A method and device for detecting a fake license plate vehicle based on a ReID algorithm and related media

Abstract

The application provides a method and device for detecting a fake license plate vehicle based on a ReID algorithm and related media, which comprises the following steps: acquiring video pictures of different road sections, detecting and tracking all vehicles in the video pictures by using a target tracking network, and recording corresponding vehicle snapshots and vehicle information; selecting any vehicle as a target vehicle, detecting the license plate of the target vehicle by using a license plate detection algorithm, and saving a target license plate snapshot of the target license plate; acquiring a target vehicle snapshot of the target license plate, and extracting a first feature vector of the target vehicle snapshot by using a vehicle re-identification algorithm; extracting a second feature vector of the target license plate snapshot by using a license plate recognition algorithm; and correlating vehicles under different road sections based on the first feature vector and the second feature vector by using a matching algorithm, and determining whether there is a fake license plate vehicle according to a correlation result. The application has the advantages of high timeliness, high accuracy, high automation, and the like, and can greatly reduce labor costs.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a method, device, and related medium for detecting cloned vehicles based on the ReID algorithm. Background Technology

[0002] In urban traffic operations, some criminals forge and illegally obtain the license plates, models, and colors of genuine vehicles to engage in smuggling, theft, and other illegal activities. Therefore, it is necessary to use technical means to efficiently detect vehicles with counterfeit license plates.

[0003] Common methods for identifying cloned vehicles mainly include identifying the vehicle, extracting the vehicle's license plate number, comparing the vehicle's registration information and driving trajectory to determine whether it is a cloned vehicle. This method requires the ability to clearly obtain the license plate information, but most cameras on the road cannot capture clear license plates, which is a major limitation. Furthermore, comparing vehicle registration information requires a lot of manpower, resulting in low work efficiency. Summary of the Invention

[0004] This invention provides a method, apparatus, computer equipment, and storage medium for detecting cloned vehicles based on the ReID algorithm, aiming to improve the timeliness and accuracy of detecting cloned vehicles.

[0005] In a first aspect, this invention provides a method for detecting cloned vehicles based on the ReID algorithm, including:

[0006] Acquire video footage from different road sections, and use a target tracking network to detect and track all vehicles in the video footage, and record the corresponding vehicle snapshots and vehicle information;

[0007] Select any one of the vehicles as the target vehicle, use a license plate detection algorithm to detect the license plate of the target vehicle, and save a snapshot of the target license plate.

[0008] Obtain a snapshot of the target vehicle with the target license plate, and use a vehicle re-identification algorithm to extract the first feature vector of the target vehicle snapshot;

[0009] The second feature vector of the target license plate snapshot is extracted using a license plate recognition algorithm;

[0010] Based on the first and second feature vectors, a matching algorithm is used to associate vehicles on different road segments, and the association results are used to determine whether there are vehicles using counterfeit license plates.

[0011] Secondly, embodiments of the present invention provide a cloned vehicle detection device based on the ReID algorithm, comprising:

[0012] The video acquisition unit is used to acquire video footage from different road sections, and to use a target tracking network to detect and track all vehicles in the video footage, and to record the corresponding vehicle snapshots and vehicle information.

[0013] The target selection unit selects any one of the vehicles as the target vehicle, uses a license plate detection algorithm to detect the license plate of the target vehicle, and saves a snapshot of the target license plate.

[0014] The first extraction unit is used to obtain a snapshot of the target vehicle with the target license plate and to extract the first feature vector of the target vehicle snapshot using a vehicle re-identification algorithm.

[0015] The second extraction unit is used to extract the second feature vector of the target license plate snapshot using a license plate recognition algorithm;

[0016] The first judgment unit is used to associate vehicles on different road segments based on the first feature vector and the second feature vector using a matching algorithm, and to determine whether there are vehicles with cloned license plates based on the association results.

[0017] Thirdly, an embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for detecting cloned vehicles based on the ReID algorithm as described in the first aspect.

[0018] Fourthly, an embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for detecting cloned vehicles based on the ReID algorithm as described in the first aspect.

