Fake-licensed vehicle detection method and device and electronic equipment

Abstract

The invention provides a fake-licensed vehicle detection method, a fake-licensed vehicle detection device and electronic equipment. The method comprises the following steps: obtaining a vehicle image of a to-be-detected vehicle collected by a first camera in a road network; performing feature extraction on the vehicle image by using the target detection model to obtain to-be-detected vehicle features of the to-be-detected vehicle; determining an upstream driving track on which the to-be-detected vehicle is driven; on the basis of the upstream driving track, predicting a downstream driving track of the to-be-detected vehicle to obtain a predicted downstream driving track; determining a prediction time period when the to-be-detected vehicle passes through the predicted downstream driving track; obtaining a downstream vehicle image collected by a second camera in the predicted downstream driving track within a prediction time period; performing feature extraction on the downstream vehicle image by using the target detection model to obtain corresponding predicted vehicle features; and if the predicted vehicle feature is consistent with the to-be-detected vehicle feature and the license plate information of the to-be-detected vehicle is not consistent with the license plate information of the vehicle in the downstream image, determining that the to-be-detected vehicle is a fake-licensed vehicle.

Description

Technical Field

[0001] The present application relates to detection technology, and more particularly to a method, device and electronic equipment for detecting cloned vehicles. Background Art

[0002] Accurate identification of vehicles with fake license plates is an important technical problem that needs to be solved in highway toll auditing. Vehicle face recognition-based detection of fake license plates is technically and implementably feasible. At present, the main technical idea of ​​vehicle face recognition is to analyze the identity of vehicles based on the similarity of feature values ​​of passing vehicle images; specifically, the passing vehicle images are processed to obtain the complete vehicle face image and cut it, and the feature values ​​of each unit after cutting are calculated to form a feature value array, and the feature value similarity is analyzed to determine the identity of the vehicle.

[0003] The current method of detecting fake license plates uses a data set to pre-train a target detection neural network, uses image processing technology to identify vehicle feature information from the collected vehicle images, uses a neural network to predict vehicle features and store them, and compares the many detection results obtained to find the target vehicle with the highest similarity. This method has certain limitations:

[0004] 1. Most vehicles of the same type have high similarity in facial features. Recognition and comparison based only on facial features will produce large errors, and many unrelated vehicles may be identified as the same vehicle, affecting the detection effect.

[0005] 2. This method requires processing and comparison of all vehicle detection data, which consumes a lot of storage and computing resources. Summary of the invention

[0006] The embodiments of the present application provide a method, device, electronic device and computer-readable storage medium for detecting cloned vehicles, which can reduce the amount of data processing, ensure detection accuracy and save computing resources.

[0007] The technical solution of the embodiment of the present application is implemented as follows:

[0008] The present application provides a method for detecting a fake license plate vehicle, including:

[0009] Obtaining a vehicle image of the vehicle to be tested captured by a first camera in the road network;

[0010] Using the target detection model, extracting features from the vehicle image to obtain the vehicle features of the vehicle to be tested;

[0011] Determining the upstream driving trajectory that the vehicle to be tested has traveled;

[0012] Based on the upstream driving trajectory, predicting the downstream driving trajectory of the vehicle to be tested to obtain a predicted downstream driving trajectory;

[0013] Determining a predicted time period for the vehicle to be tested to pass through the predicted downstream driving trajectory;

[0014] Obtaining a downstream vehicle image captured by a second camera in the predicted downstream driving trajectory within a predicted time period;

[0015] Using the target detection model, extracting features from the downstream vehicle image to obtain corresponding predicted vehicle features;

[0016] If the predicted vehicle feature is consistent with the feature of the vehicle to be tested, and the license plate information of the vehicle to be tested is inconsistent with that of the vehicle in the downstream image, it is determined that the vehicle to be tested is a fake license plate vehicle.

[0017] In the above solution, the target detection model includes at least two layers, and the at least two layers include a feature extraction layer. The target detection model is used to extract features of the vehicle image to obtain the vehicle features to be tested, including:

[0018] Performing image segmentation on the vehicle image to obtain a first image and a second image, wherein the first image includes a vehicle face of the vehicle to be tested;

[0019] Using each layer of the target detection model, extracting features from the first image to obtain the face features of the vehicle to be tested;

[0020] The feature extraction layer of the target detection model is used to extract features from the second image to obtain features of the roof of the vehicle to be tested.

[0021] In the above solution, the step of determining the upstream driving trajectory of the vehicle to be tested includes:

[0022] Extracting at least one upstream trajectory of the vehicle to be tested from the road network;

[0023] Determine a target time period for the vehicle to be tested to pass through the upstream driving trajectory;

[0024] Obtaining an upstream vehicle image captured by a camera arranged in the at least one upstream track within the target time period;

[0025] Using the feature extraction layer of the target detection model respectively, extract features from each of the upstream vehicle images to obtain corresponding upstream vehicle roof features;

[0026] Determine, based on the upstream vehicle image corresponding to at least one upstream trajectory, a target upstream trajectory where a first similarity between an upstream vehicle roof feature and the vehicle roof feature to be measured reaches a first similarity condition;

[0027] The target upstream trajectory is determined as the upstream driving trajectory of the vehicle to be tested.

[0028] In the above scheme, the predicted vehicle features include predicted vehicle features and predicted roof features, the predicted vehicle face features are obtained in the same manner as the vehicle face features to be tested, and the predicted roof features are obtained in the same manner as the vehicle roof features to be tested; the method further includes:

[0029] Determining a second similarity between the predicted vehicle roof feature and the vehicle roof feature to be measured;

[0030] If the second similarity meets the second similarity condition, determining a third similarity between the predicted vehicle face feature and the vehicle face feature to be tested;

[0031] If the third similarity meets a third similarity condition, it is determined that the predicted vehicle feature is consistent with the feature of the vehicle to be tested.

