Method and system for detecting vehicle detection replacement behavior of truck

Abstract

The invention relates to the technical field of vehicle detection, and provides a method and a system for detecting a vehicle detection replacement behavior of a truck, and the method comprises the steps: S1, collecting the unique identification information of a vehicle when the vehicle enters a detection station, and capturing a multi-angle real-time image; s2, querying a vehicle archive database based on the unique identification information, and obtaining a registration image and historical features of the vehicle; s3, carrying out license plate and VIN code consistency comparison on the real-time image and the registered image, and extracting vehicle macroscopic features and fine-grained invariant features to carry out similarity calculation; s4, if the comparison results are inconsistent or the similarity is lower than a preset threshold value, triggering real-time alarm; s5, constructing a vehicle detection data base library, and storing images, detection data and source information detected each time; and S6, carrying out time sequence consistency analysis and one-vehicle multi-image correlation analysis on the base database data, and outputting a detection-replacement suspicion report. According to the invention, the detection replacement behavior of the truck can be well detected.

Description

A method and system for detecting vehicle inspection substitution in large trucks Technical Field

[0001] This invention relates to the field of vehicle inspection technology, and more specifically, to a method and system for detecting vehicle inspection fraud in large trucks. Background Technology

[0002] The practice of using substitute inspection vehicles for large trucks, commonly known as "vehicle substitution inspection" or "truck spoofing," refers to the act of using a compliant large truck of the same model but in good condition (referred to as a "spoof truck" or "sample truck") to impersonate a non-compliant truck in order to pass official vehicle safety inspections. This allows the problematic truck to fraudulently obtain a legal inspection certificate. The proliferation of this practice is the result of multiple factors, rooted in the conflict between substantial economic interests and stringent regulatory requirements.

[0003] Currently, the identification and prevention of truck inspection fraud mainly relies on the following methods:

[0004] Manual comparison based on experience: Inspection station staff rely on visual comparison between the vehicle registration certificate photo and the actual vehicle appearance. This method has limited ability to distinguish subtle differences between vehicles of the same model and color, is inefficient, and is easily affected by subjective factors.

[0005] Single-dimensional authentication based on license plate or VIN: This method uses the license plate number or Vehicle Identification Number (VIN) as the primary or even unique identifier for verification. However, criminals can easily bypass this type of check by forging or misusing license plates, or by tampering with or obscuring the VIN.

[0006] Segmented technical inspection: Some regions or systems have adopted technologies such as video surveillance and OBD (On-Board Diagnostics) data reading. However, monitoring and analysis are often fragmented, lack real-time performance, and lack in-depth, comprehensive characterization and continuous tracking of the vehicle's "visual identity".

[0008] In summary, existing technologies have the following main shortcomings: identity authentication is limited to a single dimension and is easily circumvented; there is a lack of automated extraction and comparison capabilities for fine-grained visual features that are unique to vehicles and difficult to tamper with; real-time warnings are disconnected from in-depth post-event analysis; and the detection model lacks robustness and adaptability when facing increasingly sophisticated technical countermeasures (such as OBD cheating and sophisticated camouflage).

[0009] Therefore, there is an urgent need for a method and system for detecting vehicle inspection substitution in large trucks to solve the above problems. Summary of the Invention

[0010] The present invention provides a method and system for detecting vehicle inspection substitution behavior in large trucks, which can overcome some or all the defects of the prior art.

[0011] A method for detecting vehicle inspection substitution behavior in large trucks according to the present invention includes the following steps:

[0012] Step S1: When the vehicle enters the inspection station, collect the vehicle's unique identification information and capture real-time images from multiple angles;

[0013] Step S2: Query the vehicle file database based on the unique identifier information to obtain the vehicle's registration image and historical features;

[0014] Step S3: Compare the license plate and VIN code consistency between the real-time image and the registered image, and extract the macroscopic features and fine-grained invariant features of the vehicle to calculate the similarity.

