Processing method, device and equipment of vehicle information recognition model

Abstract

The application provides a processing method, device and equipment of a vehicle information recognition model, and is used for improving the recognition accuracy of the vehicle information recognition model. The method comprises the following steps: acquiring a vehicle image collected by a camera at a preset location and performing vehicle information recognition, wherein the preset location is a location in a target site where a vehicle management device is arranged; when there is an abnormal vehicle image with abnormal vehicle information recognition, uploading the abnormal vehicle image to a server, so that the server trains an initial neural network model according to the abnormal vehicle image after labeling of target vehicle information added by a user is performed on the abnormal vehicle image, and uses the trained model as a vehicle information recognition model, wherein the vehicle information recognition model is used for recognizing vehicle information of a vehicle corresponding to the target site; and receiving the vehicle information recognition model issued by the server.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of vehicles, in particular to a processing method, device and equipment of a vehicle information recognition model. BACKGROUND

[0002] For parking lots, parks and other places involving a large number of vehicle parking, nowadays, management personnel often introduce automatic systems to realize automatic management of parking management, charging management, etc.

[0003] Among them, cameras are configured at the entrance and exit to collect image data of vehicles, and then the vehicles can be identified, the license plates can be identified, and the data can be retained.

[0004] In the research process of the existing related technology, the inventors found that when the existing vehicle management system is applied to a specific site, the vehicle information recognition accuracy may be unstable, for example, in A parking lot, the normal vehicle information recognition effect can be achieved, while in B parking lot, the normal vehicle information recognition effect cannot be achieved; for example, in C parking lot, the normal vehicle information recognition effect cannot be achieved, while in D park, the normal vehicle information recognition effect can be achieved. SUMMARY

[0005] The present application provides a processing method, device and equipment of a vehicle information recognition model, which is used to improve the vehicle state recognition accuracy of the vehicle information recognition model, so as to ensure the normal work of the vehicle management equipment.

[0006] In a first aspect, the present application provides a processing method of a vehicle information recognition model, which comprises:

[0007] Obtaining vehicle images collected by a camera at a preset site and performing vehicle information recognition on the vehicle images, wherein the preset site is a site in a target site where a vehicle management equipment is configured;

[0008] When there is an abnormal vehicle image with abnormal vehicle information recognition, uploading the abnormal vehicle image to a server, so that the server trains an initial neural network model according to the abnormal vehicle image after labeling the target vehicle information added by a user on the abnormal vehicle image, and uses the trained model as a vehicle information recognition model, wherein the vehicle information recognition model is used to identify the vehicle information of the vehicles corresponding to the target site;

[0009] Receiving the vehicle information recognition model issued by the server.

[0010] In a second aspect, the present application provides another processing method of a vehicle information recognition model, which comprises:

[0011] receive an abnormal vehicle image uploaded by the vehicle management device, wherein the vehicle management device is configured to acquire a vehicle image collected by a camera at a preset location in a target site and perform vehicle information recognition on the vehicle image, the preset location is a location in the target site where the vehicle management device is deployed, and the abnormal vehicle image is a vehicle image with abnormal vehicle information recognition;

[0012] train an initial neural network model according to the abnormal vehicle image after the abnormal vehicle image is assigned with target vehicle information added by a user, and use the trained model as a vehicle information recognition model, wherein the vehicle information recognition model is configured to recognize vehicle information of a vehicle in the target site;

[0013] issue the vehicle information recognition model to the vehicle management device.

[0014] In a third aspect, the present application provides a processing device for a vehicle information recognition model, which comprises:

[0015] an acquisition unit configured to acquire a vehicle image collected by a camera at a preset location, wherein the preset location is a location in a target site where a vehicle management device is deployed;

[0016] an identification unit configured to perform vehicle information recognition on the vehicle image;

[0017] an uploading unit configured to upload an abnormal vehicle image to a server when the abnormal vehicle image has abnormal vehicle information recognition, so that the server trains an initial neural network model according to the abnormal vehicle image after the abnormal vehicle image is assigned with target vehicle information added by a user, and uses the trained model as a vehicle information recognition model, wherein the vehicle information recognition model is configured to recognize vehicle information of a vehicle in the target site;

[0018] a receiving unit configured to receive a vehicle information recognition model issued by the server.

[0019] In a fourth aspect, the present application provides another processing device for a vehicle information recognition model, which comprises:

[0020] a receiving unit configured to receive an abnormal vehicle image uploaded by a vehicle management device, wherein the vehicle management device is configured to acquire a vehicle image collected by a camera at a preset location and perform vehicle information recognition on the vehicle image, the preset location is a location in a target site where the vehicle management device is deployed, and the abnormal vehicle image is a vehicle image with abnormal vehicle information recognition;

[0021] a training unit configured to train an initial neural network model according to the abnormal vehicle image after the abnormal vehicle image is assigned with target vehicle information added by a user, and use the trained model as a vehicle information recognition model, wherein the vehicle information recognition model is configured to recognize vehicle information of a vehicle in the target site;

[0022] The issuing unit is configured to issue the vehicle information recognition model to the vehicle management device.

[0023] In a fifth aspect, the present application further provides a processing device for a vehicle information recognition model, comprising a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program in the memory to implement the method in the first aspect, any implementation manner of the first aspect, the second aspect, or any implementation manner of the second aspect.

[0024] In a sixth aspect, the present application further provides a computer readable storage medium, which stores a plurality of instructions, and the instructions are adapted to be loaded by a processor to execute the method in the first aspect, any implementation manner of the first aspect, the second aspect, or any implementation manner of the second aspect.