[0032] In the above solution, the at least two layers further include a feature fusion layer and a detection head, and the various layers of the target detection model are used to extract features from the first image to obtain the face features of the vehicle to be tested, including:

[0033] Using the feature extraction layer, converting the first image into at least two first feature maps;

[0034] Using the feature fusion layer to perform feature fusion processing on the at least two first feature maps to obtain at least one second feature map;

[0035] The detection head is used to perform target detection on the at least one second feature map to obtain the vehicle face features of the vehicle to be detected.

[0036] In the above scheme, the feature extraction layer includes at least one sub-extraction layer and a large separation convolution attention LSKA module layer, and the first image is converted into at least two first feature maps by using the feature extraction layer, including:

[0037] The first image is processed with features in sequence using each sub-extraction layer and a large separation convolution attention (LSKA) module layer to obtain at least two first feature maps.

[0038] In the above scheme, the detection head includes a first convolutional layer, a shared convolutional layer, a second convolutional layer and a loss function layer connected in sequence, the first convolutional layer includes at least two first sub-convolutional layers respectively connected to the shared convolutional layer, the second convolutional layer includes at least two second sub-convolutional layers respectively connected to the shared convolutional layer, and the loss function layer includes sub-loss function layers respectively connected one-to-one to the at least two second sub-convolutional layers.

[0039] In the above scheme, the method further comprises:

[0040] Build an object detection model;

[0041] Obtaining a sample vehicle image and sample vehicle features;

[0042] Using the target detection model, extracting features from the sample vehicle image to obtain corresponding sample predicted vehicle features;

[0043] Based on the error between the sample vehicle feature and the sample predicted vehicle feature, the model parameters of the target detection model are updated.

[0044] The present application provides a device for detecting a fake license plate vehicle, including:

[0045] A first acquisition module is used to obtain a vehicle image of the vehicle to be tested collected by a first camera in the road network;

[0046] A first feature extraction module is used to extract features from the vehicle image using a target detection model to obtain the vehicle features of the vehicle to be tested;

[0047] A first determination module is used to determine the upstream driving track that the vehicle to be tested has traveled;

[0048] A prediction module, used for predicting the downstream driving trajectory of the vehicle to be tested based on the upstream driving trajectory to obtain a predicted downstream driving trajectory;

[0049] A second determination module is used to determine a predicted time period for the vehicle to be tested to pass through the predicted downstream driving trajectory;

[0050] A second acquisition module is used to obtain a downstream vehicle image captured by a second camera in the predicted downstream driving trajectory within a predicted time period;

[0051] A second feature extraction module, used to extract features from the downstream vehicle image using the target detection model to obtain corresponding predicted vehicle features;

[0052] The third determination module is used to determine that the vehicle to be tested is a fake license plate vehicle if the predicted vehicle feature is consistent with the feature of the vehicle to be tested and the license plate information of the vehicle to be tested is inconsistent with that of the vehicle in the downstream image.

[0053] An embodiment of the present application provides an electronic device, including:

[0054] A memory for storing executable instructions;

[0055] The processor is used to implement the method for detecting cloned vehicles provided in the embodiment of the present application when executing the executable instructions stored in the memory.

[0056] An embodiment of the present application provides a computer-readable storage medium storing executable instructions for causing a processor to execute instructions to implement the method for detecting cloned vehicles provided in the embodiment of the present application.

[0057] An embodiment of the present application provides a computer program product storing a computer program for implementing the method for detecting cloned vehicles provided in the embodiment of the present application when executed by a processor.

[0058] The embodiments of the present application can reduce the amount of data processing by predicting trajectories and improving models, thereby ensuring detection accuracy and saving computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is an optional structural diagram of a fake license plate vehicle detection system provided in an embodiment of the present application;

[0060] Figure 2 is an optional structural diagram of an electronic device provided in an embodiment of the present application;

[0061] Figure 3 This is an optional flowchart of the method for detecting a license plate cloned vehicle provided in an embodiment of the present application;

[0062] Figure 4 is an optional structural diagram of the target detection model provided in the embodiment of the present application;

[0063] Figure 5 It is an optional structural diagram of the detection head provided in the embodiment of the present application. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0065] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0066] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described here can be implemented in an order other than that illustrated or described here.

[0067] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0068] The embodiments of the present application provide a method, device, electronic device and computer-readable storage medium for detecting cloned vehicles, which can reduce the amount of data processing, ensure detection accuracy and save computing resources.

[0069] First, the license plate-cloning vehicle detection system provided in the embodiment of the present application is described. Figure 1 , Figure 1 It is an optional structural diagram of the fake license plate vehicle detection system 100 provided in the embodiment of the present application. In order to realize supporting a fake license plate vehicle detection application, the terminal 103 is connected to the camera 101 through the network 102. In some embodiments, the terminal 103 can be a laptop, a tablet computer, a desktop computer, a smart phone, a dedicated messaging device, a portable gaming device, a smart speaker, a smart watch, etc., but is not limited thereto. The camera 101 can be an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDN, Content Delivery Network) services, and big data and artificial intelligence platforms. The network 102 can be a wide area network or a local area network, or a combination of the two. The terminal 103 and the camera 101 can be directly or indirectly connected by wired or wireless communication, which is not limited in the embodiment of the present application.

[0070] Next, the electronic device provided in the embodiment of the present application for implementing the above-mentioned method for detecting a fake license plate vehicle is described. Figure 2 , Figure 2 is an optional structural diagram of an electronic device 200 provided in an embodiment of the present application. In practical applications, the electronic device 200 can be implemented as Figure 1 The terminal 103 or the camera 101 in the embodiment of the present invention is an electronic device. Figure 1 Taking the terminal 103 shown as an example, the electronic device for implementing the method for detecting cloned vehicles in the embodiment of the present application is described.

[0071] Figure 2 The electronic device 200 shown includes: at least one processor 201, a memory 205, at least one network interface 202 and a user interface 203. The various components in the electronic device 200 are coupled together via a bus system 204. It is understood that the bus system 204 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 204 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 204 is not described in detail. Figure 2 Various buses are labeled as bus system 204 .

[0072] The processor 201 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0073] The user interface 203 includes one or more output devices 2031 that enable presentation of media content, including one or more speakers and / or one or more visual display screens. The user interface 203 also includes one or more input devices 2032, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0074] The memory 205 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard disk drives, optical disk drives, etc. The memory 205 may optionally include one or more storage devices that are physically remote from the processor 201.

