evaluation is performed with the comprehensive similarity index CSEF obtained in real time.

[0011] Preferably, S1 includes S11, S12 and S13;

[0012] S11. Install network cameras on the front of the entrances and exits of the logistics vehicle access control and weighing areas. The network cameras automatically trigger the shooting of vehicles entering and leaving the factory to obtain the front image of the vehicle. The front image includes vehicle color information, surface reflectivity, vehicle logo details and vehicle light status content information;

[0013] At the same time, at the side of the entrance and exit of the logistics vehicle access control and weighing area, based on the installed network cameras, the driver's window area of ​​the vehicle entering the factory and the vehicle leaving the factory is captured. The driver's window picture includes the driver's face, window transparency and the number of passengers in the car;

[0014] S12, connecting the network camera to the central processing center by setting up a connection network, transmitting the collected front image and the driver's window image to the central processing center for preprocessing, and obtaining a standard factory entry image set and a standard factory exit image set, wherein the preprocessing method includes denoising, size normalization, and image set packaging;

[0015] S13. Build a MongoDB database, and set a write port and a write output port respectively. Write the preprocessed standard incoming picture set and standard outgoing picture set into the MongoDB database through the write port. After receiving the standard picture data set, the MongoDB database automatically classifies and indexes the standard incoming picture set and the standard outgoing picture set, and stores the standard incoming picture set and the standard outgoing picture set into the picture storage table by setting the picture storage table and the feature data storage table.

[0016] Preferably, S2 includes S21, S22 and S23;

[0017] S21. Extract the standard factory entry picture set and the standard factory exit picture set in the picture storage table in real time through the write port set in the MongoDB database, and use the convolutional neural network CNN to perform feature extraction on the extracted standard factory entry picture set and the standard factory exit picture set to obtain the vehicle factory entry feature set and the vehicle exit feature set;

[0018] The vehicle entry feature set includes the vehicle entry appearance feature set and the vehicle entry optical feature set;

[0019] The appearance feature set of the incoming vehicle includes the color feature parameter VCP of the incoming vehicle 1 , the recognition accuracy of vehicle license plate characters entering the factory LPRA 1 The matching degree of facial features of the driver of the vehicle entering the factory is DFFMD 1 ;

[0020] The optical characteristic set of the incoming vehicle includes the HBI index of the incoming vehicle 1 And the LDC of the vehicle logo details 1 ;

[0021] The vehicle factory feature set includes a factory vehicle appearance feature set and a factory vehicle optical feature set;

[0022] The factory vehicle appearance feature set includes the factory vehicle color feature parameters VCP 2 , Factory vehicle license plate character recognition accuracy LPRA 2 The matching degree of the driver's facial features with the factory vehicle DFFMD 2 ;

[0023] The factory vehicle optical characteristic set includes the factory vehicle headlight brightness index HBI 2 And the LDC of the vehicle logo details 2 ;

[0024] S22, extract edge, color and texture features from the standard factory entry picture set and the standard factory exit picture set through the convolution operation of the convolutional neural network CNN, and obtain the factory entry vehicle color feature parameter VCP 1 and factory vehicle color characteristic parameters VCP 2 ;

[0025] Through the target detection operation of the convolutional neural network CNN, the license plate identification characters, headlight brightness and vehicle logo details of the vehicles in the standard factory entry picture set and the standard factory exit picture set are detected, and the license plate character recognition accuracy rate LPRA of the vehicles entering the factory is obtained. 1 , Factory vehicle license plate character recognition accuracy LPRA 2 , Vehicle headlight brightness index HBI1 , Factory vehicle headlight brightness index HBI 2 , LDC of vehicle logo details complexity 1 And the LDC of the vehicle logo details 2 ;

[0026] The deep convolution layer of the convolutional neural network CNN is used to extract the standard factory entry picture set and the standard factory exit picture set, and the driver's facial features are used to obtain the matching degree DFFMD of the driver's facial features of the factory entry vehicle. 1 The matching degree of the driver's facial features with the factory vehicle DFFMD 2 ;

[0027] S23. The vehicle entry feature set and the vehicle exit feature set are written into the feature data storage table of the MongoDB database through the write port of the distributed database for storage, and the timestamp and license plate characters are marked for each written vehicle entry feature set and vehicle exit feature set.

[0028] Preferably, S3 includes S31 and S32;

[0029] S31, constructing a vehicle feature similarity scoring formula, inputting the appearance feature set of incoming vehicles and the appearance feature set of outgoing vehicles into the vehicle feature similarity scoring formula, and outputting a vehicle similarity score VFSS;

[0030] The vehicle similarity score VFSS is calculated using the following vehicle feature similarity score formula;

[0031]

[0032] in, Indicates color similarity, represents the similarity of license plate character recognition, It represents the matching degree of the facial features of the vehicle driver. α, β and χ represent the preset weight values ​​of color similarity, license plate character recognition similarity and matching degree of the facial features of the vehicle driver respectively. The specific values ​​are set by the user. VCPmax represents the reference upper limit of the vehicle color feature parameter. LPRAmax represents the reference upper limit of the vehicle license plate character recognition accuracy. DFFMDmax represents the reference upper limit of the matching degree of the facial features of the vehicle driver.