[0019] This invention provides a method, apparatus, computer device, and storage medium for detecting cloned vehicles based on the ReID algorithm. The method includes: acquiring video footage from different road segments, detecting and tracking all vehicles in the video footage using a target tracking network, and recording corresponding vehicle snapshots and vehicle information; selecting any vehicle as a target vehicle, detecting the license plate of the target vehicle using a license plate detection algorithm, and saving a snapshot of the target license plate; acquiring the target vehicle snapshot of the target license plate, and extracting a first feature vector from the target vehicle snapshot using a vehicle re-identification algorithm; extracting a second feature vector from the target license plate snapshot using a license plate recognition algorithm; and, based on the first and second feature vectors, associating vehicles on different road segments using a matching algorithm, and determining whether a cloned vehicle exists based on the association results. This invention utilizes computer vision technology and deep learning algorithms to identify and track road vehicles, associating vehicles on different road segments to determine whether a cloned vehicle exists. This invention can efficiently identify cloned vehicles without querying vehicle registration information, offering advantages such as high timeliness, high accuracy, high automation, and significantly reduced labor costs. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating a method for detecting cloned vehicles based on the ReID algorithm provided in an embodiment of the present invention;

[0022] Figure 2 A schematic diagram of a sub-process of a method for detecting cloned vehicles based on the ReID algorithm provided in an embodiment of the present invention;

[0023] Figure 3 A schematic block diagram of a cloned vehicle detection device based on the ReID algorithm provided in an embodiment of the present invention;

[0024] Figure 4 This is a schematic block diagram of a method for detecting cloned vehicles based on the ReID algorithm, provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0027] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0028] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0029] Please see below. Figure 1 , Figure 1 The flowchart of a method for detecting cloned vehicles based on the ReID algorithm provided in this embodiment of the invention is shown, specifically including steps S101 to S105.

[0030] S101. Acquire video footage from different road sections, and use a target tracking network to detect and track all vehicles in the video footage, and record the corresponding vehicle snapshots and vehicle information.

[0031] S102. Select any one of the vehicles as the target vehicle, use a license plate detection algorithm to detect the license plate of the target vehicle, and save a snapshot of the target license plate of the target vehicle.

[0032] S103. Obtain a snapshot of the target vehicle and extract the first feature vector of the target vehicle snapshot using a vehicle re-identification algorithm;

[0033] S104. Extract the second feature vector of the target license plate snapshot using a license plate recognition algorithm;

[0034] S105. Based on the first feature vector and the second feature vector, a matching algorithm is used to associate vehicles on different road segments, and the association results are used to determine whether there are any vehicles using counterfeit license plates.

[0035] In this embodiment, the acquired video footage is first detected and tracked by a target tracking network. For each detected vehicle, a snapshot and vehicle information are recorded. Then, a target vehicle is selected from the video footage, and its license plate and corresponding snapshot are obtained using a license plate detection algorithm. Next, a vehicle re-identification algorithm (i.e., the ReID algorithm) and a license plate recognition algorithm are used to extract the first feature vector of the vehicle snapshot and the second feature vector of the license plate snapshot. Subsequently, the vehicle is matched and associated with the first and second feature vectors to determine whether there is a cloned vehicle.

[0036] This embodiment utilizes computer vision technology and deep learning algorithms to identify and track road vehicles, thereby associating vehicles on different road sections to determine whether they are using counterfeit license plates. This embodiment can efficiently identify counterfeit vehicles without querying vehicle registration information, and has the advantages of high timeliness, high accuracy, high degree of automation, and significant reduction in labor costs, which is conducive to market promotion.

[0037] In one specific embodiment, step S101 includes:

[0038] The system transmits images captured by cameras on different roads to a server via transmission equipment. The cameras communicate with the server, and for example, the NVIDIA Tesla 4 can communicate with cameras on the road using fiber optic technology to acquire multiple video streams in real time. It supports various video encoding and decoding formats such as H.264 / 265 and MPEG. Its powerful computing capabilities also support AI models for multi-channel video inference. Then, a target tracking network and a license plate detection network capture snapshots of vehicles and their corresponding license plates. A vehicle re-identification network and a license plate recognition network then correlate vehicles on different roads to analyze whether they are using counterfeit license plates.