[0075] The memory 205 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 205 described in the embodiment of the present application is intended to include any suitable type of memory.

[0076] In some embodiments, the memory 205 can store data to support various operations. Examples of these data include programs, modules, and data structures, or subsets or supersets thereof. In the embodiment of the present application, the memory 205 stores an operating system 2051, a network communication module 2052, a presentation module 2053, an input processing module 2054, and a fake license plate vehicle detection device 2055; specifically,

[0077] Operating system 2051, including system programs for processing various basic system services and performing hardware-related tasks, such as framework layer, core library layer, driver layer, etc., for implementing various basic services and processing hardware-based tasks;

[0078] A network communication module 2052, for reaching other computing devices via one or more (wired or wireless) network interfaces 202, exemplary network interfaces 202 include: Bluetooth, Wireless Compatibility Certification (WiFi), and Universal Serial Bus (USB);

[0079] a presentation module 2053 for enabling presentation of information (e.g., a user interface for operating peripherals and displaying content and information) via one or more output devices 2031 (e.g., a display screen, a speaker, etc.) associated with the user interface 203;

[0080] The input processing module 2054 is used to detect one or more user inputs or interactions from one of the one or more input devices 2032 and translate the detected inputs or interactions.

[0081] In some embodiments, the fake license plate vehicle detection device provided in the embodiments of the present application can be implemented in a software manner. Figure 2 The device 2055 for detecting a fake license plate vehicle stored in the memory 205 is shown, which may be software in the form of a program or a plug-in, and includes the following software modules: a first obtaining module 20551, a first extracting module 20552, a second extracting module 20553, and a second obtaining module 20554. These modules are logical, and thus may be arbitrarily combined or further separated according to the functions implemented. The functions of each module will be described below.

[0082] In other embodiments, the cloned car detection device provided in the embodiments of the present application can be implemented in hardware. As an example, the cloned car detection device provided in the embodiments of the present application can be a processor in the form of a hardware decoding processor, which is programmed to execute the cloned car detection method provided in the embodiments of the present application. For example, the processor in the form of a hardware decoding processor can adopt one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), or other electronic components.

[0083] The method for detecting cloned vehicles provided in the embodiment of the present application will be described in combination with the exemplary application and implementation of the terminal provided in the embodiment of the present application.

[0084] See also Figure 3 , Figure 3 is an optional flow chart of the method for detecting a fake license plate vehicle provided in the embodiment of the present application, which will be combined with Figure 3 The steps shown are explained.

[0085] Step 301, obtaining a vehicle image of a vehicle to be tested captured by a first camera in a road network;

[0086] Step 302, using the target detection model, extracting features from the vehicle image to obtain the vehicle features of the vehicle to be tested;

[0087] Step 303, determining the upstream driving track that the vehicle to be tested has traveled;

[0088] Step 304, predicting the downstream driving trajectory of the vehicle to be tested based on the upstream driving trajectory to obtain a predicted downstream driving trajectory;

[0089] Step 305, determining a predicted time period for the vehicle to be tested to pass through the predicted downstream driving trajectory;

[0090] Step 306, obtaining a downstream vehicle image captured by a second camera in the predicted downstream driving trajectory within a predicted time period;

[0091] Step 307, using the target detection model, extracting features from the downstream vehicle image to obtain corresponding predicted vehicle features;

[0092] Step 308: If the predicted vehicle feature is consistent with the feature of the vehicle to be tested, and the license plate information of the vehicle to be tested is inconsistent with that of the vehicle in the downstream image, then it is determined that the vehicle to be tested is a fake license plate vehicle.

[0093] In actual implementation, any camera point in the road network is selected as a detection point for fake license plate vehicles, that is, it is determined as the first camera. For any vehicle passing record detected at this point, the vehicle image and detection time are extracted to form the image data to be detected. In one embodiment, weather data and the like can also be used as vehicle image attribute data. In an embodiment, weather data can be synchronously acquired from meteorological monitoring, road condition monitoring and other systems through the detection time. Here, the vehicle corresponding to the vehicle passing record is determined as the vehicle to be tested. Then, the target detection model is used to extract features from the vehicle image to obtain the features of the vehicle to be tested. Here, the target detection model can be, for example, a YOLO-V8 network. Corresponding feature recognition and extraction are performed through target detection. In actual implementation, license plate recognition is also performed on the vehicle image to obtain the license plate information of the vehicle to be tested.

[0094] In step 303, the upstream driving trajectory of the vehicle to be tested is further determined. Here, the upstream driving trajectory is the driving path of the vehicle to be tested before the license plate-copying vehicle monitoring point, which can be specifically composed of the positions of multiple cameras located upstream of the first camera in the road network. Here, based on all vehicle images captured by multiple cameras upstream, the license plate information of all cameras passing through the multiple cameras upstream can be determined, and based on the license plate information corresponding to the vehicle to be tested, the upstream camera passed by the vehicle to be tested can be determined, thereby determining the upstream driving trajectory based on the upstream camera.

[0095] In some embodiments, step 303 includes: extracting at least one upstream trajectory of the vehicle to be tested from the road network; determining a target time period for the vehicle to be tested to pass through the upstream driving trajectory; obtaining an upstream vehicle image captured by a camera set in the at least one upstream trajectory within the target time period; using the target detection model, performing feature extraction on the upstream vehicle image to obtain corresponding upstream vehicle features; based on the corresponding upstream vehicle image in at least one upstream trajectory, determining a target upstream trajectory whose similarity between the upstream vehicle features and the features of the vehicle to be tested reaches a target similarity threshold; and determining the target upstream trajectory as the upstream driving trajectory of the vehicle to be tested.