[0025] From the above, the present application has the following beneficial effects:

[0026] The present application configures a back transmission mechanism in the parking lot. If the vehicle management device deployed in the parking lot detects a vehicle image with vehicle information recognition exception, the abnormal vehicle image is transmitted back to the cloud server. The cloud server trains the vehicle information recognition model based on the abnormal vehicle images transmitted back, and then issues the trained vehicle information recognition model to the vehicle management device deployed in the parking lot. Since the vehicles and their vehicle information of the fixed parking lot are usually highly similar, the vehicle information recognition model is trained based on the abnormal vehicle images with vehicle information recognition exception in the parking lot. Therefore, the vehicle information recognition model has strong pertinence to the vehicles and their vehicle information in the parking lot, and the vehicle information recognition accuracy of the vehicle information recognition model in the parking lot can be improved to a certain extent. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 FIG. 1 is a scenario schematic diagram of the processing method of the vehicle information recognition model according to the present application;

[0028] Figure 2 FIG. 2 is a flowchart of the processing method of the vehicle information recognition model according to the present application;

[0029] Figure 3 FIG. 3 is another flowchart of the processing method of the vehicle information recognition model according to the present application;

[0030] Figure 4 FIG. 4 is a flowchart of the vehicle information recognition method for a vehicle image according to the present application;

[0031] Figure 5 FIG. 5 is another flowchart of the vehicle information recognition method for a vehicle image according to the present application;

[0032] Figure 6 This is another flowchart illustrating vehicle information recognition from vehicle images according to this application;

[0033] Figure 7 This is another flowchart illustrating vehicle information recognition from vehicle images according to this application;

[0034] Figure 8 This is another flowchart illustrating vehicle information recognition from vehicle images according to this application;

[0035] Figure 9 This is another flowchart illustrating vehicle information recognition from vehicle images according to this application;

[0036] Figure 10 This is another flowchart illustrating vehicle information recognition from vehicle images according to this application;

[0037] Figure 11 This is a schematic diagram of a processing device for the vehicle information recognition model of this application;

[0038] Figure 12 This is another structural schematic diagram of the processing device for the vehicle information recognition model of this application;

[0039] Figure 13 This is a schematic diagram of the processing device for the vehicle information recognition model of this application. Detailed Implementation

[0040] First, before introducing this application, let me first introduce the relevant content regarding the application background.

[0041] The vehicle information recognition model processing method, apparatus, and computer-readable storage medium provided in this application can be applied to vehicle information recognition model processing equipment, such as vehicle management equipment deployed in a specific parking lot, or on a server, to improve the vehicle information recognition accuracy of the vehicle information recognition model, thereby ensuring the normal operation of the vehicle management equipment.

[0042] The vehicle information recognition model processing method mentioned in this application can be executed by a vehicle information recognition model processing device, or a vehicle management device or server device that integrates the device. The device can be implemented in hardware or software.

[0043] The vehicle management equipment mentioned in this application can be applied to different types of parking lots such as commercial parking lots (paid parking lots), park parking lots, and company parking lots. It can be used for automated management of parking lots, such as vehicle entry and exit gate control, parking management, fee management, and vehicle monitoring.

[0044] by Figure 1A scene schematic diagram of the processing method of the vehicle information identification model of the present application is shown by way of example. The vehicle management device 101 can be deployed in the parking lot of a logistics company park for running a vehicle management system. The vehicle management system can be configured with cameras at various locations in the parking lot of the park, such as the entrance and exit, to collect vehicle images entering and exiting the parking lot of the park. When an abnormal vehicle image with abnormal vehicle information identification is identified, the processing method of the vehicle information identification model provided by the present application can be triggered. The abnormal vehicle image is uploaded to the cloud server 103 through the network 102. The server 103 trains the vehicle information identification model according to the abnormal vehicle image uploaded by the vehicle management device 101. The vehicle information identification model is configured to the vehicle management device 101, so that the vehicle information identification model has strong pertinence to the vehicles and their vehicle information in the parking lot of the park, and the vehicle information identification accuracy of the vehicle information identification model in the parking lot of the park can be improved to a certain extent.

[0045] Among them, the vehicle image for identifying vehicle information is generally multiple images, such as dynamic images or videos, so as to determine the vehicle information from a dynamic perspective. Of course, for individual vehicle information identification methods, only a single image or a single picture can be used.

[0046] In the present application, the vehicle management device can be understood as a local workstation configured in a parking lot, which can typically be a desktop computer. The desktop computer can provide data processing functions and human-computer interaction functions for staff to view vehicle images (monitoring images) collected by cameras or input relevant operation instructions. After the local desktop computer, notebook computer and other devices are configured with the application program of the vehicle management device involved in the processing method of the vehicle information identification model, they can be used as the vehicle management device in the present application (one of the vehicle information identification model processing devices provided by the present application). The vehicle management device itself can be further extended with a camera for collecting vehicle images, or other devices such as a card reader, a gate, an indicator light, etc., to realize automatic management of vehicle access control, parking management, toll management, vehicle monitoring, etc. in the parking lot.

[0047] Next, the processing method of the vehicle information identification model provided by the present application will be introduced.

[0048] Firstly, refer to Figure 2 , Figure 2 is a flowchart of the processing method of the vehicle information identification model of the present application shown from the vehicle management device side of the local side. The processing method of the vehicle information identification model provided by the present application can specifically include the following steps:

[0049] Step S201: Acquire vehicle images captured by the camera at a preset location and perform vehicle information recognition on the vehicle images, wherein the preset location is a location in the target site where the vehicle management equipment is configured.

[0050] Step S202: When there is an abnormal vehicle image with abnormal vehicle information recognition, the abnormal vehicle image is uploaded to the server so that after the server assigns the target vehicle information added by the user to the abnormal vehicle image, it trains an initial neural network model based on the abnormal vehicle image and uses the trained model as the vehicle information recognition model. The vehicle information recognition model is used to identify the vehicle information of the vehicle corresponding to the target site.

[0051] Step S203: Receive the vehicle information recognition model sent by the server.