[0039] In one embodiment, such as Figure 2 As shown, step S101 includes steps S201 to S205.

[0040] S201. Use a target detection network to detect all vehicles in the video frame;

[0041] S202. Capture a snapshot of the detected vehicle and use a license plate detection model to check whether the license plate in the vehicle snapshot is complete.

[0042] S203. If the license plate in the detected vehicle snapshot is complete, record the corresponding vehicle snapshot and the vehicle information at the time of capture.

[0043] S204. If the license plate in the vehicle snapshot is incomplete, then take a new vehicle snapshot and continue to check whether the license plate in the vehicle snapshot is complete until a complete license plate is found in the captured vehicle snapshot.

[0044] S205. Use the Deepsort target tracking model to track the detected vehicles.

[0045] In this embodiment, when detecting and tracking all vehicles in the video footage, a target detection network is first used to detect all vehicles in each frame of the captured road footage. Then, a Deep Sort target tracking model is used to track the detected vehicles. Specifically, the changes in vehicle position before and after are compared and calculated. Combined with the direction of vehicle movement, the same vehicle in the previous and current frames is identified, and each vehicle is detected and tracked. When taking snapshots, only one snapshot is saved for each vehicle. To ensure that the vehicle is complete in the snapshot and is not severely obscured by other vehicles, a license plate detection model is used to detect the captured vehicle snapshots. If a complete license plate is detected, the snapshot is saved; otherwise, the snapshot is discarded, and a new snapshot is taken. At the same time as taking the snapshot, the vehicle information corresponding to the vehicle is recorded, such as the time and location of the snapshot, and the unique ID number assigned by the target tracking network.

[0046] In a specific embodiment, the object detection network is initialized with parameters using a pre-trained model on the publicly available COCO dataset, and then trained using real road scene images. This allows the object detection network to converge faster during training and yields a model with high detection accuracy. Furthermore, the DeepSort object tracking model is an improvement upon Sort object tracking, introducing an appearance feature extraction network. This network is trained offline on a pedestrian re-identification dataset. During real-time object tracking, it extracts the appearance features of the target and performs nearest-neighbor matching, effectively improving object tracking performance under occlusion conditions.

[0047] In one embodiment, step S102 includes:

[0048] A license plate detection network was constructed by training vehicle images containing labeled license plates;

[0049] The license plate detection network is updated using perspective transformation parameters according to the following formula:

[0050]

[0051] In the formula, T(q) represents the position of the license plate rectangle after perspective transformation, q represents the position of the standard license plate rectangle, and v1, v2, v3, v4, v5, and v6 are all perspective transformation parameters.

[0052] The updated license plate detection network is used to detect the license plate of the target vehicle, and a perspective transformation function is used to correct the detected license plate, thereby obtaining a snapshot of the target license plate.

[0053] In this embodiment, the license plate detection network is first trained using vehicle images containing labeled license plates. A labeled license plate refers to a license plate with at least multiple feature points labeled, such as the top left, top right, bottom right, and bottom left corners. Secondly, considering that different road cameras may distort license plates when capturing snapshots, perspective transformation parameters are added during network training to more accurately detect distorted license plate feature points and improve network accuracy. Then, the license plate is corrected into a uniform rectangle using the `getPerspectiveTransform` function in OpenCV (the perspective transformation function) and saved. It should be noted that the `getPerspectiveTransform` function can project an irregular quadrilateral plane onto a rectangular plane using perspective transformation; therefore, in this embodiment, a distorted license plate can be corrected into a rectangular license plate.

[0054] In one embodiment, step S103 includes:

[0055] The target vehicle snapshot is input into the vehicle re-identification network, and passes through the first convolutional layer, the first pooling layer and the first fully connected layer in the vehicle re-identification network in sequence. The first feature vector of the target vehicle snapshot is output by the first fully connected layer.