[0015] Step S4: If the comparison results are inconsistent or the similarity is lower than the preset threshold, a real-time alarm will be triggered;

[0016] Step S5: Construct a vehicle detection data base library to store images, detection data, and source information for each detection;

[0017] Step S6: Perform time-series consistency analysis and one-vehicle-multiple-image correlation analysis on the base database data, and output a report on suspected substitution inspection.

[0018] As a preferred approach, the extraction of fine-grained invariant features employs a multi-branch fusion network, including a license plate region recognition branch, a truck bed modification recognition branch, and a crash bar detail recognition branch.

[0019] As a preferred method, when recognizing license plate areas, a coordinate system based on the width of the vehicle's rear is established to calculate the relative positions of characters. Anti-interference design is implemented to eliminate interference from temporary stickers, dirt, or other contaminants. Specifically:

[0020] 1) Establishment of the reference coordinate system;

[0021] Let the width of the vehicle's front be W and the height be H. Establish a normalized coordinate system:

[0022]

[0023] x, y are the original pixel coordinates of the image, x ' ,y ' These are the normalized coordinates;

[0024] 2) Character position feature vector;

[0025] For each character c i Detect the center coordinates of the bounding box:

[0026] p i =(x ' i ,y ' i )

[0027] Forming a position feature matrix:

[0028]

[0029] P represents the feature matrix of the center positions of all characters on the license plate, and n is the total number of characters on the license plate;

[0030] 3) Calculation of positional similarity;

[0031] Use Hausdorff distance to measure location differences:

[0032]

[0033] d H P represents the position matrix of two characters A and P B The Hausdorff distance between two sets of points measures the maximum difference in their spatial distribution; p a p b They are P A and P B The coordinates of the points in the matrix; Sup represents the supremum, i.e., the maximum distance; inf represents the infrem, i.e., the minimum distance;

[0034] Similarity conversion:

[0035]

[0036] sim pos λ represents the similarity of license plate positions, and λ represents the attenuation coefficient.

[0037] As a preferred method, during the truck bed modification identification process, a difference map is generated based on the height of the cargo bed side panels, the shape of the rear bumper beam, and the detection area of ​​the suspension fixing points. Specifically:

[0038] (1) Diagram showing the differences in the structure of the truck bed;

[0039] Generate a difference heatmap using a pre-trained Diffusion model:

[0040]

[0041] in, Indicates query image I query With reference image I refThe difference map in the feature space, where φ(·) is the feature encoder;

[0042] (2) Similarity of the modified areas;

[0043] Calculate the Intersection over Union (IoU) ratio of the sensitive modification region:

[0044]

[0045] M query M ref These represent the modified region masks in the query image and the reference image, respectively;

[0046] Comprehensive assessment of structural similarity:

[0047]

[0048] α is the weighting coefficient, and β is the adjustment factor. It is the Frobenius norm.

[0049] As a preferred method, when identifying details of the bumper strip, local feature descriptors are used, and the RANSAC algorithm is employed to remove outliers. Specifically:

[0050] a) Local feature descriptors;

[0051] Extracting keypoint descriptors using the ORB+FREAK algorithm:

[0052]

[0053] Let d represent the set of extracted key points. i The binary descriptor represents the i-th key point, and m represents the total number of key points extracted.

[0054] b) Feature matching degree calculation;

[0055] Hamming distance d H match:

[0056]

[0057] Indicates an indicator function;

[0058] RANSAC interior rate ρ:

[0059]

[0060] Where, N inliersN represents the number of interior points. matches The initial total number of matching point pairs is represented by H, which is the homography matrix, and the error threshold is ∈ = 3px;

[0061] c) Similarity in details of the anti-collision strips;

[0062] First, calculate the LBP feature histogram similarity:

[0063]

[0064] h A (k), h B (k) represents the value of the k-th bin in the LBP histograms of images A and B, respectively; K represents the total number of bins in the histogram;

[0065] Detailed similarity decomposition:

[0066] sim detail =γ·ρ+(1-γ)·sim texture γ=0.7

[0067] γ is the weighting coefficient.