[0096] Here, at least one upstream trajectory of the vehicle to be tested in the road network is determined. For each upstream trajectory, the upstream vehicle image captured by the camera in the corresponding upstream trajectory within the target time period is obtained. Here, the N-level cameras located upstream of the first camera can be selected as an upstream trajectory. N is a positive integer greater than or equal to 1. The target time period is the time range for the vehicle to be tested to pass through the upstream trajectory. Specifically, based on the detection time t and the path mileage from the upstream detection point to the cloned vehicle detection point, the maximum travel time T1 is calculated as the time range for upstream trajectory extraction. Here, the passing vehicle pictures of all cameras with all n-level detection points upstream in the [t-T1, t] time period can be extracted to obtain the upstream vehicle image of the upstream trajectory.

[0097] Then, the target detection model is used to extract features from the upstream vehicle image to obtain the upstream vehicle features. Next, for each upstream trajectory, the target upstream trajectory is calculated such that the similarity between the upstream vehicle features corresponding to the upstream vehicle images collected by each camera in the upstream trajectory and the features of the vehicle to be tested reaches the target similarity threshold. Here, the similarity can be calculated using cosine similarity.

[0098] Next, in step 304, based on the upstream driving trajectory, the downstream driving trajectory of the vehicle to be tested is predicted to obtain a predicted downstream driving trajectory. Here, a vehicle trajectory prediction model can be used for prediction. The encoder-decoder structure of the Transformer can be modified to form an improved vehicle trajectory prediction model. The modification content includes: (1) In the encoding area, the attention module is replaced by a gated convolution encoder GConv with time complexity. GConv includes two parts: a channel-by-channel processor (Deepwise Processor) and a gated convolution channel mixer (Gated Convolution ChannelMixer); (2) In the decoding area, a convolutional neural network auxiliary detector is used for learning and training through a joint training method. The model of the encoder-decoder structure of the modified Transformer is trained using a vehicle trajectory sample data set to obtain a vehicle trajectory prediction model. In some embodiments, the vehicle trajectory prediction model is also trained using sample data. The sample data includes basic data and a vehicle trajectory prediction data set. Here, the basic data includes highway network topology data and camera point data; based on the network topology and camera point, the camera point topology structure is obtained, that is, the road connection relationship and driving direction of the camera point; the vehicle trajectory prediction data set includes the vehicle license plate information, vehicle driving trajectory, driving time period, and in one embodiment, the vehicle trajectory prediction data set also includes weather, etc. The vehicle driving trajectory is identified based on the same license plate information for consistency, and is composed of camera points with vehicle passing records and vehicle passing time, and a time series is formed in the order of vehicle passing time. For the relationship between adjacent points in the time series, connectivity detection is performed based on the camera topology structure. If it is not connected, the accuracy and integrity of the detection point, time, and license plate information are quality inspected, so that there is a connectivity relationship between adjacent points in the time series; this data set is used for training the improved vehicle trajectory prediction model; in the embodiment, the historical data sampling is 1-2 years of historical data of the entire road network.

[0099] In actual implementation, the predicted time period of the vehicle to be tested passing through the predicted downstream driving trajectory is determined. Here, the predicted time period can be determined based on the speed limit of the road network, and the traffic time interval in which the vehicle to be tested completely passes through the downstream driving trajectory is determined as the predicted time period. Here, the travel time threshold T2 is set, and the passing vehicle pictures collected by the downstream waypoint camera within the predicted time [t, t+T2] are extracted, that is, the downstream vehicle image. Using the target detection model, the downstream vehicle image is feature extracted to obtain the corresponding predicted vehicle features.

[0100] In actual implementation, if the predicted vehicle feature is consistent with the feature of the vehicle to be tested, and the license plate information of the vehicle to be tested and the vehicle in the downstream image is inconsistent, then the vehicle to be tested is determined to be a fake license plate vehicle. Here, the similarity between the predicted vehicle feature and the feature of the vehicle to be tested can be calculated. If the similarity reaches a corresponding threshold, it is determined that the predicted vehicle feature is consistent with the feature of the vehicle to be tested. Here, the license plate information includes the license plate number. When it is determined that the vehicle features are consistent, it can be determined that the two are the same vehicle, and then the license plate information of the vehicle in the downstream image with consistent vehicle features is identified. If the license plate information is inconsistent with the license plate information of the vehicle to be tested detected at the first camera, it is determined that the license plate of the vehicle to be tested has been changed midway, thereby determining that the vehicle to be tested is a fake license plate vehicle.

[0101] In an embodiment of the present application, a vehicle image of a vehicle to be tested captured by a first camera in a road network is obtained; a target detection model is used to perform feature extraction on the vehicle image to obtain the vehicle features of the vehicle to be tested; an upstream driving trajectory that the vehicle to be tested has traveled is determined; based on the upstream driving trajectory, a downstream driving trajectory of the vehicle to be tested is predicted to obtain a predicted downstream driving trajectory; a predicted time period in which the vehicle to be tested passes through the predicted downstream driving trajectory is determined; a downstream vehicle image captured by a second camera in the predicted downstream driving trajectory within the predicted time period is obtained; a feature extraction is performed on the downstream vehicle image using the target detection model to obtain corresponding predicted vehicle features; if the predicted vehicle features are consistent with the features of the vehicle to be tested, and the license plate information of the vehicle to be tested and the vehicle in the downstream image is inconsistent, then it is determined that the vehicle to be tested is a fake license plate vehicle. By improving the predicted trajectory and model, the amount of data processing can be reduced, and computing resources can be saved while ensuring detection accuracy.

[0102] In some embodiments, the target detection model includes at least two layers, and the at least two layers include a feature extraction layer. The step 302 includes: performing image segmentation on the vehicle image to obtain a first image and a second image, wherein the first image includes the face of the vehicle to be tested; using each layer of the target detection model, performing feature extraction on the first image to obtain the face features of the vehicle to be tested; using the feature extraction layer of the target detection model, performing feature extraction on the second image to obtain the roof features of the vehicle to be tested.