[0056] In this embodiment, the vehicle re-identification network mainly consists of convolutional layers (i.e., the first convolutional layer), pooling layers (i.e., the first pooling layer), and fully connected layers (i.e., the first fully connected layer). Its network output calculates two loss functions: one branch outputs a 1024-dimensional feature vector for calculating the triplet loss, and the other branch outputs a 576-dimensional vector for calculating the cross-entropy loss, thus obtaining the first feature vector. In a specific embodiment, the vehicle re-identification network can be jointly trained using publicly available datasets, including the VeRi and VehicleID datasets, as well as vehicle data from real-world road scenarios, until the network converges, resulting in a vehicle re-identification model with strong representational capabilities suitable for real-world road scenarios. Then, convolutional layer calculations are performed on each captured vehicle snapshot, and pooling layers are used to reduce the computational load and accelerate the network's operation. Finally, a 1024-bit feature vector, i.e., the first feature vector, is obtained through the first fully connected layer. The first feature vector extracted from each snapshot is mapped to the vehicle's unique ID number, and this information is saved.

[0057] In one embodiment, step S104 includes:

[0058] The target license plate snapshot is input into the license plate recognition network, and then sequentially passes through the second convolutional layer, the second pooling layer, and the second fully connected layer in the license plate recognition network. The second feature vector of the target license plate snapshot is output by the second fully connected layer.

[0059] In this embodiment, the license plate recognition network mainly consists of a classification neural network composed of a convolutional layer (i.e., the second convolutional layer), a pooling layer (i.e., the second pooling layer), and a fully connected layer (i.e., the second fully connected layer). The training set uses actual license plate number data, the network training input is a license plate number image, and the output is the license plate number. This trains a license plate recognition classification network model with strong representation ability suitable for actual road scenarios, i.e., the license plate recognition network. Then, convolution calculation is performed on each captured license plate snapshot, and the second pooling layer is used to reduce the computational load of the model and speed up the network's operation. Finally, a 256-dimensional feature vector is obtained through the second fully connected layer, i.e., the second feature vector. The second feature vector of each captured snapshot is mapped to the unique ID number of the vehicle, and this information is saved.

[0060] In one embodiment, step S105 includes:

[0061] Calculate the first Euclidean distance between the first feature vectors under different road segments, and calculate the second Euclidean distance between the second feature vectors under different road segments;

[0062] Determine whether a vehicle with a fake license plate exists based on the first and second Euclidean distances.

[0063] Furthermore, it also includes:

[0064] When the first Euclidean distance is less than a preset first distance threshold, it is determined that vehicles in different road segments are associated and matched.

[0065] When the second Euclidean distance is less than the preset second distance threshold, it is determined that the license plates of vehicles on different road segments are associated and matched.

[0066] If vehicles on different road segments are not associated and matched, but their license plates are associated and matched, then it is determined that there is a vehicle using a cloned license plate;

[0067] If vehicles on different road segments are associated and matched, and vehicle license plates are also associated and matched, then the presence of cloned vehicles can be determined based on the vehicle information.

[0068] In this embodiment, when determining whether a target vehicle is a cloned vehicle, a first Euclidean distance is first calculated for the first feature vectors of vehicle snapshots captured from different cameras. If the first Euclidean distance is less than a preset first distance threshold, it indicates that the two vehicles have a high degree of similarity in appearance, i.e., the two vehicles are associated and matched; otherwise, it indicates that the two vehicles have a low degree of similarity in appearance, i.e., the two vehicles are not associated and matched. Next, a second Euclidean distance is calculated for the second feature vector of the license plate snapshot corresponding to the vehicle. If the second Euclidean distance is less than a preset second distance threshold, it indicates that the two license plates are similar, i.e., the two license plates are associated and matched; otherwise, it indicates that the two license plates are not associated and matched. Then, based on the association matching results, it is determined whether a cloned vehicle exists. Specifically: if two license plates are associated and matched but the two vehicles are not associated and matched, it is determined to be a cloned vehicle. For example, if two vehicles have similar license plates but significant differences in appearance, it is considered a cloned vehicle. If two license plates are associated and matched and the two vehicles are associated and matched, it is determined to be a non-cloned vehicle. For example, if two vehicles have similar license plates but significant differences in appearance, it is considered a non-cloned vehicle.