[0052] Secondly, see Figure 3 , Figure 3 This is another flowchart illustrating the processing method of the vehicle information recognition model of this application, shown from the cloud-side server side. The processing method of the vehicle information recognition model provided by this application may also include the following steps:

[0053] Step S301: Receive abnormal vehicle images uploaded by vehicle management equipment. The vehicle management equipment is used to acquire vehicle images captured by a camera at a preset location and perform vehicle information recognition on the vehicle images. The preset location is a location in the target site where the vehicle management equipment is configured. The abnormal vehicle images are vehicle images with abnormal vehicle information recognition.

[0054] Step S302: After assigning the target vehicle information added by the user to the abnormal vehicle image, an initial neural network model is trained based on the abnormal vehicle image, and the trained model is used as the vehicle information recognition model. The vehicle information recognition model is used to identify the vehicle information of the vehicle corresponding to the target site.

[0055] Step S303: Send the vehicle information recognition model to the vehicle management equipment.

[0056] From the above Figure 2 as well as Figure 3As can be seen from the embodiments shown, this application configures a feedback mechanism on-site. If the vehicle management equipment deployed in the parking lot detects a vehicle image with abnormal vehicle information recognition, the abnormal vehicle image is sent back to the cloud server. The cloud server then trains a vehicle information recognition model on these returned abnormal vehicle images and distributes the trained vehicle information recognition model back to the vehicle management equipment deployed in the parking lot. Since vehicles entering and exiting a fixed area and their vehicle information usually have a high degree of similarity, the current vehicle information model is trained from abnormal vehicle images with abnormal vehicle information recognition at that area. Therefore, it is highly targeted to the vehicles and their vehicle information at that area, and can improve the vehicle information recognition accuracy of the vehicle information recognition model at that area to a certain extent.

[0057] The process of the vehicle information recognition model in this application will be further described, including the steps involved and possible implementation methods in practical applications.

[0058] In this application, the target site for the vehicle management equipment can be the parking lot mentioned above, such as commercial parking lots (paid parking lots), park parking lots, company parking lots, and other different types of parking lots.

[0059] Cameras are deployed at multiple locations within the parking lot to capture vehicle images. Typically, cameras can be deployed at the entrances and exits of the parking lot, or at locations within the parking lot that vehicles frequently pass through or where there is a wide field of view, which meet the location selection requirements.

[0060] A camera can be understood not only as a standalone camera, but also as a camera included in the components of certain devices, which can capture images of vehicles at preset locations in parking lots.

[0061] It is understood that, in this application, acquiring vehicle images captured by a camera at a preset location can be understood not only as capturing vehicle images in real time at a preset location via a camera, but also as retrieving vehicle images from a storage device that stores those vehicle images after the deployed camera has captured them at the preset location.

[0062] Once the vehicle image is obtained, vehicle status recognition can be performed within the image.

[0063] For example, vehicle information in images can be identified using artificial intelligence (AI), specifically through a neural network model. This neural network model is trained on a large number of different vehicle images. Specifically, a large number of vehicle images can be collected, and vehicle information can be labeled in these images as a training set. The images in the training set are then sequentially input into the activated neural network model, allowing the model to perform forward propagation, identify vehicle information in the images, and calculate the loss function based on the output recognition results. Backpropagation is then performed to optimize the model parameters. When the training requirements such as recognition accuracy, number of training iterations, or training time are met, the model training is complete. At this point, the model can be called a vehicle information recognition model and deployed in various parking lots to serve the vehicle management equipment and vehicle management system of the parking lots.

[0064] In this application, to distinguish it from the vehicle information recognition model mentioned below, the vehicle information recognition model used in step S201 to perform vehicle information recognition on the vehicle image can be referred to as the initial vehicle information recognition model. This initial vehicle information recognition model is generally a general model used by various parking lots, or it can be a vehicle information recognition model obtained by updating the general model to a specific parking lot in a "one-to-one" manner through subsequent data processing in this application after it has been configured to a specific parking lot.

[0065] In practical applications, it is understandable that vehicle information recognition models cannot always accurately identify vehicle information. Due to factors such as ambient light, shadows, lens smudges, aging equipment components, and signal interference, environmental, hardware, and software conditions can all affect images, thereby impacting the accuracy of vehicle information recognition models. Therefore, in some vehicle images, vehicle information may not be recognized or the recognition effect may be poor. The development process of vehicle information recognition models can be understood as improving their vehicle information recognition accuracy as much as possible.

[0066] In this application, vehicle images with abnormal vehicle information recognition can be directly considered, and these images can be used as abnormal vehicle images to update or optimize the vehicle information recognition model, so as to provide a "one-to-one" update service for the local vehicle information recognition scenario in the parking lot.

[0067] Specifically, when a vehicle management device detects abnormal vehicle images with anomalous vehicle information recognition, it can upload these images to a server via the network. After receiving the abnormal vehicle images uploaded by the vehicle management device, the server can retrain a vehicle information recognition model for the parking lot corresponding to the vehicle management device based on these images.

[0068] During this training process, staff can annotate vehicle information on these abnormal vehicle images. Secondly, the server can obtain the vehicle information recognition model currently used by the vehicle management device, and use this model to train and update itself based on these abnormal vehicle images; alternatively, the server can obtain an initialized neural network model and train it solely based on these abnormal vehicle images; or, the server can obtain any neural network model and train it based on these abnormal vehicle images.

[0069] For different types of abnormal vehicle images, the server can also perform uniformization processing, which is understandable. In real-world environments, there are many abnormal vehicle images showing no vehicles, resting, or working, but fewer showing vehicles arriving or leaving their posts. This can lead to uneven data distribution, and training with unevenly distributed data will result in unsatisfactory recognition performance of the final model. Therefore, data uniformization can be performed by obtaining the class with the least amount of data, then examining the distribution of data in other classes, and equalizing the data in each class. Data from classes with larger amounts of data is removed and not included in the training process.