[0103] Here, the target detection model is an improved structure based on YOLO-V8. It includes at least two layers, and at least two layers include a feature extraction layer. In actual implementation, the vehicle face portion of the vehicle image is cut to form a first image containing a partial image of the vehicle face and a second image containing the remaining image. Then, the vehicle image is input into the target detection model, and the first image is feature extracted using each layer of the target detection model, and the face features of the vehicle to be tested are output from the last layer of the target detection model. The feature extraction layer of the target detection model is used to extract features from the second image to obtain the roof features of the vehicle to be tested output by the feature extraction layer. The face features of the vehicle to be tested and the roof features of the vehicle to be tested are determined as the features of the vehicle to be tested. Specifically, the vehicle image is input into the improved target detection model; the first image is output as the features of the vehicle to be tested via the complete network structure; the second image is only processed by the feature extraction layer of the target detection model, and the feature map FM is output. v , use global average pooling (GeMpooling) for aggregation processing to obtain the roof feature FV to be tested v .

[0104] In some embodiments, determining the upstream driving trajectory that the vehicle to be tested has traveled includes: extracting at least one upstream trajectory of the vehicle to be tested from the road network; determining a target time period for the vehicle to be tested to pass through the upstream driving trajectory; obtaining an upstream vehicle image captured by a camera set in the at least one upstream trajectory within the target time period; performing feature extraction on each of the upstream vehicle images using the feature extraction layer of the target detection model to obtain corresponding upstream roof features; determining a target upstream trajectory in which a first similarity between the upstream roof features and the roof features to be tested meets a first similarity condition based on the corresponding upstream vehicle images in at least one upstream trajectory; and determining the target upstream trajectory as the upstream driving trajectory of the vehicle to be tested.

[0105] In actual implementation, the upstream vehicle image is input into the feature extraction layer in the target detection model, and the corresponding upstream roof feature Q1 is output; then, the upstream roof feature of each upstream trajectory and the roof feature FV to be tested are calculated. v The first similarity Where ∥·∥ is the Euclidean norm of the feature vector, and FVv·Q1 is the dot product of the feature vector. Here, the first similarity condition is the maximum similarity. The path with the maximum similarity is taken as the upstream driving trajectory of the vehicle.

[0106] In some embodiments, the predicted vehicle features include predicted vehicle features and predicted roof features, the predicted vehicle face features are acquired in the same manner as the vehicle face features to be tested, and the predicted roof features are acquired in the same manner as the roof features to be tested; the method further includes: determining a second similarity between the predicted roof features and the roof features to be tested; if the second similarity meets a second similarity condition, determining a third similarity between the predicted vehicle face features and the vehicle face features to be tested; if the third similarity meets a third similarity condition, determining that the predicted vehicle features are consistent with the vehicle features to be tested.

[0107] In actual implementation, the upstream driving trajectory is combined with the detection point of the cloned vehicle, and the combined trajectory is input into the vehicle trajectory prediction model to output multiple predicted downstream waypoints. Multiple predicted downstream waypoints constitute the predicted downstream driving trajectory. In actual implementation, the travel time threshold T2 is set, and the downstream vehicle image captured by the downstream waypoint camera in the time period [t, t+T2] is extracted. The downstream vehicle image is segmented to obtain a first downstream image containing the vehicle face and a second downstream image containing the rest. The second downstream image is input into the target detection model to output the predicted vehicle face features and the predicted roof features Q2. Calculate the roof feature vector Q2 and the roof feature FV to be tested. v The second similarity A similarity ranking threshold is set, and downstream vehicle images that rank above the threshold are used as verification samples; the first downstream image corresponding to the verification sample is processed by the target detection model to obtain predicted vehicle face features, and the third similarity between the predicted vehicle face features and the vehicle face features to be tested is calculated; if the third similarity meets the third similarity condition, it is determined that the predicted vehicle features are consistent with the vehicle features to be tested. In some embodiments, the first downstream image corresponding to the verification sample is processed by the target detection model through feature extraction, feature fusion, detection head, and NMS, and a predicted vehicle face feature map is output; specifically, the vehicle brand logo, wear feature type, etc. in the predicted vehicle face feature map are detected to see if they are consistent with the FM v If the data with the same anchor frame position and attributes exists, it is marked as the same vehicle. Then, the license plate information marked as the same vehicle is checked to see if it is consistent; if not, it is identified as a fake license plate vehicle.

[0108] In some embodiments, the feature extraction layer includes at least one sub-extraction layer and a large separation convolutional attention LSKA module layer, and using the feature extraction layer to convert the first image into at least two first feature maps includes: using each sub-extraction layer and the large separation convolutional attention LSKA module layer to perform feature extraction processing on the first image in sequence to obtain at least two first feature maps.

[0109] For example, see Figure 4 , Figure 4 It is an optional structural diagram of the target detection model provided by the embodiment of the present application. The target detection model includes a feature extraction layer 401, a feature fusion layer 402 and a detection head 403. Here, the feature extraction layer 401 includes four sub-extraction layers 4011 connected in sequence and a large separation convolution attention LSKA module layer 4012. Among them, the sub-extraction layer 4011 is a RepViT structure. The feature extraction layer 402 includes an upsampling module, a c2f module, a fusion module and a convolution module.

[0110] In some embodiments, the at least two layers also include a feature fusion layer and a detection head, and the various layers of the target detection model are used to perform feature extraction on the first image to obtain the vehicle face features of the vehicle to be tested, including: using the feature extraction layer to convert the first image into at least two first feature maps; using the feature fusion layer to perform feature fusion processing on the at least two first feature maps to obtain at least one second feature map; using the detection head to perform target detection on the at least one second feature map to obtain the vehicle face features of the vehicle to be tested.

[0111] In some embodiments, the target detection model is also trained. Specifically, the method further includes: constructing a target detection model; obtaining a sample vehicle image and a sample vehicle feature; using the target detection model, extracting features from the sample vehicle image to obtain corresponding sample predicted vehicle features; and updating the model parameters of the target detection model based on the error between the sample vehicle feature and the sample predicted vehicle feature.

[0112] Here, the target detection data set is used to train the target detection model. Here, the target detection data set includes sample vehicle images and sample vehicle features. In one embodiment, the sample vehicle images include a sample first image and a sample second image, and the corresponding sample vehicle features include a sample vehicle face feature and a sample vehicle roof feature. In actual implementation, the vehicle face portion of the vehicle picture captured by the camera is cut to obtain, and the cut vehicle face portion of the picture is calibrated with the vehicle face wear characteristic state, vehicle brand logo and its name. The data set consists of the cut vehicle face portion of the picture, the vehicle face wear characteristic state, the vehicle brand logo, the vehicle brand name, and the vehicle license plate; the data set is used for training the improved target detection model.