[0069] Furthermore, when two license plates and two vehicles are associated and matched, it is possible to determine whether they are cloned vehicles based on vehicle information. For example, by referring to the appearance time and location information of the two vehicles, it can be determined whether they are cloned vehicles. Specifically, if the two vehicles appear at the same time but have different location information, it is determined that there is a cloned vehicle. Or, if the appearance time of the two vehicles is similar but the location information is significantly different, it can also be determined that there is a cloned vehicle.

[0070] Figure 3 This is a schematic block diagram of a cloned vehicle detection device 300 based on the ReID algorithm provided in an embodiment of the present invention. The device 300 includes:

[0071] The video acquisition unit 301 is used to acquire video footage from different road sections, and to use a target tracking network to detect and track all vehicles in the video footage, and to record the corresponding vehicle snapshots and vehicle information.

[0072] The target selection unit 302 selects any one of the vehicles as the target vehicle, uses a license plate detection algorithm to detect the license plate of the target vehicle, and saves a snapshot of the target license plate.

[0073] The first extraction unit 303 is used to obtain a snapshot of the target vehicle with the target license plate and to extract the first feature vector of the target vehicle snapshot using a vehicle re-identification algorithm.

[0074] The second extraction unit 304 is used to extract the second feature vector of the target license plate snapshot using a license plate recognition algorithm;

[0075] The first judgment unit 305 is used to associate vehicles on different road segments based on the first feature vector and the second feature vector using a matching algorithm, and to determine whether there are vehicles with cloned license plates based on the association results.

[0076] In one embodiment, such as Figure 4 As shown, the video acquisition unit 301 includes:

[0077] The vehicle detection unit 401 is used to detect all vehicles in the video frame using a target detection network;

[0078] The snapshot capture unit 402 is used to capture a snapshot of the detected vehicle and detect whether the license plate in the vehicle snapshot is complete using a license plate detection model.

[0079] The information recording unit 403 is used to determine if the license plate in the detected vehicle snapshot is complete, and then record the corresponding vehicle snapshot and the vehicle information at the time of capture.

[0080] The re-capture unit 404 is used to determine if the license plate in the detected vehicle snapshot is incomplete, and then re-capture the vehicle snapshot and continue to check whether the license plate in the detected vehicle snapshot is complete until a complete license plate is found in the captured vehicle snapshot.

[0081] The vehicle tracking unit 405 is used to track detected vehicles using the Deepsort target tracking model.

[0082] In one embodiment, the target selection unit 302 includes:

[0083] The network construction unit is used to train and construct a license plate detection network using vehicle images containing labeled license plates;

[0084] A license plate detection unit is used to detect the license plate of the target vehicle, including updating the parameters of the license plate detection network according to the following formula using perspective transformation parameters:

[0085]

[0086] In the formula, T(q) represents the position of the license plate rectangle after perspective transformation, q represents the position of the standard license plate rectangle, and v1, v2, v3, v4, v5, and v6 are all perspective transformation parameters.

[0087] The license plate correction unit is used to detect the license plate of the target vehicle using the updated license plate detection network, and to correct the detected license plate using a perspective transformation function, thereby obtaining a snapshot of the target license plate.

[0088] In one embodiment, the first extraction unit 303 includes:

[0089] The first feature vector output unit is used to input the target vehicle snapshot into the vehicle re-identification network, and sequentially pass it through the first convolutional layer, the first pooling layer and the first fully connected layer in the vehicle re-identification network, and output the first feature vector of the target vehicle snapshot by the first fully connected layer.

[0090] In one embodiment, the second extraction unit 304 includes:

[0091] The second feature vector output unit is used to input the target license plate snapshot into the license plate recognition network, and sequentially pass it through the second convolutional layer, the second pooling layer and the second fully connected layer in the license plate recognition network, and output the second feature vector of the target license plate snapshot by the second fully connected layer.

[0092] In one embodiment, the first determination unit 305 includes:

[0093] The distance calculation unit is used to calculate the first Euclidean distance between the first feature vectors under different road segments, and to calculate the second Euclidean distance between the second feature vectors under different road segments;

[0094] The distance determination unit is used to determine whether a vehicle with a fake license plate exists based on the first Euclidean distance and the second Euclidean distance.