[0068] As a preferred method, similarity calculation adopts a weighted multi-feature fusion approach, with the following formula:

[0069] sim total =w p ·sim pos +w s ·sim struct +w d ·sim detail

[0070] Where, sim pos Indicates the similarity of license plate positions, sim struct Indicating similarity in truck bed structure, sim detail Indicates the similarity of details of the bumper strips, w p w s w d These are the weighting coefficients.

[0071] As a preferred approach, the timing consistency analysis is as follows:

[0072] If the appearance features of the same vehicle are compared over time in the images of previous inspections, and a sudden change in features occurs in adjacent inspections while the vehicle model remains the same, it is determined to be a suspected case of substitute inspection.

[0073] This invention provides a system for detecting vehicle inspection substitution behavior in large trucks, which employs the aforementioned method for detecting vehicle inspection substitution behavior in large trucks and includes:

[0074] The data acquisition module is used to acquire the vehicle's unique identification information and capture multi-angle real-time images of the vehicle when it enters the inspection station.

[0075] The data query module is used to query the vehicle file database based on the unique identifier information to obtain the vehicle's registration image and historical feature information;

[0076] The feature comparison module is used to compare the license plate and VIN code consistency between the real-time image and the registered image, and to extract and compare the macroscopic features and fine-grained invariant features of the vehicle based on the vehicle re-identification or fine-grained image recognition model.

[0077] The real-time early warning module is used to trigger a real-time alarm and notify regulatory personnel when the feature comparison module detects inconsistencies in the license plate and VIN code or a similarity of vehicle appearance features below a preset threshold.

[0078] The data base module is used to store the inspection records of each inspection station, including vehicle images, inspection data, timestamps and source information;

[0079] The data analysis module is used to perform time-series consistency analysis and correlation analysis on the vehicle historical inspection records in the database to identify vehicles or gangs suspected of substitute inspections.

[0080] Preferably, the fine-grained invariant feature includes at least one of the following:

[0081] The location and shape of the vehicle inspection sticker, personalized stickers or decals on the vehicle body, scratches or dents on the vehicle body, tire brand and wear condition, guardrail style, and gantry shape.

[0082] The beneficial effects of this invention are as follows:

[0083] This invention represents a leap from license plate recognition to vehicle recognition, significantly improving the immutability and accuracy of identity authentication. By introducing vehicle re-identification and fine-grained image recognition technologies, this invention no longer relies solely on easily forged license plates or VIN codes. Instead, it comprehensively extracts and compares the vehicle's inherent, difficult-to-replicate microscopic visual features (such as body scratches, personalized stickers, and component wear). This authentication method, based on the vehicle's inherent visual DNA, fundamentally solves the problem of license plate counterfeiting and significantly improves the accuracy of identifying counterfeit vehicles of the same model.

[0084] This invention not only designs an online blocking mechanism for real-time image comparison and alarm at the inspection station, but also constructs a base library that aggregates all historical inspection data, supporting time-series consistency analysis and multi-image correlation analysis for a single vehicle.

[0085] To address common evasion tactics such as obscured license plates, modified truck beds, and camouflaged exteriors, this invention designs dedicated recognition branches (license plate location, truck bed structure, and bumper details) and employs robust algorithms based on Hausdorff distance, difference map generation, and local feature matching. Finally, a scientific weighted fusion strategy is used to derive a comprehensive similarity score, effectively eliminating interference from temporary adhesives, mud, and changes in lighting, ensuring the stability and reliability of judgments in various complex scenarios.