[0113] Then, the improved target detection model is trained. In some embodiments, the method further includes: constructing a target detection model; obtaining a sample vehicle image and a sample vehicle feature; using the target detection model to extract features from the sample vehicle image to obtain corresponding sample predicted vehicle features; and updating the model parameters of the target detection model based on the error between the sample vehicle feature and the sample predicted vehicle feature. Specifically, see Figure 4 , build a target detection model; this model is an improved structure based on YOLO-V8, which includes three parts, namely feature extraction layer, feature fusion layer, and detection head; the improvements include the feature extraction layer structure and the detection head structure, and the feature fusion module still uses the c2f module in the original structure to splice feature maps of different layers and output the fused features. Specifically, the feature extraction layer structure is replaced with the RepViT network structure, that is, in the original lightweight convolutional neural network architecture MobileNetV3, the original block structure composed of 1x1 dilated convolution, deep convolution, and 1x1 projection layers is transformed to separate the mixed spatial information token mixer (and the mixed channel information channel mixer, forming a feature extraction network with a total of 4 layers; the LSKA module is added before the last pooling layer in the SPPF module; the original MobileNetV3 structure consists of 1x1 dilated convolution, deep convolution, and 1x1 projection layers. The 1x1 dilated convolution layer and the 1x1 projection layer can realize the interaction between channels, and the deep convolution helps to fuse spatial information; the former and the latter correspond to the channel mixer and the token mixer, respectively. The deep convolution is moved up to separate the two. At the same time, structural reparameterization is adopted to introduce a multi-branch topology structure for the deep convolution during training to improve performance. During the inference process, the multi-branch structure of the deep convolution can be merged into a single-branch structure to eliminate the additional computational and memory costs brought by multiple branches.

[0114] In order to improve the multi-scale fusion capability of the SPPF module, the LSKA module is added before the last pooling layer. The LSKA attention mechanism captures the extensive contextual information of the image by utilizing large and separable convolution kernels and spatial dilation convolutions, generates an attention map, and weights the original features through this attention map, thereby enhancing the network’s attention to important features and improving the performance of the model.

[0115] In some embodiments, see Figure 5 , Figure 5 It is an optional structural diagram of the detection head provided in an embodiment of the present application. The detection head includes a first convolutional layer 501, a shared convolutional layer 502, a second convolutional layer 503 and a loss function layer 504 connected in sequence, the first convolutional layer 501 includes at least two first sub-convolutional layers respectively connected to the shared convolutional layer 502, the second convolutional layer 503 includes at least two second sub-convolutional layers respectively connected to the shared convolutional layer 502, and the loss function layer 504 includes sub-loss function layers respectively connected one-to-one to the at least two second sub-convolutional layers.

[0116] In actual implementation, see Figure 5 , Figure 5It is an optional structural diagram of the detection head provided in the embodiment of the present application. The detection head adopts the LSCD lightweight shared convolution module to replace the convolution module in the original YOLO-V8 structure, reducing the number of detection head parameters and thus reducing the model size. Here, the second subconvolution layer is a two-dimensional convolution layer.

[0117] In the embodiment of the present application, the decoupling head used in YOLO-V8 can reduce the convergence speed of the model, but the result is a sudden increase in the number of parameters. The number of parameters of the detection head part of the entire network accounts for 40% of the entire network; by replacing the original detection head with LSCD and replacing the original convolution module with a lightweight shared convolution module, the number of parameters of the detection head can be significantly reduced because the shared convolution can share parameters. The two sub-loss function layers can be regression loss and classification loss respectively.

[0118] See also Figure 4 ,feature Figure 1-5 All are first feature maps, features Figure 1-5 After feature fusion processing, feature maps 11, 14 and 17 are obtained as the second feature maps. When the image data enters the target detection model, it first passes through a four-layer feature extraction network to convert the image features into feature maps, and obtains feature Figure 1 ,feature Figure 2 ,feature Figure 3 ,feature Figure 4 , after a SPPF module to obtain the characteristics Figure 5 , the purpose is to improve the receptive field and target detection accuracy; after an upsampling stage, feature map 6 is obtained, the purpose is to be able to detect at different scales, while helping to detect objects of different sizes and improving detection accuracy. Figure 3 The result of upsampling is fused to feature map 7; after passing through the c2f module, feature map 8 is obtained. The c2f module can increase the nonlinear ability and representation ability of the network, thereby improving the network's ability to model complex data. Feature map 8 is upsampled once to obtain feature map 9, which is combined with feature map 9. Figure 2 After fusion, feature map 10 is obtained; after one convolution, feature map 11 is obtained by fusing feature map 8 with feature map 13; after one convolution, feature map 14 is obtained by fusing feature map 8 with feature map 14; after one convolution, feature map 14 is obtained by fusing feature map 8 with feature map 8 with feature map 13. Figure 5 The feature map 16 is obtained by fusion, and the feature map 17 is obtained by the c2f module. Finally, the feature maps 11, 14 and 17 are used in the head detection stage.

[0119] Next, the loss function in YOLO-V8 is used, namely the classification loss VFL Loss, the regression loss CIOU Loss+DFL, and the hyperparameter Reg_max defaults to 16; the training data set is the target detection car face data set; the dynamic allocation strategy based on task allocation learning TAL is adopted to select high-quality positive samples, and the model is trained based on the loss function; the feature maps output by the three detection heads are processed by non-maximum suppression (NMS) to form the optimal result.

[0120] The embodiment of the present application adopts a deep learning algorithm to analyze, screen and process the vehicle passing pictures and attribute information obtained by the highway roadside or gantry cameras. It mainly improves the backbone network of the target detection model to make the entire network more capable of extracting image features and the reasoning stage faster; improves the detection head part of the target detection model, and uses shared convolution to greatly reduce the number of parameters, thereby reducing the training cost of the model. At the same time, the complete information of the passing car picture is input into the model, and the features of the vehicle part and the roof part are separately identified. The feature similarity detection is mainly based on the roof features and supplemented by the vehicle face features, thereby improving the accuracy of the detection of fake license plate vehicles. In the process of fake license plate vehicle detection, the number of samples for feature similarity verification is reduced through trajectory prediction to avoid wasting resources and computing power by calling too much camera data.