[0095] In one embodiment, the distance determination unit includes:

[0096] The first determination unit is used to determine the vehicle association matching under different road segments when the first Euclidean distance is less than a preset first distance threshold.

[0097] The second determination unit is used to determine the vehicle license plate association matching under different road segments when the second Euclidean distance is less than a preset second distance threshold.

[0098] The third determination unit is used to determine the existence of a cloned vehicle if vehicles on different road segments are not associated and matched but their license plates are associated and matched.

[0099] The second judgment unit is used to determine whether there is a cloned vehicle based on the vehicle information if vehicles on different road segments are associated and matched and vehicle license plates are associated and matched.

[0100] Since the embodiments of the apparatus and the embodiments of the method correspond to each other, please refer to the description of the embodiments of the method for the embodiments of the apparatus, which will not be repeated here.

[0101] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed, can perform the steps provided in the above embodiments. The storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0102] This invention also provides a computer device, which may include a memory and a processor. The memory stores a computer program, and when the processor calls the computer program in the memory, it can implement the steps provided in the above embodiments. Of course, the computer device may also include various network interfaces, power supplies, and other components.

[0103] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0104] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for detecting cloned vehicles based on the ReID algorithm, characterized in that, include: Acquire video footage from different road sections, and use a target tracking network to detect and track all vehicles in the video footage, and record the corresponding vehicle snapshots and vehicle information; Select any one of the vehicles as the target vehicle, use a license plate detection algorithm to detect the license plate of the target vehicle, and save a snapshot of the target vehicle's license plate. Obtain a snapshot of the target vehicle and extract the first feature vector of the target vehicle snapshot using a vehicle re-identification algorithm; The second feature vector of the target license plate snapshot is extracted using a license plate recognition algorithm; Based on the first and second feature vectors, a matching algorithm is used to associate vehicles on different road segments, and the association results are used to determine whether there are vehicles using counterfeit license plates. The process of using a target tracking network to detect and track all vehicles in the video footage, and recording corresponding vehicle snapshots and information, includes: An object detection network is used to detect all vehicles in the video footage; The system captures a snapshot of the detected vehicle and uses a license plate detection model to check whether the license plate in the snapshot is complete. If the license plate in the vehicle snapshot is complete, record the corresponding vehicle snapshot and the vehicle information at the time of capture. If the license plate in the vehicle snapshot is incomplete, a new vehicle snapshot is captured, and the process of checking whether the license plate in the vehicle snapshot is complete continues until a complete license plate is captured in the new vehicle snapshot. The detected vehicles are tracked using a Deep sort target tracking model; The step of selecting any one of the vehicles as the target vehicle, detecting the license plate of the target vehicle using a license plate detection algorithm, and saving a snapshot of the target license plate includes: A license plate detection network was constructed by training vehicle images containing labeled license plates; The license plate detection network is updated using perspective transformation parameters according to the following formula: In the formula, T(q) represents the position of the license plate rectangle after perspective transformation, q represents the position of the standard license plate rectangle, and v1, v2, v3, v4, v5, and v6 are all perspective transformation parameters. The updated license plate detection network is used to detect the license plate of the target vehicle, and the detected license plate is corrected using a perspective transformation function to obtain a snapshot of the target license plate. The step of associating vehicles on different road segments using a matching algorithm based on the first and second feature vectors, and determining whether there are cloned vehicles based on the association results, includes: Calculate the first Euclidean distance between the first feature vectors under different road segments, and calculate the second Euclidean distance between the second feature vectors under different road segments; Determine whether a vehicle with a fake license plate exists based on the first and second Euclidean distances; When the first Euclidean distance is less than a preset first distance threshold, it is determined that vehicles in different road segments are associated and matched. When the second Euclidean distance is less than the preset second distance threshold, it is determined that the license plates of vehicles on different road segments are associated and matched. If vehicles on different road segments are not associated and matched, but their license plates are associated and matched, then it is determined that there is a vehicle using a cloned license plate; If vehicles on different road segments are associated and matched, and vehicle license plates are also associated and matched, then the presence of cloned vehicles can be determined based on the vehicle information.