[0086] By directly embedding the risk assessment model into the vehicle online inspection process, the system can automatically conduct risk assessment and early warning the moment an inspection occurs, and immediately trigger an audible and visual alarm to interrupt the process. This dynamic risk interception mode transforms supervision from traditional post-event spot checks to real-time blocking during the process, greatly reducing the room for cheaters to operate and improving the response speed and prevention effectiveness of supervision. Attached Figure Description

[0087] Figure 1 is a flowchart of a method for detecting vehicle inspection substitution behavior in a large truck according to an embodiment. Detailed Implementation

[0088] To further understand the content of this invention, a detailed description of the invention will be provided in conjunction with the accompanying drawings and embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0089] Example

[0090] As shown in Figure 1, this embodiment provides a method for detecting vehicle inspection substitution behavior in large trucks, which includes the following steps:

[0091] Step S1: When the vehicle enters the inspection station, collect the vehicle's unique identification information and capture real-time images from multiple angles; specifically:

[0092] 1.1) Trigger signal: When a vehicle enters the inspection station, the system triggers the signal or the scanner / VIN code reader obtains the vehicle's unique identifier (such as license plate number or VIN code).

[0093] 1.2) Image capture: Call the high-definition cameras deployed on the detection line to capture real-time images of the vehicle from multiple angles (including at least the front view, rear view, and close-up of the vehicle identification number).

[0094] Step S2: Query the vehicle file database based on the unique identifier information to obtain the vehicle's registration image and historical features; specifically: using the obtained license plate number / VIN code as keywords, query the vehicle data base or vehicle management office business database in real time to obtain the registration image and historical features in the vehicle file.

[0095] Step S3: Compare the license plate and VIN code consistency between the real-time image and the registered image, and extract the vehicle's macro features (model, brand, color) and fine-grained invariant features (annual inspection sticker placement and shape, vehicle body personal stickers or decals, vehicle body scratches or dents, tire brand and wear condition, guardrail style, and gantry shape) for similarity calculation.

[0096] Step S4: If the comparison results are inconsistent or the similarity is lower than the preset threshold, a real-time alarm will be triggered.

[0097] Step S5: Build a vehicle detection data base library to store the images, detection data and source information for each detection.

[0098] The method for constructing the data base is as follows:

[0099] Data source: Compilation of every test record from all testing stations.

[0100] Stored content (each record should contain):

[0101] Unique identifiers: license plate number, VIN code.

[0102] Timestamp: Detection time.

[0103] Visual data: High-resolution images and video clips captured from multiple angles during each detection.

[0104] Test data: All test results, including emissions data, braking data, chassis data, etc.

[0105] Source information: Testing station ID, testing line ID.

[0106] Step S6: Perform time-series consistency analysis and multi-image association analysis on the base database data, and output a suspected substitute inspection report. The suspected substitute inspection report can be submitted to law enforcement agencies for targeted post-incident crackdowns on vehicle owners and inspection stations suspected of substitute inspections.

[0107] One vehicle, multiple images: Correlation analysis

[0108] (6.1) Using vehicle Re-ID technology, search for other vehicles in the massive database that are highly similar to the target vehicle (especially vehicles that have been warned) but have different license plate numbers.

[0109] (6.2) Discover the pattern: Identify vehicles that frequently appear in different inspection records and act as multiple "problem cars" or "substitute cars".

[0110] Anomalies in the correlation between test data and vehicle profiles: A "vehicle condition profile" is created for each vehicle, for example, an old truck whose emissions data have been on the verge of exceeding standards for many years. If the vehicle's test data suddenly becomes perfect in a certain test, but its appearance characteristics do not match the historical records, then this test is highly suspected of being a substitute test.

[0111] In this embodiment, fine-grained invariant feature extraction employs a multi-branch fusion network, including a license plate region recognition branch, a truck bed modification recognition branch, and a bumper strip detail recognition branch. The key model selections are: License plate region: YOLOv11 + key point detection (for locating character corner points); Modification recognition: Mask R-CNN (for truck bed structure segmentation); Detail features: LBP + SIFT + depth hashing (for bumper strip rust and welding points).