[0121] The following further describes the exemplary structure of the fake license plate vehicle detection device 2055 provided in the embodiment of the present application implemented as a software module. In some embodiments, for example Figure 2 As shown, the software modules stored in the fake license plate vehicle detection device 2055 of the memory 205 may include:

[0122] The first obtaining module 20551 is used to obtain a vehicle image of the vehicle to be tested collected by a first camera in the road network;

[0123] The first feature extraction module 20552 is used to extract features from the vehicle image using a target detection model to obtain the vehicle features of the vehicle to be tested;

[0124] The first determination module 20553 is used to determine the upstream driving track that the vehicle to be tested has traveled;

[0125] A prediction module 20554 is used to predict the downstream driving trajectory of the vehicle to be tested based on the upstream driving trajectory to obtain a predicted downstream driving trajectory;

[0126] The second determination module 20555 is used to determine the predicted time period of the vehicle to be tested passing through the predicted downstream driving trajectory;

[0127] The second obtaining module 20556 is used to obtain the downstream vehicle image captured by the second camera in the predicted downstream driving trajectory within the predicted time period;

[0128] A second feature extraction module 20557 is used to extract features from the downstream vehicle image using the target detection model to obtain corresponding predicted vehicle features;

[0129] The third determination module 20558 is used to determine that the vehicle to be tested is a fake license plate vehicle if the predicted vehicle feature is consistent with the feature of the vehicle to be tested and the license plate information of the vehicle to be tested is inconsistent with that of the vehicle in the downstream image.

[0130] In some embodiments, the target detection model includes at least two layers, and the at least two layers include a feature extraction layer. The first feature extraction module 20552 is also used to perform image segmentation on the vehicle image to obtain a first image and a second image, wherein the first image includes the face of the vehicle to be tested; using the various layers of the target detection model, feature extraction is performed on the first image to obtain the face features of the vehicle to be tested; using the feature extraction layer of the target detection model, feature extraction is performed on the second image to obtain the roof features of the vehicle to be tested.

[0131] In some embodiments, the first determination module 20553 is also used to extract at least one upstream trajectory of the vehicle to be tested from the road network; determine a target time period for the vehicle to be tested to pass through the upstream driving trajectory; obtain an upstream vehicle image captured by a camera set in the at least one upstream trajectory within the target time period; respectively use the feature extraction layer of the target detection model to extract features from each of the upstream vehicle images to obtain corresponding upstream roof features; based on the corresponding upstream vehicle image in at least one upstream trajectory, determine a target upstream trajectory in which the first similarity between the upstream roof feature and the roof feature to be tested meets a first similarity condition; and determine the target upstream trajectory as the upstream driving trajectory of the vehicle to be tested.

[0132] In some embodiments, the predicted vehicle features include predicted vehicle features and predicted roof features, the predicted vehicle face features are acquired in the same manner as the vehicle face features to be tested, and the predicted roof features are acquired in the same manner as the roof features to be tested; the device also includes: a comparison module for determining a second similarity between the predicted roof features and the roof features to be tested; if the second similarity meets a second similarity condition, determining a third similarity between the predicted vehicle face features and the vehicle face features to be tested; if the third similarity meets a third similarity condition, determining that the predicted vehicle features are consistent with the vehicle features to be tested.

[0133] In some embodiments, the at least two layers also include a feature fusion layer and a detection head, and the various layers of the target detection model are used to perform feature extraction on the first image to obtain the vehicle face features of the vehicle to be tested, including: using the feature extraction layer to convert the first image into at least two first feature maps; using the feature fusion layer to perform feature fusion processing on the at least two first feature maps to obtain at least one second feature map; using the detection head to perform target detection on the at least one second feature map to obtain the vehicle face features of the vehicle to be tested.

[0134] In some embodiments, the feature extraction layer includes at least one sub-extraction layer and a large separation convolutional attention LSKA module layer, and the first feature extraction module 20552 is also used to use each sub-extraction layer and the large separation convolutional attention LSKA module layer to perform feature extraction processing on the first image in sequence to obtain at least two first feature maps.

[0135] In some embodiments, the detection head includes a first convolutional layer, a shared convolutional layer, a second convolutional layer and a loss function layer connected in sequence, the first convolutional layer includes at least two first sub-convolutional layers respectively connected to the shared convolutional layer, the second convolutional layer includes at least two second sub-convolutional layers respectively connected to the shared convolutional layer, and the loss function layer includes sub-loss function layers respectively connected one-to-one to the at least two second sub-convolutional layers.

[0136] In some embodiments, the device also includes: a model training module for constructing a target detection model; obtaining sample vehicle images and sample vehicle features; using the target detection model to extract features from the sample vehicle images to obtain corresponding sample predicted vehicle features; and updating the model parameters of the target detection model based on the error between the sample vehicle features and the sample predicted vehicle features.

[0137] It should be noted that the description of the device of the embodiment of the present application is similar to the description of the above-mentioned method embodiment, and has similar beneficial effects as the method embodiment, so it will not be repeated.

[0138] The embodiment of the present application provides a computer program product, which includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above-mentioned cloned vehicle detection method of the embodiment of the present application.

[0139] An embodiment of the present application provides a computer-readable storage medium storing executable instructions, wherein executable instructions are stored. When the executable instructions are executed by a processor, the processor will be caused to execute the cloned vehicle detection method provided in the embodiment of the present application.

[0140] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface storage, optical disk, or CD-ROM; or it may be various devices including one or any combination of the above memories.

[0141] In some embodiments, executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0142] As an example, executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions).

[0143] By way of example, executable instructions may be deployed to be executed on one computing device, or on multiple computing devices located at one site, or on multiple computing devices distributed across multiple sites and interconnected by a communication network.