2. The method for detecting cloned vehicles based on the ReID algorithm according to claim 1, characterized in that, The step of obtaining a snapshot of the target vehicle with the target license plate and extracting a first feature vector of the target vehicle snapshot using a vehicle re-identification algorithm includes: The target vehicle snapshot is input into the vehicle re-identification network, and passes through the first convolutional layer, the first pooling layer and the first fully connected layer in the vehicle re-identification network in sequence. The first feature vector of the target vehicle snapshot is output by the first fully connected layer.

3. The method for detecting cloned vehicles based on the ReID algorithm according to claim 1, characterized in that, The second feature vector extracted from the target license plate snapshot using the license plate recognition algorithm includes: The target license plate snapshot is input into the license plate recognition network, and then sequentially passes through the second convolutional layer, the second pooling layer, and the second fully connected layer in the license plate recognition network. The second feature vector of the target license plate snapshot is output by the second fully connected layer.

4. A device for detecting cloned vehicles based on the ReID algorithm, characterized in that, include: The video acquisition unit is used to acquire video footage from different road sections, and to use a target tracking network to detect and track all vehicles in the video footage, and to record the corresponding vehicle snapshots and vehicle information. The target selection unit selects any one of the vehicles as the target vehicle, uses a license plate detection algorithm to detect the license plate of the target vehicle, and saves a snapshot of the target license plate of the target vehicle. The first extraction unit is used to obtain a snapshot of the target vehicle with the target license plate and to extract the first feature vector of the target vehicle snapshot using a vehicle re-identification algorithm. The second extraction unit is used to extract the second feature vector of the target license plate snapshot using a license plate recognition algorithm; The first judgment unit is used to associate vehicles on different road segments based on the first feature vector and the second feature vector using a matching algorithm, and to determine whether there are vehicles with cloned license plates based on the association results. The video acquisition unit includes: The vehicle detection unit is used to detect all vehicles in the video frame using a target detection network; The snapshot capture unit is used to capture a snapshot of the detected vehicle and to detect whether the license plate in the vehicle snapshot is complete using a license plate detection model. The information recording unit is used to determine if the license plate in the detected vehicle snapshot is complete, and then record the corresponding vehicle snapshot and the vehicle information at the time of capture. The re-capture unit is used to determine if the license plate in the detected vehicle snapshot is incomplete, and then re-capture the vehicle snapshot and continue to check whether the license plate in the detected vehicle snapshot is complete until a complete license plate is found in the captured vehicle snapshot. The vehicle tracking unit is used to track detected vehicles using a deep sort target tracking model; The target selection unit includes: The network construction unit is used to train and construct a license plate detection network using vehicle images containing labeled license plates; A license plate detection unit is used to detect the license plate of the target vehicle, including updating the parameters of the license plate detection network according to the following formula using perspective transformation parameters: In the formula, T(q) represents the position of the license plate rectangle after perspective transformation, q represents the position of the standard license plate rectangle, and v1, v2, v3, v4, v5, and v6 are all perspective transformation parameters. The license plate correction unit is used to detect the license plate of the target vehicle using the updated license plate detection network, and to correct the detected license plate using a perspective transformation function, thereby obtaining a snapshot of the target license plate. The first determination unit includes: The distance calculation unit is used to calculate the first Euclidean distance between the first feature vectors under different road segments, and to calculate the second Euclidean distance between the second feature vectors under different road segments; The distance determination unit is used to determine whether a vehicle with a fake license plate exists based on the first Euclidean distance and the second Euclidean distance. The first determination unit is used to determine the vehicle association matching under different road segments when the first Euclidean distance is less than a preset first distance threshold. The second determination unit is used to determine the vehicle license plate association matching under different road segments when the second Euclidean distance is less than a preset second distance threshold. The third determination unit is used to determine the existence of a cloned vehicle if vehicles on different road segments are not associated and matched but their license plates are associated and matched. The second judgment unit is used to determine whether there is a cloned vehicle based on the vehicle information if vehicles on different road segments are associated and matched and vehicle license plates are associated and matched.

5. A computer device, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for detecting cloned vehicles based on the ReID algorithm as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements a method for detecting cloned vehicles based on the ReID algorithm as described in any one of claims 1 to 3.

Images