[0070] During training, the server can automatically allocate Graphics Processing Units (GPUs). For training tasks involving different abnormal vehicle images, the GPU resource usage during training is analyzed by setting different batch sizes, resulting in a linear GPU utilization rate. all =G base +batch_size*G batch When multiple tasks are available, GPU resources are dynamically distributed to new training tasks based on the known batch size and tasks to avoid training failure due to memory overflow. Once the allocated GPU resources are determined, training tasks can be performed via a web page with one-click delivery.

[0071] After the server completes the training of the model and obtains the new vehicle information recognition model, it can distribute it to the vehicle management equipment at the specific parking lot, that is, to the vehicle management equipment that originally uploaded the abnormal vehicle images used to train the model. After receiving the distributed new vehicle information recognition model, the vehicle management equipment can configure itself to perform vehicle information recognition on the new vehicle images using the new vehicle information recognition model.

[0072] For example, when vehicle management device X uploads images of abnormal vehicles, it can configure a device identifier X for the images. The device identifier X can indicate that these images were uploaded from vehicle management device X. Secondly, when the server is training the model, it can also distinguish the model based on the device identifier X. The trained model can also be configured with the device identifier X to identify the "one-to-one" correspondence between the vehicle information recognition model and vehicle management device X. This is more suitable for vehicle information recognition in the parking lot where vehicle management device X is located.

[0073] As can be seen from the above, the vehicle information recognition model proposed in this application can be configured with a one-to-one vehicle information recognition model for each parking lot. Since the vehicles entering and exiting fixed parking lots and their vehicle information usually have a high degree of similarity, and the vehicle information recognition model is trained from abnormal vehicle images of the parking lot, it has strong targeting for the vehicles and their vehicle information in the parking lot, and can improve the vehicle information recognition accuracy of the vehicle information recognition model in the parking lot to a certain extent.

[0074] Secondly, in practical applications, the configuration scenarios of this vehicle information recognition model can have further advantages in other aspects.

[0075] For example, as mentioned above, when the server is training the vehicle information recognition model, staff can annotate the vehicle information on abnormal vehicle images uploaded by the vehicle management equipment. In practical applications, this annotation process involves little or no algorithm content, so the requirements for staff annotation are very low. Staff can be non-algorithm personnel, non-programmers, or even non-technical personnel. They only need to use annotation tools in remote work scenarios (such as web-based operations) to annotate the vehicle information on abnormal vehicle images distributed by the server. Obviously, this training method not only avoids the large amount of manpower required to configure the vehicle information recognition model for parking lots, but also distributes the model training workload into different parts. For example, it can be further distributed to non-technical personnel to perform vehicle information annotation operations, which is very beneficial for the maintenance of the vehicle information recognition model and can significantly reduce manpower costs.

[0076] For example, the configuration scenario of this vehicle information recognition model is not only about training the vehicle information recognition model in the cloud, but more importantly, it proposes a new business model for vehicle information recognition models. After the vehicle information recognition model is configured in a specific parking lot, the model can be updated and maintained remotely in a "one-to-one" manner, achieving an automated iterative update and deployment model business model. Compared with the traditional business model where the vehicle information recognition model is configured in a specific parking lot and then requires staff to be dispatched to the site for model update and maintenance services, the new business model proposed in this application has high practicality and application value due to its significantly reduced labor costs and significantly improved update and maintenance efficiency.

[0077] Next, we will introduce the possible implementation methods of this application in practical applications to identify abnormal vehicle images.

[0078] Specifically, in this application, vehicle information recognition may include vehicle status information recognition, license plate information recognition, and cargo compartment loading information recognition, as detailed below:

[0079] I. Vehicle Status Information Recognition

[0080] Vehicle status information recognition compares the identified vehicle status information with preset standard vehicle status information, such as arriving at work, loading, unloading, resting, and leaving work. If they do not match, the current vehicle image is confirmed as an abnormal vehicle image. Furthermore, other information can be combined to determine whether the vehicle status information recognition result is normal.

[0081] In one exemplary implementation, see [reference] Figure 4 The diagram illustrates a process for vehicle information recognition from vehicle images, as described in this application. Specifically, vehicle information recognition from vehicle images may include the following steps:

[0082] Step S401: Obtain vehicle status information sent by the vehicle, user terminal or vehicle sensing device, wherein the vehicle status information is used to indicate the current motion state or preset motion state of the vehicle.

[0083] It is understandable that in practical applications, vehicles or users in vehicles can proactively inform the vehicle management equipment of their vehicle status. For example, the vehicle management equipment can be configured with receiving devices in some locations in the parking lot, such as paid parking spaces, unloading checkpoints, and loading checkpoints. When driving a vehicle, the user can operate the vehicle's onboard terminal to search for the receiving devices of the surrounding vehicle management equipment and send the vehicle status information triggered by the user. This vehicle status information is used to indicate the current vehicle status or the preset vehicle status that the vehicle is about to enter.

[0084] Alternatively, vehicle status information can be collected passively. For example, vehicle sensing devices, such as geomagnetic coils that can sense the approach of vehicles, can be installed at locations like paid parking spaces, unloading checkpoints, and loading checkpoints. When a vehicle enters a location, the vehicle status corresponding to the preset function of that location can be defaulted. For example, if a vehicle is in a paid parking space, the vehicle sensing information collected by that paid parking space can be used to indicate that a vehicle is in a parked state. Or, if a vehicle is in a loading checkpoint, the vehicle sensing information collected by that loading checkpoint can be used to indicate that a vehicle is in a loading state.

[0085] The vehicle's sensing equipment can be specific to sensors such as geomagnetic coils, infrared sensors, and ultrasonic sensors, and can be adjusted according to actual needs.