[0112] In this embodiment, during license plate area recognition, a coordinate system based on the width of the vehicle's rear is established to calculate the relative positions of characters. Anti-interference design is implemented to eliminate interference from temporary adhesives, dirt, and other contaminants. Specifically:

[0113] 1) Establishment of the reference coordinate system;

[0114] Let the width of the vehicle's front be W and the height be H. Establish a normalized coordinate system:

[0115]

[0116] x, y are the original pixel coordinates of the image, x ' ,y ' These are the normalized coordinates;

[0117] 2) Character position feature vector;

[0118] For each character c i Detect the center coordinates of the bounding box:

[0119] p i =(x ' i ,y ' i )

[0120] Forming a position feature matrix:

[0121]

[0122] P represents the feature matrix of the center positions of all characters on the license plate, and n is the total number of characters on the license plate;

[0123] 3) Calculation of positional similarity;

[0124] Use Hausdorff distance to measure location differences:

[0125]

[0126] d H P represents the position matrix of two charactersA and P B The Hausdorff distance between two sets of points measures the maximum difference in their spatial distribution; p a p b They are P A and P B The coordinates of the points in the matrix; Sup represents the supremum, i.e., the maximum distance; inf represents the infrem, i.e., the minimum distance;

[0127] Similarity conversion:

[0128]

[0129] sim pos λ represents the similarity of license plate positions, and λ represents the attenuation coefficient.

[0130] In this embodiment, during the identification of modified cargo beds, a difference map is generated for the height of the cargo bed side panels, the shape of the rear bumper beam, and the detection area of ​​the suspension fixing points. Specifically:

[0131] (1) Diagram showing the differences in the structure of the truck bed;

[0132] Generate a difference heatmap using a pre-trained Diffusion model:

[0133]

[0134] in, Indicates query image I query With reference image I ref The difference map in the feature space, where φ(·) is the feature encoder;

[0135] (2) Similarity of the modified areas;

[0136] Calculate the Intersection over Union (IoU) ratio of the sensitive modification region:

[0137]

[0138] M query M ref These represent the modified region masks in the query image and the reference image, respectively;

[0139] Comprehensive assessment of structural similarity:

[0140]

[0141] α is the weighting coefficient, and β is the adjustment factor. It is the Frobenius norm.

[0142] In this embodiment, during the identification of details of the anti-collision strip, local feature descriptors are used, and the RANSAC algorithm is employed to remove outliers. Specifically:

[0143] a) Local feature descriptors;

[0144] Extracting keypoint descriptors using the ORB+FREAK algorithm:

[0145]

[0146] Let d represent the set of extracted key points. i The binary descriptor represents the i-th key point, and m represents the total number of key points extracted.

[0147] b) Feature matching degree calculation;

[0148] Hamming distance d H match:

[0149]

[0150] Indicates an indicator function;

[0151] RANSAC interior rate ρ:

[0152]

[0153] Where, N inliers N represents the number of interior points. matches The initial total number of matching point pairs is represented by H, which is the homography matrix, and the error threshold is ∈ = 3px;

[0154] c) Similarity in details of the anti-collision strips;

[0155] First, calculate the LBP feature histogram similarity:

[0156]

[0157] h A (k), h B (k) represents the value of the k-th bin in the LBP histograms of images A and B, respectively; K represents the total number of bins in the histogram;

[0158] Detailed similarity decomposition:

[0159] sim detail =γ·ρ+(1-γ)·sim texture γ=0.7

[0160] γ is the weighting coefficient.

[0161] In this embodiment, similarity calculation adopts a weighted multi-feature fusion method, and the formula is:

[0162] sim total =w p ·sim pos +w s ·sim struct +w d ·sim detail

[0163] Where, sim pos Indicates the similarity of license plate positions, sim struct Indicating similarity in truck bed structure, sim detail Indicates the similarity of details of the bumper strips, w p w s w d These are the weighting coefficients. The similarity calculations here are mainly used for correlation analysis of multiple images of a single vehicle.

[0164] In this embodiment, the timing consistency analysis is as follows:

[0165] If the appearance features of the same vehicle are compared over time in the images of previous inspections, and a sudden change in features occurs in adjacent inspections while the vehicle model remains the same, it is determined to be a suspected case of substitute inspection.

[0166] This embodiment provides a system for detecting vehicle inspection substitution behavior in large trucks, which employs the aforementioned method for detecting vehicle inspection substitution behavior in large trucks and includes:

[0167] The data acquisition module is used to acquire the vehicle's unique identification information and capture multi-angle real-time images of the vehicle when it enters the inspection station.