[0033] Preferably, S32 includes S321 and S322;

[0034] S321, the vehicle headlight brightness index HBI 1 and the vehicle headlight brightness index HBI 2 And the LDC of the vehicle logo details 1 And the LDC of the vehicle logo details2 Calculate the difference respectively and output the headlight brightness difference HBI diff Difference between details and car logo LDC diff ;

[0035] Headlight brightness difference HBI diff Difference between details and car logo LDC diff Obtained through the following algorithm formula;

[0036]

[0037] Preferably, S322, by constructing a vehicle optical characteristic formula, the headlight brightness difference HBI diff Difference between details and car logo LDC diff , input into the vehicle optical characteristic formula, calculate and output the vehicle optical characteristic consistency index VOFCI;

[0038] The vehicle optical characteristic consistency index VOFCI is calculated and obtained by the following vehicle optical characteristic formula;

[0039]

[0040] Among them, e represents the exponential function, δ represents the weight coefficient of the headlight brightness, and ε represents the weight coefficient of the complexity of the car logo details.

[0041] Preferably, S4 includes S41 and S42;

[0042] S41, calculating and outputting a comprehensive similarity index CSEF by combining the obtained vehicle similarity score VFSS and the vehicle optical feature consistency index VOFCI;

[0043] The comprehensive similarity index CSEF is calculated by the following algorithm formula;

[0044]

[0045] Preferably, S42, the user preliminarily sets a similarity threshold T, and then performs a preliminary comparative evaluation with the obtained comprehensive similarity index CSEF, and generates a response mechanism according to the evaluation result. The specific evaluation content is as follows;

[0046] When the comprehensive similarity index CSEF ≥ similarity threshold T, it means the similarity is qualified and the vehicle is automatically released.

[0047] When the comprehensive similarity index CSEF is less than the similarity threshold T, it means that the similarity is unqualified and passage is prohibited. At this time, a response mechanism is generated to dynamically adjust the similarity threshold T.

[0048] Preferably, S5 includes S51 and S52;

[0049] S51. When the similarity is found to be unqualified in the preliminary comparison and evaluation, a dynamic adjustment mechanism of the similarity threshold T is triggered, and a dynamic adjustment threshold Tnew is calculated and outputted based on the historical comprehensive similarity index CSEF by constructing a dynamic threshold adjustment formula;

[0050] The dynamic adjustment threshold Tnew is calculated by the following dynamic threshold adjustment formula:

[0051]

[0052] Among them, ζ represents the adjustment coefficient, which is set by the user according to the actual situation, Average(CSEF) represents the historical comprehensive similarity index CSEF, and Max(CSEF) represents the upper limit value of the comprehensive similarity index CSEF.

[0053] Preferably, S52, performing a secondary comparative evaluation based on the obtained dynamic adjustment threshold Tnew and the comprehensive similarity index CSEF output in real time, and the specific evaluation content is as follows;

[0054] When the comprehensive similarity index CSEF ≥ the dynamic adjustment threshold Tnew, it means that the similarity is qualified and the vehicle is automatically released;

[0055] When the comprehensive similarity index CSEF is less than the dynamically adjusted threshold Tnew, it means that the similarity is unqualified. At this time, passage is prohibited and an early warning message is generated to prompt the staff to conduct manual verification.

[0056] The present invention provides an AI-based vehicle feature detection method, which has the following beneficial effects:

[0057] (1) This method can significantly improve the efficiency and accuracy of image data acquisition by installing network cameras at the exit and entrance of the factory to collect comprehensive images of vehicles entering and leaving the factory, and transmit these image data to the central processing center through the network for preprocessing. The network camera can obtain the front image of the vehicle, including information such as vehicle color, surface reflectivity, vehicle logo details and headlight status, while the depth camera can accurately capture the image of the driver's window area and record the driver's face, window transparency and the number of passengers in the car. During the preprocessing process, the image data undergoes denoising, size normalization and image set packaging steps to generate standard entry image sets and standard exit image sets, which are stored in the MongoDB database. Through this automated data collection and management method, the system reduces time delays and human errors in traditional manual operations, and significantly improves the efficiency and reliability of the logistics weighing system.

[0058] (2) This method uses the MongoDB database to store standard factory entry and factory exit pictures. The system uses the convolutional neural network (CNN) to perform deep feature extraction on the standard factory entry and factory exit pictures to generate detailed vehicle entry and exit feature sets. In the feature extraction process, CNN extracts edge, color and texture features in the image through convolution operations, thereby accurately obtaining key parameters such as the vehicle's color feature parameters VCP, license plate character recognition accuracy LPR), driver facial feature matching DFFMD, headlight brightness index HBI and vehicle logo detail complexity LDC. Compared with traditional methods, the application of CNN greatly improves the accuracy and adaptability of feature extraction, especially in the case of lighting changes, image noise and complex backgrounds, it can still maintain a high level of recognition accuracy. This feature extraction process lays a solid foundation for subsequent similarity analysis, reduces the occurrence of misjudgments, and enhances the system's ability to identify abnormal situations.