[0086] Step S402: Perform vehicle tracking in consecutive frames of the vehicle image and identify the vehicle's motion trajectory.

[0087] On the other hand, this application can also perform dynamic vehicle recognition and detect vehicle status through image processing.

[0088] For example, vehicles can be identified in an image first, and then the identified vehicles can be tracked to track their movement trajectory. It can be understood that during tracking, multiple dense tracking points will be identified and their movement trajectories will be tracked. The movement trajectories of these tracking points can then be analyzed to obtain the vehicle's movement trajectory.

[0089] For example, using a pre-defined corner detection algorithm (Comer Detection) and based on a pre-defined detection range for the images, corner detection can be performed on the vehicles in the aforementioned N frames of images, yielding a large number of corner points. These corner points are specific image feature points, such as the intersection of two or more edges, points where the image gradient rate or gradient change rate reaches a threshold, points where the object edge is discontinuous, or pixels corresponding to a local maximum of the first derivative (i.e., the grayscale gradient).

[0090] Then, the preset optical flow algorithm is called to calculate the optical flow of these corner points, which serves as the vehicle's motion trajectory. The optical flow method is suitable for motion tracking and has high detection accuracy.

[0091] Taking the trajectory of a corner point in two images as an example, the coordinates of that corner point in the current image are (x... i y i The displacement of this corner point is (Δx) i Δy i Then the trajectory of that corner point is ((x i y i ), (x i +Δx i y i +Δy i )) , and so on, calculate the motion trajectory of the corner point in N frames of images, and denot it as (P i ), i∈(0,N).

[0092] In addition, as mentioned above, staff can annotate vehicle information on abnormal vehicle images uploaded to the server. In practical applications, if these images are consecutive frames, they can be annotated using optical flow maps. An optical flow map is an image composed of optical flow information that can show the motion trajectory of each feature point. The optical flow map of the feature points of the image region corresponding to the vehicle status information to be annotated can be overlaid and the corresponding semantic information can be configured to complete the annotation. Of course, license plate information and cargo loading information mentioned later can also be annotated in this way.

[0093] Step S403: Detect whether the motion trajectory matches the current motion state or the preset motion state. If not, trigger step S404.

[0094] Understandably, this application can configure corresponding movement trajectory ranges for different vehicle states. For example, if it is in a parked state, it has the characteristic of a movement trajectory that remains unchanged for a long time; if it is in a loading state, it has the movement trajectory characteristics of parking at the unloading gate, opening the door, and unloading.

[0095] In this way, we can compare whether the vehicle's motion trajectory identified in the vehicle image matches the motion state indicated in the vehicle status information. If they match, the vehicle status recognition is normal; if they do not match, the image recognition may be incorrect, and an abnormal situation may have occurred.

[0096] In practical applications, there are several typical cases of abnormal vehicle status recognition, such as:

[0097] 1. Originally required to be on duty continuously, but currently there is no vehicle available;

[0098] 2. When the current checkpoint is idle (no vehicles), the vehicle will be in loading, unloading, or resting status.

[0099] 3. When vehicles are at the checkpoint (continuously on duty, continuously resting, or continuously loading and unloading), there is no vehicle present.

[0100] Step S404: Confirm the vehicle image as an abnormal vehicle image with abnormal vehicle information recognition.

[0101] If a vehicle image is detected whose movement trajectory does not match the corresponding vehicle status information, the image can be identified as an abnormal vehicle image.

[0102] II. License Plate Information Recognition

[0103] License plate recognition compares the identified license plate information with preset standard license plate information, such as license plate size and the number of characters (letters). If they do not match, the current vehicle image is confirmed to be abnormal. Furthermore, other information can be combined to determine whether the license plate recognition result is normal.

[0104] In yet another exemplary implementation, see [reference] Figure 5 The diagram illustrates another process for vehicle information recognition from vehicle images according to this application. Specifically, vehicle information recognition from vehicle images in this application may include the following steps:

[0105] Step S501: When the preset location of the vehicle image is the exit location of the target site and the license plate recognition result of the vehicle image is normal, check whether there are still the same license plate recognition results among the license plate recognition results of historical vehicle images within the preset time period. If not, trigger step S502.

[0106] For example, if there are driving routes or driving directions planned on the driving roads in the target site, one or more exit locations can be set up along the driving routes or driving directions. That is to say, in this application, the exit locations can be exit locations set up between the entire parking lot and the outside of the site, or exit locations set up within the parking lot.

[0107] In practical applications, it can be assumed that when a vehicle passes through an exit location, it must have passed through the corresponding entrance location some time before. Therefore, the vehicle must have left a vehicle image and its license plate recognition result at the entrance location.

[0108] Therefore, if the vehicle management equipment detects that the license plate recognition results of historical vehicle images within a preset time period do not include the current vehicle image, it is clear that there was an abnormality in the license plate recognition of the current vehicle in the previous historical vehicle images. Therefore, although the current vehicle image can normally recognize the license plate, since there is no corresponding historical license plate recognition result, it can be considered that the license plate has a special situation, which can trigger the feedback mechanism of this application, that is, trigger the subsequent step S602 to identify the current vehicle image as an abnormal vehicle image.

[0109] Step S502: The vehicle image is confirmed as an abnormal vehicle image with abnormal vehicle information recognition.

[0110] After step S502 is triggered, the current vehicle image can be identified as an abnormal vehicle image.

[0111] In yet another exemplary implementation, see [reference] Figure 6 The diagram illustrates a process for vehicle information recognition from a vehicle image, as described in this application. Specifically, vehicle information recognition from a vehicle image may further include the following steps:

[0112] Step S601: Detect whether the license plate in the vehicle image is complete. If the license plate in the vehicle image is missing, then trigger step S602.