[0168] The data query module is used to query the vehicle file database based on the unique identifier information to obtain the vehicle's registration image and historical feature information;

[0169] The feature comparison module is used to compare the license plate and VIN code consistency between the real-time image and the registered image, and to extract and compare the macroscopic features and fine-grained invariant features of the vehicle based on the vehicle re-identification or fine-grained image recognition model.

[0170] The real-time early warning module is used to trigger a real-time alarm and notify regulatory personnel when the feature comparison module detects inconsistencies in the license plate and VIN code or a similarity of vehicle appearance features below a preset threshold.

[0171] The data base module is used to store the inspection records of each inspection station, including vehicle images, inspection data, timestamps and source information;

[0172] The data analysis module is used to perform time-series consistency analysis and correlation analysis on the vehicle historical inspection records in the database to identify vehicles or gangs suspected of substitute inspections.

[0173] This embodiment achieves a leap from license plate recognition to vehicle recognition, greatly improving the immutability and accuracy of identity authentication. By introducing vehicle re-identification and fine-grained image recognition technologies, this embodiment no longer relies solely on easily forged license plates or VIN codes, but comprehensively extracts and compares the vehicle's inherent, difficult-to-replicate microscopic visual features. This authentication method based on the vehicle's intrinsic visual DNA fundamentally solves the problem of license plate spoofing and significantly improves the accuracy of identifying spoofing of the same model of vehicle.

[0174] The present invention and its embodiments have been described above illustratively. This description is not restrictive, and the figures shown are only one embodiment of the present invention; the actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for detecting vehicle inspection substitution behavior in large trucks, characterized in that: The process includes the following steps: Step S1: When a vehicle enters the inspection station, collect the vehicle's unique identification information and capture real-time images from multiple angles; Step S2: Based on the unique identification information, query the vehicle's file database to obtain the vehicle's registration image and historical features; Step S3: Compare the license plate and VIN code consistency between the real-time image and the registration image, and extract the vehicle's macroscopic features and fine-grained invariant features for similarity calculation; Step S4: If the comparison results are inconsistent or the similarity is lower than a preset threshold, trigger a real-time alarm; Step S5: Construct a vehicle inspection data base database to store the images, inspection data, and source information for each inspection; Step S6: Perform time-series consistency analysis and one-vehicle-multiple-image association analysis on the base database data, and output a suspected substitute inspection report.

2. The method for detecting vehicle inspection substitution behavior in large trucks according to claim 1, characterized in that: Fine-grained invariant feature extraction employs a multi-branch fusion network, including a license plate region recognition branch, a truck bed modification recognition branch, and a crash bar detail recognition branch.

3. The method for detecting vehicle inspection substitution behavior in large trucks according to claim 2, characterized in that: When recognizing license plates, a coordinate system based on the width of the vehicle's rear is established to calculate the relative positions of characters. Anti-interference design is implemented to eliminate interference from temporary stickers, dirt, or other contaminants. Specifically: 1) Establishing a reference coordinate system; assuming the width of the vehicle's front is W and the height is H, a normalized coordinate system is established: x, y are the original pixel coordinates of the image, x ' ,y ' 1) Normalized coordinates; 2) Character position feature vector; for each character c i Detect bounding box center coordinates: p i =(x ' i ,y ' i Forming a position feature matrix: P represents the feature matrix of the center positions of all characters on the license plate, and n is the total number of characters on the license plate; 3) Calculation of positional similarity; using Hausdorff distance to measure positional differences: d H P represents the position matrix of two characters A and P B The Hausdorff distance between two sets of points measures the maximum difference in their spatial distribution; p a p b They are P A and P B The coordinates of the points in the matrix; Sup represents the supremum, i.e., the maximum distance; inf represents the infremum, i.e., the minimum distance; similarity transformation: sim pos λ represents the similarity of license plate positions, and λ represents the attenuation coefficient.