[0144] In summary, the embodiments of the present application can reduce the amount of data processing, ensure detection accuracy and save computing resources.

[0145] The above is only an embodiment of the present application and is not intended to limit the protection scope of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. A method for detecting fake license plate vehicles, characterized in that: include: Obtaining a vehicle image of the vehicle to be tested captured by a first camera in the road network; Using the target detection model, extracting features from the vehicle image to obtain the vehicle features of the vehicle to be tested; Determining the upstream driving trajectory that the vehicle to be tested has traveled; Based on the upstream driving trajectory, predicting the downstream driving trajectory of the vehicle to be tested to obtain a predicted downstream driving trajectory; Determining a predicted time period for the vehicle to be tested to pass through the predicted downstream driving trajectory; Obtaining a downstream vehicle image captured by a second camera in the predicted downstream driving trajectory within a predicted time period; Using the target detection model, extracting features from the downstream vehicle image to obtain corresponding predicted vehicle features; If the predicted vehicle feature is consistent with the feature of the vehicle to be tested, and the license plate information of the vehicle to be tested is inconsistent with that of the vehicle in the downstream image, it is determined that the vehicle to be tested is a fake license plate vehicle.

2. The method according to claim 1, characterized in that The target detection model includes at least two layers, and the at least two layers include a feature extraction layer. The target detection model is used to extract features of the vehicle image to obtain the vehicle features to be tested, including: Performing image segmentation on the vehicle image to obtain a first image and a second image, wherein the first image includes a vehicle face of the vehicle to be tested; Using each layer of the target detection model, extracting features from the first image to obtain the face features of the vehicle to be tested; The feature extraction layer of the target detection model is used to extract features from the second image to obtain features of the roof of the vehicle to be tested.

3. The method according to claim 2, characterized in that The step of determining the upstream driving trajectory of the vehicle to be tested includes: Extracting at least one upstream trajectory of the vehicle to be tested from the road network; Determine a target time period for the vehicle to be tested to pass through the upstream driving trajectory; Obtaining an upstream vehicle image captured by a camera arranged in the at least one upstream track within the target time period; Using the feature extraction layer of the target detection model respectively, extract features from each of the upstream vehicle images to obtain corresponding upstream vehicle roof features; Determine, based on the upstream vehicle image corresponding to at least one upstream trajectory, a target upstream trajectory where a first similarity between an upstream vehicle roof feature and the vehicle roof feature to be measured reaches a first similarity condition; The target upstream trajectory is determined as the upstream driving trajectory of the vehicle to be tested.

4. The method according to claim 2, characterized in that: The predicted vehicle features include predicted vehicle features and predicted roof features, the predicted vehicle face features are acquired in the same manner as the vehicle face features to be tested, and the predicted roof features are acquired in the same manner as the vehicle roof features to be tested; the method further includes: Determining a second similarity between the predicted vehicle roof feature and the vehicle roof feature to be measured; If the second similarity meets the second similarity condition, determining a third similarity between the predicted vehicle face feature and the vehicle face feature to be tested; If the third similarity meets the third similarity condition, it is determined that the predicted vehicle feature is consistent with the feature of the vehicle to be tested.

5. The method according to claim 2, characterized in that: The at least two layers further include a feature fusion layer and a detection head. The various layers of the target detection model are used to extract features from the first image to obtain the face features of the vehicle to be tested, including: Using the feature extraction layer, converting the first image into at least two first feature maps; Using the feature fusion layer to perform feature fusion processing on the at least two first feature maps to obtain at least one second feature map; The detection head is used to perform target detection on the at least one second feature map to obtain the vehicle face features of the vehicle to be tested.

6. The method according to claim 5, characterized in that The feature extraction layer includes at least one sub-extraction layer and a large separation convolution attention LSKA module layer, and the first image is converted into at least two first feature maps by using the feature extraction layer, including: The first image is processed by performing feature extraction in sequence using each sub-extraction layer and a large separation convolution attention LSKA module layer to obtain at least two first feature maps.

7. The method according to claim 5, characterized in that The detection head includes a first convolutional layer, a shared convolutional layer, a second convolutional layer and a loss function layer connected in sequence, the first convolutional layer includes at least two first sub-convolutional layers respectively connected to the shared convolutional layer, the second convolutional layer includes at least two second sub-convolutional layers respectively connected to the shared convolutional layer, and the loss function layer includes sub-loss function layers respectively connected one-to-one to the at least two second sub-convolutional layers.

8. The method according to claim 1, characterized in that The method further comprises: Build an object detection model; Obtaining a sample vehicle image and sample vehicle features; Using the target detection model, extracting features from the sample vehicle image to obtain corresponding sample predicted vehicle features; Based on the error between the sample vehicle feature and the sample predicted vehicle feature, the model parameters of the target detection model are updated.

9. A device for detecting fake license plate vehicles, characterized in that: include: A first acquisition module is used to obtain a vehicle image of the vehicle to be tested collected by a first camera in the road network; A first feature extraction module is used to extract features from the vehicle image using a target detection model to obtain the vehicle features of the vehicle to be tested; A first determination module is used to determine the upstream driving track that the vehicle to be tested has traveled; A prediction module, used for predicting the downstream driving trajectory of the vehicle to be tested based on the upstream driving trajectory to obtain a predicted downstream driving trajectory; A second determination module is used to determine a predicted time period for the vehicle to be tested to pass through the predicted downstream driving trajectory; A second acquisition module is used to obtain a downstream vehicle image captured by a second camera in the predicted downstream driving trajectory within a predicted time period; A second feature extraction module, used to extract features from the downstream vehicle image using the target detection model to obtain corresponding predicted vehicle features; The third determination module is used to determine that the vehicle to be tested is a fake license plate vehicle if the predicted vehicle feature is consistent with the feature of the vehicle to be tested and the license plate information of the vehicle to be tested is inconsistent with that of the vehicle in the downstream image.

10. An electronic device, characterized in that: include: A memory for storing executable instructions; The processor is used to implement the method for detecting cloned vehicles as described in any one of claims 1 to 8 when executing the executable instructions stored in the memory.

Images