[0113] It is understandable that in practical applications, driving habits or camera angles of some vehicles may affect the position or display effect of the license plate in the image. Since the number of license plates is fixed, if the license plate identified by the model is missing, this part of the data is definitely data that the model has difficulty judging. Therefore, the image can also be used as an image of an abnormal vehicle.

[0114] Specifically, it can detect whether the border shape of the license plate in the image is rectangular or a similar shape. If it is neither rectangular nor a similar shape, then there is obviously a missing license plate.

[0115] Step S602: Confirm the vehicle image as an abnormal vehicle image with abnormal vehicle information recognition.

[0116] Once a vehicle image with a missing license plate is detected, it can be identified as an abnormal vehicle image.

[0117] III. Identification of Carriage Loading Information

[0118] The vehicle loading information recognition function compares the identified loading information with preset standard loading information, such as vehicle size, loading rate changes, and loaded items. If they do not match, the current vehicle image is confirmed as an abnormal vehicle image. Furthermore, other information can be combined to determine whether the vehicle loading information recognition result is normal.

[0119] In yet another exemplary implementation, see [reference] Figure 7 The diagram illustrates another process for vehicle information recognition from vehicle images according to this application. Specifically, vehicle information recognition from vehicle images in this application may include the following steps:

[0120] Step S701: Identify the cargo compartment loading information in the vehicle image to obtain the cargo compartment loading information identification result;

[0121] In this application, the loading information of the vehicle compartment is identified by an initial vehicle information recognition model. This loading information is used to indicate the loading status of the vehicle compartment. For example, it can typically be represented by the loading rate. For a logistics vehicle that has completed loading, its compartment should obviously be loaded with a large number of logistics items, thus having a high loading rate. For a logistics vehicle that has completed unloading, its compartment should not be loaded with any logistics items or should be loaded with very few logistics items, thus having a low loading rate.

[0122] Step S702: Determine whether the size of the detection box indicated in the vehicle loading information is smaller than the preset detection box size. If so, trigger step S703.

[0123] It is understandable that the identification results of the carriage loading information can not only indicate whether the identification is normal, the normally identified carriage loading information (such as the carriage load rate), and the reasons for identification abnormalities (such as the image not being able to be opened, the image not containing the carriage, the identification time being too long), but also explain some situations in the identification process.

[0124] For example, in this application, the size of the vehicle compartment detection frame during the recognition process can be used to further determine whether the vehicle compartment loading information recognition result is normal.

[0125] It is understandable that in the application scenario of this application, the normal vehicle image captured should not only include the vehicle compartment, but also clearly reflect the condition of the vehicle compartment. Therefore, if the vehicle compartment detection box obtained during the recognition process is very small or there is no vehicle compartment detection box at all, it is obviously not in line with the actual application scenario. The current vehicle image is a difficult sample to recognize for the initial vehicle information recognition model, which leads to recognition errors or failure to recognize the vehicle.

[0126] The area enclosed by the carriage detection frame is the carriage image identified in the image. The detection frame is generally configured as a rectangle and indicated by coordinate values ​​(x, y).

[0127] Step S703: Confirm the vehicle image as an abnormal vehicle image with abnormal vehicle information recognition.

[0128] Therefore, when the detected vehicle detection box size indicated in the recognition result is smaller than the preset detection box size, the corresponding vehicle image can be identified as an abnormal vehicle image.

[0129] In yet another exemplary implementation, see [reference] Figure 8 The diagram illustrates another process for vehicle information recognition from vehicle images according to this application. Specifically, vehicle information recognition from vehicle images in this application may include the following steps:

[0130] Step S801: Identify the license plate, vehicle status, and cargo loading information in the vehicle image to obtain the license plate recognition result, vehicle status recognition result, and cargo loading information recognition result.

[0131] Understandably, this application may also combine license plate recognition results, vehicle status recognition results, and cargo compartment loading information recognition results to determine whether vehicle information is being recognized correctly.

[0132] For example, the vehicle information recognition model mentioned in this application includes an initial vehicle information recognition model, which, in addition to having the function of recognizing the loading information of the vehicle compartment, may also have the function of recognizing the license plate and vehicle status.

[0133] Alternatively, the vehicle information recognition model mentioned in this application can also be configured with corresponding sub-recognition models for vehicle loading information recognition, license plate recognition, and vehicle status recognition.

[0134] Step S802: Determine whether the license plate recognition result, vehicle status recognition result, and cargo compartment loading information recognition result match. If they do, then trigger step S803.

[0135] After obtaining the license plate recognition result, vehicle status recognition result, and cargo loading information recognition result, this application can determine the abnormal vehicle image based on whether the features of the three match.

[0136] For example, information about the cargo compartment is only available when the vehicle is arriving at the port and departing from the post, and license plate information is only available during the arrival and departure processes. That is, in the vehicle status from entering the port to leaving the post, there will be a time sequence of license plate recognition results, cargo compartment loading information recognition results, and license plate recognition results. If the time sequence does not meet this time sequence characteristic, the current vehicle image can be regarded as an abnormal vehicle image with recognition anomalies.

[0137] For example, if there is no license plate recognition result or vehicle compartment detection result in the vehicle's arrival or departure status, then the current vehicle image is an abnormal vehicle image with recognition failure.

[0138] For example, in the vehicle status of arriving at or leaving the post, if the current license plate recognition result does not have a matching license plate recognition result in vehicle images taken at other locations within the preset event, the current vehicle image can also be regarded as an abnormal vehicle image with recognition anomalies.

[0139] For example, when a vehicle is in a state of arrival or departure but there is a situation where the vehicle loading rate does not conform to the preset vehicle arrival or departure status, the current vehicle image can be used as an abnormal vehicle image for identification.

[0140] For example, if a loading vehicle maintains a very low loading rate for an extended period of time, the current vehicle image can be used to identify abnormal vehicles.