4. The method for detecting vehicle inspection substitution behavior in large trucks according to claim 3, characterized in that: When identifying the modified cargo bed, a difference map is generated for the height of the cargo bed sideboard, the shape of the rear anti-collision beam, and the detection area of ​​the suspension fixing point. Specifically: (1) Cargo bed structure difference map; a difference heat map is generated using a pre-trained Diffusion model: in, Indicates query image I query With reference image I ref In the feature space, the difference map is φ(·), which is the feature encoder; (2) Modification region similarity; calculate the intersection-union ratio (IoU) of the modification-sensitive region: M query M ref These represent the modified region masks in the query image and the reference image, respectively; comprehensive structural similarity evaluation: α is the weighting coefficient, and β is the adjustment factor. It is the Frobenius norm.

5. The method for detecting vehicle inspection substitution behavior in large trucks according to claim 4, characterized in that: When identifying details of the bumper strip, local feature descriptors are used, and the RANSAC algorithm is employed to remove outliers. Specifically: a) Local feature descriptors; the ORB+FREAK algorithm is used to extract key point descriptors. Let d represent the set of extracted key points. i b) The binary descriptor of the i-th keypoint, where m represents the total number of extracted keypoints; d) Feature matching degree calculation; Hamming distance d H match: Indicates the indicator function; RANSAC interior rate ρ: Where, N inliers N represents the number of interior points. matches Represents the total number of initial matching point pairs, H is the homography matrix, and the error threshold is ∈=3px; c) Anti-collision bar detail similarity; First calculate the LBP feature histogram similarity: h A (k), h B (k) represents the value of the k-th bin in the LBP histograms of images A and B, respectively; K represents the total number of bins in the histogram; detail similarity decomposition: sim detail =γ·ρ+(1-γ)·sim texture γ = 0.7γ is the weighting coefficient.

6. The method for detecting vehicle inspection substitution behavior in large trucks according to claim 5, characterized in that: Similarity calculation uses a weighted multi-feature fusion method, and the formula is: sim total =w p ·sim pos +w s ·sim struct +w d ·sim detail Where, sim pos Indicates the similarity of license plate positions, sim struct Indicating similarity in truck bed structure, sim detail Indicates the similarity of details of the bumper strips, w p w s w d These are the weighting coefficients.

7. The method for detecting vehicle inspection substitution behavior in large trucks according to claim 6, characterized in that: Temporal consistency analysis involves comparing the appearance features of the same vehicle across different detection images over time. If a sudden change in features occurs in adjacent detections while the vehicle model remains unchanged, it is considered a suspected case of substitute detection.

8. A system for detecting vehicle inspection substitution behavior in large trucks, characterized in that: It employs a method for detecting vehicle inspection substitution behavior of large trucks as described in any one of claims 1-7, and includes: a data acquisition module, used to acquire the unique identification information of the vehicle when it enters the inspection station, and capture multi-angle real-time images of the vehicle; a data query module, used to query the vehicle file database based on the unique identification information to obtain the vehicle's registration image and historical feature information; a feature comparison module, used to compare the license plate and VIN code consistency between the real-time image and the registration image, and extract and compare the vehicle's macroscopic features and fine-grained invariant features based on a vehicle re-identification or fine-grained image recognition model; a real-time early warning module, used to trigger a real-time alarm and notify supervisory personnel when the feature comparison module detects inconsistencies in the license plate and VIN code or that the similarity of the vehicle's appearance features is lower than a preset threshold; a data base module, used to store the inspection records of each inspection station, including vehicle images, inspection data, timestamps, and source information; and a data analysis module, used to perform time-series consistency analysis and correlation analysis on the vehicle historical inspection records in the data base to identify suspected inspection substitution vehicles or inspection substitution gangs.

9. The vehicle inspection substitution detection system for large trucks according to claim 8, characterized in that: Fine-grained invariant features include at least one of the following: the location and shape of the annual inspection sticker, the vehicle body's personalized stickers or decals, the vehicle body's scratches or dents, the tire brand and wear condition, the guardrail style, and the gantry shape.

Images