[0059] (3) By constructing a vehicle feature similarity scoring formula and a vehicle optical feature formula, this method can enable the system to conduct a comprehensive similarity assessment of the vehicle's entry and exit feature sets. Specifically, the system calculates the vehicle similarity score VFSS and the vehicle optical feature consistency index VOFCI, and further obtains the comprehensive similarity index CSEF through comprehensive calculation. Then, the comprehensive similarity index is preliminarily compared and evaluated with the preset similarity threshold T to determine whether the vehicle meets the factory conditions. When the preliminary evaluation results show that the similarity is unqualified, the system will trigger the dynamic threshold adjustment mechanism, calculate and output the dynamic adjustment threshold Tnew based on historical data, and conduct a secondary comparison and evaluation. This dynamic adjustment mechanism allows the system to flexibly respond to different environmental conditions and changes in vehicle features, thereby significantly improving the consistency and accuracy of vehicle identification. At the same time, the system automatically releases qualified vehicles and generates warning information when unqualified vehicles are detected, reminding staff to conduct manual verification to ensure the safety and efficiency of the vehicle management process. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 Schematic diagram of the steps of the vehicle feature detection method based on AI of the present invention. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0062] Example 1

[0063] See also Figure 1 The present invention provides a vehicle feature detection method based on AI. To achieve the above purpose, the present invention is implemented by the following technical scheme, including the following steps:

[0064] S1. Use a network camera to capture front images of vehicles entering the factory and vehicles leaving the factory, and use a depth camera to collect driver window images of vehicles entering the factory and vehicles leaving the factory, and send the collected front images and driver window images to a central processing center through a network for preprocessing, obtain a standard factory entry image set and a standard factory exit image set, and store the preprocessed standard factory entry image set and standard factory exit image set in a database;

[0065] S2, extracting the standard factory entry picture set and the standard factory exit picture set stored in the database, applying the convolutional neural network CNN to perform feature extraction on the standard factory entry picture set and the standard factory exit picture set, and generating a vehicle factory entry feature set and a vehicle factory exit feature set;

[0066] S3, constructing a vehicle feature similarity scoring formula and a vehicle optical feature formula, then inputting the extracted vehicle entry feature set and vehicle exit feature set into the vehicle feature similarity scoring formula and the vehicle optical feature formula, and outputting a vehicle similarity score VFSS and a vehicle optical feature consistency index VOFCI;

[0067] S4, based on the obtained vehicle similarity score VFSS and vehicle optical feature consistency index VOFCI, a comprehensive calculation is performed to output a comprehensive similarity index CSEF, and then a preliminary comparative evaluation is performed between the preset similarity threshold T and the comprehensive similarity index CSEF;

[0068] S5. Based on the preliminary comparative evaluation results, a dynamic threshold adjustment formula is further constructed, the similarity threshold T is input into the dynamic threshold adjustment formula, the dynamic adjustment threshold Tnew is output, and then a secondary comparative evaluation is performed with the comprehensive similarity index CSEF obtained in real time.

[0069] In this embodiment, the method uses a network camera to capture the front image of the vehicle entering and leaving the factory, and collects pictures of the driver's window area through a depth camera to fully capture the appearance characteristics and driver information of the vehicle. Subsequently, these images are transmitted to the central processing center through the network, and after pre-processing such as denoising and size normalization, standardized factory entry and factory exit picture sets are generated and stored in the MongoDB database for structured management. This process ensures the accuracy of data collection and the efficiency of transmission, laying the foundation for subsequent feature analysis. Next, the system extracts a standard picture data set from the database and applies a convolutional neural network CNN to perform in-depth feature extraction. Through the convolution operation and deep learning algorithm of the convolutional neural network CNN, the system can accurately extract the multi-dimensional features of the vehicle. These features are organized into a vehicle entry feature set and a factory exit feature set, providing detailed data support for vehicle feature consistency analysis. This feature extraction process ensures that the system can maintain a high level of recognition accuracy under different environments, significantly improving the accuracy and reliability of vehicle recognition. On this basis, the system constructs a vehicle feature similarity scoring formula and a vehicle optical feature formula, inputs the extracted entry and exit feature sets into them, and calculates the vehicle similarity score VFSS and the vehicle optical feature consistency index VOFCI. These indicators comprehensively reflect the consistency of the vehicle's features when entering and leaving the factory. Subsequently, the system calculates the comprehensive similarity index CSEF based on these results, and makes a preliminary evaluation with the preset similarity threshold T to determine the eligibility of the vehicle. Through this step, the system can effectively identify vehicles that have undergone significant changes during the entry and exit process, thereby improving the accuracy and safety of management. Finally, in order to adapt to possible environmental changes and fluctuations in vehicle features, the system introduces a dynamic threshold adjustment mechanism. When the preliminary evaluation results show that the similarity is unqualified, the system will adjust the similarity threshold T in real time through the dynamic threshold adjustment formula constructed by historical data, generate a dynamic adjustment threshold Tnew, and conduct a secondary evaluation with the comprehensive similarity index CSEF obtained in real time. This dynamic adjustment mechanism makes the system more adaptable and flexible, can effectively respond to different actual situations, reduce the occurrence of misjudgments, and improve the robustness of the system.