[0141] For example, when the license plate and the size of the vehicle body are opposite, it is obviously not in line with the actual situation, and the current vehicle image can also be used as an abnormal vehicle image for identification.

[0142] Step S803: Confirm the vehicle image as an abnormal vehicle image with abnormal vehicle information recognition.

[0143] Therefore, when the temporal change characteristics of the license plate recognition result, vehicle status recognition result, and cargo loading information recognition result do not conform to the preset temporal change characteristics, the corresponding vehicle image can be identified as an abnormal vehicle image.

[0144] Secondly, in addition to combining the vehicle status information, license plate information, and cargo loading information mentioned above to determine whether the vehicle information identification is abnormal, other reference factors can also be used, such as the reliability of the vehicle information identification results and the blurriness of the vehicle image itself.

[0145] In yet another exemplary implementation, see [reference] Figure 9 The diagram shown illustrates another process for vehicle information recognition from vehicle images according to this application. In this application, vehicle information recognition from vehicle images may further include the following steps:

[0146] Step S901: Obtain the confidence level of the vehicle state recognition result, wherein the confidence level is used to indicate the credibility of the vehicle state recognition result. If the confidence level is lower than the preset confidence level threshold, step S1002 is triggered.

[0147] It is understandable that when a neural network model is used to identify vehicle information from an input vehicle image, the model output can include not only the specific content of the vehicle information, but also the confidence level of the specific content of the vehicle information. The confidence level is used to indicate the degree of credibility. The higher the confidence level, the higher the degree of credibility. On the other hand, the lower the confidence level, the more unstable the model is in the recognition of the image or the recognition may be inaccurate. The image may also be regarded as an abnormal vehicle image.

[0148] Step S902: The vehicle image is confirmed as an abnormal vehicle image with abnormal vehicle information recognition.

[0149] Thus, a confidence threshold can be set for the confidence level. When the confidence level is lower than the threshold, the vehicle image can be identified as an abnormal vehicle image regardless of whether the vehicle status can be identified.

[0150] In yet another exemplary implementation, see [reference] Figure 10 The diagram shown illustrates another process for vehicle information recognition from vehicle images according to this application. In this application, vehicle information recognition from vehicle images may further include the following steps:

[0151] Step S1001: Obtain the blur detection value of the vehicle image, wherein the blur detection value is used to indicate the blur degree of the vehicle image. If the blur detection value is higher than the preset blur threshold, step S1002 is triggered.

[0152] This is understandable; one can also judge abnormal vehicle images based on the degree of blurriness.

[0153] Specifically, blur detection values ​​can be used to indicate the degree of blur in vehicle images. For example, objects in normal vehicle images have clear outlines and high contrast, while objects in highly blurred vehicle images lack obvious contrast and there is a lack of significant pixel value variation between pixels. In practical applications, the second derivative of the image can be calculated to obtain the edges, and then the variance of the edges can be calculated to obtain the variance value. The variance value can be used to determine whether the image is blurry and the degree of blur.

[0154] Step S1002: Confirm the vehicle image as an abnormal vehicle image with abnormal vehicle information recognition.

[0155] When a vehicle image with a blur detection value higher than the blur threshold is detected, it can also be identified as an abnormal vehicle image. It is understandable that although vehicle information may be identifiable in such blurry vehicle images, most blurry vehicle images are still difficult to identify. Therefore, they can also be used as abnormal vehicle images to train new and more targeted vehicle information recognition models. Secondly, it can also improve the diversity of data samples, thereby improving the training effect of the vehicle information recognition model.

[0156] To facilitate better implementation of the vehicle information recognition model processing method provided in this application, this application also provides a vehicle information recognition model processing device.

[0157] See Figure 11 , Figure 11 This is a schematic diagram of a processing device for the vehicle information recognition model of this application, shown from the perspective of a local vehicle management device. In this application, the processing device 1100 for the vehicle information recognition model may specifically include the following structure:

[0158] The acquisition unit 1101 is used to acquire vehicle images captured by the camera at a preset location, wherein the preset location is a location in the target site where the vehicle management equipment is configured;

[0159] The recognition unit 1102 is used to recognize vehicle information from vehicle images;

[0160] The uploading unit 1103 is used to upload abnormal vehicle images to the server when there are abnormal vehicle images with abnormal vehicle information recognition. After the server assigns the target vehicle information added by the user to the abnormal vehicle images, it trains an initial neural network model based on the abnormal vehicle images and uses the trained model as the vehicle information recognition model. The vehicle information recognition model is used to identify the vehicle information of the vehicle corresponding to the target site.

[0161] The receiving unit 1104 is used to receive the vehicle information recognition model sent by the server.

[0162] In one exemplary implementation, the identification unit 1102 is specifically used for:

[0163] Acquire vehicle status information sent by the vehicle, user terminal or vehicle sensing device, wherein the vehicle status information is used to indicate the current motion state or preset motion state of the vehicle.

[0164] Vehicle tracking is performed within consecutive frames of a vehicle image, and the vehicle's trajectory is identified.

[0165] Detect whether the motion trajectory matches the current motion state or the preset motion state;

[0166] If not, the vehicle image will be identified as an abnormal vehicle image due to a vehicle information recognition anomaly.

[0167] In yet another exemplary implementation, the identification unit 1102 is specifically used for:

[0168] When the preset location of the vehicle image is the exit of the target site and the license plate recognition result of the vehicle image is normal, check whether there are still the same license plate recognition results among the license plate recognition results of historical vehicle images within a preset time period.

[0169] If not, the vehicle image will be identified as an abnormal vehicle image due to a vehicle information recognition anomaly.

[0170] In yet another exemplary implementation, the identification unit 1102 is specifically used for:

[0171] Detect whether the license plate in the vehicle image is complete;

[0172] If the license plate is missing in the vehicle image, the vehicle image will be identified as an abnormal vehicle image with abnormal vehicle information recognition.