[0070] Example 2

[0071] This embodiment is explained in Example 1, please refer to Figure 1 , specifically: S1 includes S11, S12 and S13;

[0072] S11. Install network cameras on the front of the entrances and exits of the logistics vehicle access control and weighing areas. The network cameras automatically trigger the shooting of vehicles entering and leaving the factory to obtain the front image of the vehicle. The front image includes vehicle color information, surface reflectivity, vehicle logo details and vehicle light status content information;

[0073] At the same time, at the side of the entrance and exit of the logistics vehicle access control and weighing area, based on the installed network cameras, the driver's window area of ​​the vehicle entering the factory and the vehicle leaving the factory is captured. The driver's window picture includes the driver's face, window transparency and the number of passengers in the car;

[0074] S12, connecting the network camera to the central processing center by setting up a connection network, transmitting the collected front image and the driver's window image to the central processing center for preprocessing, and obtaining a standard factory entry image set and a standard factory exit image set, wherein the preprocessing method includes denoising, size normalization, and image set packaging;

[0075] S13. Build a MongoDB database, and set a write port and a write output port respectively. Write the preprocessed standard incoming picture set and standard outgoing picture set into the MongoDB database through the write port. After receiving the standard picture data set, the MongoDB database automatically classifies and indexes the standard incoming picture set and the standard outgoing picture set, and stores the standard incoming picture set and the standard outgoing picture set into the picture storage table by setting the picture storage table and the feature data storage table.

[0076] In this embodiment, the method installs a network camera at the entrance and exit of the logistics vehicle access control and weighing area, and the system can fully capture the front image of the vehicle and detailed information of the driver's window area, including vehicle color, surface reflectivity, vehicle logo details, headlight status, as well as driver's facial features, window transparency and the number of passengers in the car. This multi-dimensional data collection ensures the comprehensiveness and accuracy of vehicle feature information. Subsequently, by setting up a network, the system can efficiently transmit the collected image data to the central processing center for preprocessing. The preprocessing steps include denoising, size normalization and image set packaging, and finally generate a standard factory entry picture set and a standard factory exit picture set. The optimization of this link ensures the high quality and consistency of the data, laying a solid foundation for subsequent feature analysis. In terms of data storage, by building a MongoDB database, the system realizes automatic classification and index management of the preprocessed picture set, ensuring fast access and efficient storage of data. Compared with the traditional centralized storage method, the distributed database can better handle the management needs of large-scale data and improve the scalability and stability of the system. In summary, through these steps, the system achieves comprehensive data collection, fast transmission and efficient storage in the early stage of vehicle feature detection, which not only improves the accuracy and speed of data processing, but also significantly enhances the overall performance and reliability of the system. Compared with traditional technical means, these improvements bring a higher level of automation and stronger adaptability, thereby improving the efficiency and safety of vehicle management.

[0077] Example 3

[0078] This embodiment is explained in Example 2. Please refer to Figure 1 , specifically: S2 includes S21, S22 and S23;

[0079] S21. Extract the standard factory entry picture set and the standard factory exit picture set in the picture storage table in real time through the write port set in the MongoDB database, and use the convolutional neural network CNN to perform feature extraction on the extracted standard factory entry picture set and the standard factory exit picture set to obtain the vehicle factory entry feature set and the vehicle exit feature set;

[0080] The vehicle entry feature set includes the vehicle entry appearance feature set and the vehicle entry optical feature set;

[0081] The appearance feature set of the incoming vehicle includes the color feature parameter VCP of the incoming vehicle 1 , the recognition accuracy of vehicle license plate characters entering the factory LPRA 1 The matching degree of facial features of the driver of the vehicle entering the factory is DFFMD 1 ;

[0082] The optical characteristic set of the incoming vehicle includes the HBI index of the incoming vehicle 1 And the LDC of the vehicle logo details 1 ;

[0083] The vehicle factory feature set includes a factory vehicle appearance feature set and a factory vehicle optical feature set;

[0084] The factory vehicle appearance feature set includes the factory vehicle color feature parameters VCP 2 , Factory vehicle license plate character recognition accuracy LPRA 2 The matching degree of the driver's facial features with the factory vehicle DFFMD 2 ;

[0085] The factory vehicle optical characteristic set includes the factory vehicle headlight brightness index HBI 2 And the LDC of the vehicle logo details 2 ;

[0086] S22, extract edge, color and texture features from the standard factory entry picture set and the standard factory exit picture set through the convolution operation of the convolutional neural network CNN, and obtain the factory entry vehicle color feature parameter VCP 1 and factory vehicle color characteristic parameters VCP 2 ;

[0087] Through the target detection operation of the convolutional neural network CNN, the license plate identification characters, headlight brightness and vehicle logo details of the vehicles in the standard factory entry picture set and the standard factory exit picture set are detected, and the license plate character recognition accuracy rate LPRA of the vehicles entering the factory is obtained. 1, Factory vehicle license plate character recognition accuracy LPRA 2 , Vehicle headlight brightness index HBI 1 , Factory vehicle headlight brightness index HBI 2 , LDC of vehicle logo details complexity 1 And the LDC of the vehicle logo details 2 ;

[0088] Through the deep convolution layer of the convolutional neural network CNN, the driver's facial features are extracted from the standard factory entry picture set and the standard factory exit picture set to obtain the facial feature matching degree DFFMD of the driver of the factory vehicle. 1 The matching degree of the driver's facial features with the factory vehicle DFFMD 2 ;

[0089] S23. The extracted vehicle entry feature set and vehicle exit feature set are written into the feature data storage table of the MongoDB database through the write port of the distributed database for storage, and the timestamp and license plate characters are marked for each written vehicle entry feature set and vehicle exit feature set.

[0090] In this embodiment, the method uses the real-time write function of the MongoDB database, and the system can quickly extract the standard factory entry and factory exit picture sets in the picture storage table, and use the convolutional neural network CNN to perform in-depth feature extraction on these picture sets. The convolutional neural network CNN effectively extracts the appearance and optical features of the vehicle through convolution operations, such as color parameters, license plate character recognition accuracy, headlight brightness, vehicle logo detail complexity, and driver's facial feature matching. These multidimensional features are sorted into factory entry and factory exit feature sets, respectively, to ensure that the key feature information of the vehicle is fully and accurately identified when entering and leaving the factory. Next, the system stores the extracted vehicle feature set through the write port of the distributed database, and timestamps and license plate character marks each written feature set, which not only realizes efficient data storage, but also provides a reliable basis for subsequent data comparison and analysis. Compared with traditional technical means, the optimization of this process significantly improves the accuracy of data processing and the automation level of the system. The use of distributed databases not only improves the speed of data access, but also enhances the scalability and stability of the system, and can better cope with large-scale data management needs.

[0091] Example 4

[0092] This embodiment is explained in Example 3, please refer to Figure 1 , specifically: S3 includes S31 and S32;

[0093] S31, constructing a vehicle feature similarity scoring formula, inputting the appearance feature set of incoming vehicles and the appearance feature set of outgoing vehicles into the vehicle feature similarity scoring formula, and outputting a vehicle similarity score VFSS;

[0094] The vehicle similarity score VFSS is calculated using the following vehicle feature similarity score formula;

[0095]

[0096] in, Indicates color similarity, represents the similarity of license plate character recognition, It represents the matching degree of the facial features of the vehicle driver. α, β and χ represent the preset weight values ​​of color similarity, license plate character recognition similarity and matching degree of the facial features of the vehicle driver respectively. The specific values ​​are set by the user. VCPmax represents the reference upper limit of the vehicle color feature parameter. LPRAmax represents the reference upper limit of the vehicle license plate character recognition accuracy. DFFMDmax represents the reference upper limit of the matching degree of the facial features of the vehicle driver.

[0097] S32 includes S321 and S322;

[0098] S321, by the vehicle headlight brightness index HBI 1 and the vehicle headlight brightness index HBI 2 And the LDC of the vehicle logo details 1 And the LDC of the vehicle logo details 2 Calculate the difference respectively and output the headlight brightness difference HBI diff Difference between details and car logo LDC diff ;

[0099] Headlight brightness difference HBI diff Difference between details and car logo LDC diff Obtained through the following algorithm formula;

[0100]

[0101] S322, by constructing the vehicle optical characteristic formula, the headlight brightness difference HBI diff Difference between details and car logo LDC diff , input into the vehicle optical characteristic formula, calculate and output the vehicle optical characteristic consistency index VOFCI;

[0102] The vehicle optical characteristic consistency index VOFCI is calculated and obtained by the following vehicle optical characteristic formula;

[0103]

[0104] Among them, e represents the exponential function, δ represents the weight coefficient of the headlight brightness, and ε represents the weight coefficient of the complexity of the car logo details.

[0105] In this embodiment, the method constructs a vehicle feature similarity scoring formula, inputs the appearance feature set of the incoming and outgoing vehicles into the formula, and calculates the vehicle similarity score VFSS. The scoring formula combines key parameters such as color similarity, license plate character recognition similarity, and driver facial feature matching, and optimizes according to the weights set by the user to ensure that each feature is accurately reflected in the similarity calculation. In the further refinement step of S32, S321 calculates the headlight brightness difference HBI of the incoming and outgoing vehicles. diff Difference between details and car logo LDC diff , further refining the analysis of the vehicle's optical features. Based on these differences, S322 calculates the vehicle optical feature consistency index VOFCI through the vehicle optical feature formula, which comprehensively reflects the consistency of the vehicle's optical features. This process not only ensures a comprehensive analysis of the vehicle's appearance and optical features, but also improves the system's sensitivity and recognition ability to changes in vehicle features through precise calculation of the formula.

[0106] Example 5

[0107] This embodiment is explained in Example 4. Please refer to Figure 1 , specifically: S4 includes S41 and S42;

[0108] S41, calculating and outputting a comprehensive similarity index CSEF by combining the obtained vehicle similarity score VFSS and the vehicle optical feature consistency index VOFCI;

[0109] The comprehensive similarity index CSEF is calculated by the following algorithm formula;

[0110]

[0111] S42, the user preliminarily sets a similarity threshold T, and then preliminarily compares and evaluates it with the obtained comprehensive similarity index CSEF, and generates a response mechanism based on the evaluation result. The specific evaluation content is as follows;

[0112] When the comprehensive similarity index CSEF ≥ similarity threshold T, it means the similarity is qualified and the vehicle is automatically released.

[0113] When the comprehensive similarity index CSEF is less than the similarity threshold T, it means that the similarity is unqualified and passage is prohibited. At this time, a response mechanism is generated to dynamically adjust the similarity threshold T.

[0114] In this embodiment, the method generates a comprehensive similarity index CSEF by calculating the vehicle similarity score VFSS and the vehicle optical feature consistency index VOFCI. This comprehensive index effectively integrates the multi-dimensional characteristics of the vehicle and provides a more comprehensive data basis for evaluating the consistency of the vehicle. Then, the user is allowed to set a preliminary similarity threshold T and compare it with the calculated comprehensive similarity index CSEF to determine whether the vehicle meets the release conditions. If the comprehensive similarity index CSEF is higher than or equal to the similarity threshold T, the system will automatically release the vehicle; if it is lower than the threshold, the system will trigger a response mechanism, including prohibiting passage and dynamically adjusting the similarity threshold T. This method greatly improves the judgment accuracy and reliability of the logistics weighing system by comprehensively evaluating multiple characteristic parameters of the vehicle. Compared with the traditional method that relies on a single feature or simple rule judgment, the new method can adapt to different vehicles and environmental changes more flexibly, significantly reducing the risk of misjudgment and missed judgment. The mechanism of dynamically adjusting the threshold further enhances the adaptability of the system, enabling the system to maintain efficient operation under different conditions. Ultimately, these improvements have improved the intelligence level of the system, optimized the vehicle management process, and ensured the safety and efficiency of factory entrance and exit management.

[0115] Example 6

[0116] This embodiment is explained in Example 5. Please refer to Figure 1 , specifically: S5 includes S51 and S52;

[0117] S51. When the similarity is found to be unqualified in the preliminary comparison and evaluation, a dynamic adjustment mechanism of the similarity threshold T is triggered, and a dynamic adjustment threshold Tnew is calculated and outputted based on the historical comprehensive similarity index CSEF by constructing a dynamic threshold adjustment formula;

[0118] The dynamic adjustment threshold Tnew is calculated by the following dynamic threshold adjustment formula:

[0119]

[0120] Among them, ζ represents the adjustment coefficient, which is set by the user according to the actual situation, Average(CSEF) represents the historical comprehensive similarity index CSEF, and Max(CSEF) represents the upper limit value of the comprehensive similarity index CSEF.

[0121] S52, performing a secondary comparative evaluation based on the obtained dynamic adjustment threshold Tnew and the comprehensive similarity index CSEF output in real time, the specific evaluation content is as follows;

[0122] When the comprehensive similarity index CSEF ≥ the dynamic adjustment threshold Tnew, it means that the similarity is qualified and the vehicle is automatically released;

[0123] When the comprehensive similarity index CSEF is less than the dynamically adjusted threshold Tnew, it means that the similarity is unqualified. At this time, passage is prohibited and an early warning message is generated to prompt the staff to conduct manual verification.

[0124] In this embodiment, the method triggers the dynamic adjustment mechanism of the similarity threshold T when the preliminary comparison evaluation shows that the similarity of the vehicle is unqualified. The system calculates and generates a new dynamic threshold Tnew according to the distribution of the historical comprehensive similarity index CSEF by constructing a dynamic threshold adjustment formula. This step ensures that the system can adaptively adjust the judgment criteria to cope with the interference caused by environmental changes or abnormal conditions. Then, a secondary comparison evaluation is performed based on the obtained dynamic adjustment threshold Tnew and the comprehensive similarity index CSEF output in real time. If the comprehensive similarity index CSEF reaches or exceeds the dynamic adjustment threshold Tnew, the system will automatically release the vehicle; otherwise, the system will prohibit passage and generate an early warning message to remind the staff to conduct manual verification. This mechanism improves the system's adaptability to complex environments and changing conditions by dynamically adjusting the threshold, ensuring the flexibility and accuracy of the vehicle management process. Compared with the traditional fixed threshold judgment method, the dynamic adjustment mechanism significantly improves the robustness and intelligence level of the system. Traditional methods are usually unable to cope with dynamic changes in the environment and are prone to misjudgment. However, through the reference and real-time adjustment of historical data, the system can maintain efficient operation and reduce the errors caused by fixed thresholds. This improvement not only optimizes the safety and efficiency of vehicle management, but also reduces the frequency of manual intervention and improves the reliability and operational efficiency of the overall system.

[0125] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The AI-based vehicle feature detection method is characterized by: The following steps are involved: S1. Use a network camera to capture front images of vehicles entering the factory and vehicles leaving the factory, and collect driver window images of vehicles entering the factory and vehicles leaving the factory at the same time, and send the collected front images and driver window images to a central processing center through a network connection for preprocessing, obtain a standard factory entry image set and a standard factory exit image set, and store the preprocessed standard factory entry image set and standard factory exit image set in a database; S2, extracting the standard factory entry picture set and the standard factory exit picture set stored in the database, applying the convolutional neural network CNN to perform feature extraction on the standard factory entry picture set and the standard factory exit picture set, and generating a vehicle factory entry feature set and a vehicle factory exit feature set; S3, constructing a vehicle feature similarity scoring formula and a vehicle optical feature formula, then inputting the extracted vehicle entry feature set and vehicle exit feature set into the vehicle feature similarity scoring formula and the vehicle optical feature formula, and outputting a vehicle similarity score VFSS and a vehicle optical feature consistency index VOFCI; S4, based on the obtained vehicle similarity score VFSS and vehicle optical feature consistency index VOFCI, a comprehensive calculation is performed to output a comprehensive similarity index CSEF, and then a preliminary comparative evaluation is performed between the preset similarity threshold T and the comprehensive similarity index CSEF; S5. Based on the preliminary comparative evaluation results, a dynamic threshold adjustment formula is further constructed, the similarity threshold T is input into the dynamic threshold adjustment formula, the dynamic adjustment threshold Tnew is output, and then a secondary comparative evaluation is performed with the comprehensive similarity index CSEF obtained in real time.

2. The AI-based vehicle feature detection method according to claim 1, characterized in that: Said S1 includes S11, S12 and S13; S11. Install a network camera on the front of the entrance and exit of the logistics vehicle access control and weighing area. The network camera automatically triggers the shooting of vehicles entering and leaving the factory to obtain the front image of the vehicle, which includes vehicle color information, surface reflectivity, vehicle logo details and vehicle light status content information; At the same time, at the side of the entrance and exit of the logistics vehicle access control and weighing area, based on the installed network camera, the driver's window area of ​​the incoming and outgoing vehicles is captured, and the driver's window image includes the driver's face, window transparency and the number of passengers in the car; S12, connecting the network camera to the central processing center by setting up a connection network, transmitting the collected front image and the driver's window image to the central processing center for preprocessing, and obtaining a standard factory entry image set and a standard factory exit image set, wherein the preprocessing method includes denoising, size normalization and image set packaging; S13. Build a MongoDB database, and set a write port and a write output port respectively. Write the preprocessed standard incoming picture set and standard outgoing picture set into the MongoDB database through the write port. After receiving the standard picture data set, the MongoDB database automatically classifies and indexes the standard incoming picture set and the standard outgoing picture set, and stores the standard incoming picture set and the standard outgoing picture set into the picture storage table by setting the picture storage table and the feature data storage table.

3. The AI-based vehicle feature detection method according to claim 2, characterized in that: The S2 includes S21, S22 and S23; S21. Through the write port set in the MongoDB database, the standard factory entry picture set and the standard factory exit picture set in the picture storage table are extracted in real time, and the convolutional neural network CNN is used to extract features of the standard factory entry picture set and the standard factory exit picture set to obtain the vehicle entry feature set and the vehicle exit feature set; The vehicle entry feature set includes an entry vehicle appearance feature set and an entry vehicle optical feature set; The appearance feature set of the incoming vehicle includes the color feature parameter VCP1 of the incoming vehicle, the recognition accuracy rate LPRA1 of the license plate character of the incoming vehicle, and the facial feature matching degree DFFMD1 of the driver of the incoming vehicle; The optical feature set of the incoming vehicle includes the vehicle headlight brightness index HBI1 and the vehicle logo detail complexity LDC1; The vehicle factory feature set includes a factory vehicle appearance feature set and a factory vehicle optical feature set; The factory vehicle appearance feature set includes the factory vehicle color feature parameter VCP2, the factory vehicle license plate character recognition accuracy LPRA2 and the factory vehicle driver facial feature matching degree DFFMD2; The factory vehicle optical feature set includes the factory vehicle headlight brightness index HBI2 and the factory vehicle logo detail complexity LDC2; S22, extracting edge, color and texture features from the standard incoming picture set and the standard outgoing picture set through a convolution operation of a convolutional neural network CNN, and obtaining an incoming vehicle color feature parameter VCP1 and an outgoing vehicle color feature parameter VCP2; Through the target detection operation of the convolutional neural network CNN, the license plate identification characters, headlight brightness and vehicle logo details of the vehicles in the standard factory-entry picture set and the standard factory-exit picture set are detected, and the license plate character recognition accuracy rate LPRA1 of the factory-entry vehicle, the license plate character recognition accuracy rate LPRA2 of the factory-exit vehicle, the headlight brightness index HBI1 of the factory-entry vehicle, the headlight brightness index HBI2 of the factory-exit vehicle, the vehicle logo detail complexity LDC1 of the factory-entry vehicle and the vehicle logo detail complexity LDC2 of the factory-exit vehicle are obtained; The facial features of the drivers in the standard factory entry picture set and the standard factory exit picture set are extracted through the deep convolution layer of the convolutional neural network CNN, and the facial feature matching degree DFFMD1 of the drivers of the factory entry vehicles and the facial feature matching degree DFFMD2 of the drivers of the factory exit vehicles are obtained; S23. The vehicle entry feature set and the vehicle exit feature set are written into the feature data storage table of the MongoDB database through the write port of the distributed database for storage, and the timestamp and license plate characters are marked for each written vehicle entry feature set and vehicle exit feature set.

4. The AI-based vehicle feature detection method according to claim 3, characterized in that: The S3 includes S31 and S32; S31, constructing a vehicle feature similarity scoring formula, inputting the appearance feature set of incoming vehicles and the appearance feature set of outgoing vehicles into the vehicle feature similarity scoring formula, and outputting a vehicle similarity score VFSS; The vehicle similarity score VFSS is calculated by the following vehicle feature similarity score formula; in, Indicates color similarity, represents the similarity of license plate character recognition, It represents the matching degree of the facial features of the vehicle driver. α, β and χ represent the preset weight values ​​of color similarity, license plate character recognition similarity and matching degree of the facial features of the vehicle driver respectively. The specific values ​​are set by the user. VCPmax represents the reference upper limit of the vehicle color feature parameter. LPRAmax represents the reference upper limit of the vehicle license plate character recognition accuracy. DFFMDmax represents the reference upper limit of the matching degree of the facial features of the vehicle driver.

5. The AI-based vehicle feature detection method according to claim 4, characterized in that: The S32 includes S321 and S322; S321, respectively perform difference calculation on the headlight brightness index HBI1 of the incoming vehicle and the headlight brightness index HBI2 of the outgoing vehicle, as well as the vehicle logo detail complexity LDC1 of the incoming vehicle and the vehicle logo detail complexity LDC2 of the outgoing vehicle, and output the headlight brightness difference HBI diff Difference between details and car logo LDC diff ; The headlight brightness difference HBI diff Difference between details and car logo LDC diff Obtained through the following algorithm formula; 6. The AI-based vehicle feature detection method according to claim 5, characterized in that: S322, by constructing the vehicle optical characteristic formula, the headlight brightness difference HBI diff Difference between details and car logo LDC diff , input into the vehicle optical characteristic formula, calculate and output the vehicle optical characteristic consistency index VOFCI; The vehicle optical characteristic consistency index VOFCI is calculated and obtained by the following vehicle optical characteristic formula; Among them, e represents the exponential function, δ represents the weight coefficient of the headlight brightness, and ε represents the weight coefficient of the complexity of the car logo details.

7. The AI-based vehicle feature detection method according to claim 1, characterized in that: The S4 includes S41 and S42; S41, calculating and outputting a comprehensive similarity index CSEF by combining the obtained vehicle similarity score VFSS and the vehicle optical feature consistency index VOFCI; The comprehensive similarity index CSEF is calculated by the following algorithm formula:

8. The AI-based vehicle feature detection method according to claim 1, characterized in that: S42, the user preliminarily sets a similarity threshold T, and then performs a preliminary comparative evaluation with the obtained comprehensive similarity index CSEF, and generates a response mechanism according to the evaluation result. The specific evaluation content is as follows; When the comprehensive similarity index CSEF ≥ similarity threshold T, it means the similarity is qualified and the vehicle is automatically released. When the comprehensive similarity index CSEF is less than the similarity threshold T, it means that the similarity is unqualified and passage is prohibited. At this time, a response mechanism is generated to dynamically adjust the similarity threshold T.

9. The AI-based vehicle feature detection method according to claim 8, characterized in that: The S5 includes S51 and S52; S51. When the similarity is found to be unqualified in the preliminary comparison and evaluation, a dynamic adjustment mechanism of the similarity threshold T is triggered, and a dynamic adjustment threshold Tnew is calculated and outputted based on the historical comprehensive similarity index CSEF by constructing a dynamic threshold adjustment formula; The dynamic adjustment threshold Tnew is calculated and obtained by the following dynamic threshold adjustment formula: Among them, ζ represents the adjustment coefficient, which is set by the user according to the actual situation, Average(CSEF) represents the historical comprehensive similarity index CSEF, and Max(CSEF) represents the upper limit value of the comprehensive similarity index CSEF.

10. The AI-based vehicle feature detection method according to claim 1, characterized in that: S52, performing a secondary comparative evaluation based on the obtained dynamic adjustment threshold Tnew and the comprehensive similarity index CSEF output in real time, the specific evaluation content is as follows; When the comprehensive similarity index CSEF ≥ the dynamic adjustment threshold Tnew, it means that the similarity is qualified and the vehicle is automatically released; When the comprehensive similarity index CSEF is less than the dynamically adjusted threshold Tnew, it means that the similarity is unqualified. At this time, passage is prohibited and an early warning message is generated to prompt the staff to conduct manual verification.

Images