[0173] In yet another exemplary implementation, the identification unit 1102 is specifically used for:

[0174] Identify the cargo compartment loading information in the vehicle image to obtain the cargo compartment loading information identification result;

[0175] Determine whether the size of the detection box indicated in the vehicle loading information is smaller than the preset detection box size;

[0176] If so, the vehicle image will be identified as an abnormal vehicle image with abnormal vehicle information recognition.

[0177] In yet another exemplary implementation, the identification unit 1102 is specifically used for:

[0178] The system identifies the license plate, vehicle status, and cargo loading information in vehicle images, and obtains the license plate recognition results, vehicle status recognition results, and cargo loading information recognition results.

[0179] Determine whether the license plate recognition result, vehicle status recognition result, and cargo loading information recognition result match;

[0180] If so, the vehicle image will be identified as an abnormal vehicle image with abnormal vehicle information recognition.

[0181] In yet another exemplary implementation, the identification unit 1102 is specifically used for:

[0182] Obtain the confidence level of the vehicle information recognition results, where the confidence level is used to indicate the degree of credibility of the vehicle information recognition results;

[0183] If the confidence level is lower than the preset confidence threshold, the vehicle image will be identified as an abnormal vehicle image with abnormal vehicle information recognition.

[0184] In yet another exemplary implementation, the identification unit 1102 is specifically used for:

[0185] Obtain the blur detection value of the vehicle image, where the blur detection value is used to indicate the degree of blur of the vehicle image;

[0186] If the blur detection value is higher than the preset blur threshold, the vehicle image is identified as an abnormal vehicle image with abnormal vehicle information recognition.

[0187] See Figure 12 , Figure 12 This is another schematic diagram of the processing device for the vehicle information recognition model of this application, shown from the server side on the cloud side. In this application, the processing device 1200 for the vehicle information recognition model may also include the following structure:

[0188] The receiving unit 1201 is used to receive abnormal vehicle images uploaded by the vehicle management device. The vehicle management device is used to acquire vehicle images captured by the camera at a preset location and to identify vehicle information in the vehicle images. The preset location is a location in the target site where the vehicle management device is configured, and the abnormal vehicle images are vehicle images with abnormal vehicle information identification.

[0189] The training unit 1202 is used to train an initial neural network model based on the abnormal vehicle image after the user adds target vehicle information to the abnormal vehicle image, and to use the trained model as a vehicle information recognition model. The vehicle information recognition model is used to identify the vehicle information of the vehicle corresponding to the target site.

[0190] Unit 1203 is used to send vehicle information recognition models to vehicle management equipment.

[0191] In one exemplary implementation, the abnormal vehicle image is a vehicle image whose motion trajectory does not conform to the current motion state or the preset motion state. The current motion state or the preset motion state is obtained from the vehicle status information sent by the vehicle, the user terminal, or the vehicle sensing device. The vehicle's motion trajectory is obtained by tracking and identifying the vehicle in consecutive frames of the vehicle image.

[0192] In another exemplary implementation, the abnormal vehicle image is defined as follows: when the preset location of the image is the exit of the target site and the license plate recognition result of the image is normal, no vehicle image with the same license plate recognition result is detected in the license plate recognition results of historical vehicle images within a preset time period.

[0193] In yet another exemplary implementation, the abnormal vehicle image is a vehicle image in which the license plate is missing.

[0194] In another exemplary implementation, the abnormal vehicle information is a vehicle image whose detection box size is smaller than the preset detection box size, as indicated in the vehicle loading information recognition result. The vehicle loading information recognition result is obtained by recognizing the vehicle loading information in the vehicle image.

[0195] In another exemplary implementation, the abnormal vehicle image is a vehicle image in which the license plate recognition result, vehicle status recognition result, and vehicle loading information recognition result do not match. The license plate recognition result, vehicle status recognition result, and vehicle loading information recognition result are obtained by recognizing the license plate, vehicle status, and vehicle loading information in the vehicle image.

[0196] In another exemplary implementation, the abnormal vehicle image is a vehicle image whose confidence level of the vehicle information recognition result is lower than a preset confidence threshold, wherein the confidence level is used to indicate the credibility of the vehicle information recognition result.

[0197] In another exemplary implementation, the abnormal vehicle image is a vehicle image whose blur detection value is higher than a preset blur threshold, wherein the blur detection value is used to indicate the degree of blur of the vehicle image.

[0198] This application also provides a processing device for vehicle information recognition models, see [link / reference]. Figure 13 , Figure 13 This diagram illustrates a structural schematic of a processing device for the vehicle information recognition model of this application. In practical applications, this device can be either a local vehicle management device or a cloud-based server, as mentioned above. Specifically, the processing device for the vehicle information recognition model of this application includes a processor 1301, a memory 1302, and an input / output device 1303. The processor 1301 executes the computer program stored in the memory 1302 to implement, for example... Figures 1 to 10 The processing steps of the vehicle information recognition model in any embodiment correspond to each step; or, when the processor 1301 executes the computer program stored in the memory 1302, it implements as follows: Figure 11 or Figure 12 Corresponding to the functions of each unit in the embodiment, the memory 1302 is used to store the functions executed by the processor 1301 as described above. Figures 1 to 10 The computer program required for the processing method of the vehicle information recognition model in any embodiment.

[0199] For example, a computer program may be divided into one or more modules / units, one or more of which are stored in memory 1302 and executed by processor 1301 to complete this application. One or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in a computer device.

[0200] The processing device for the vehicle information recognition model may include, but is not limited to, processor 1301, memory 1302, and input / output device 1303. Those skilled in the art will understand that the illustrations are merely examples of the processing device for the vehicle information recognition model and do not constitute a limitation on the processing device. It may include more or fewer components than illustrated, or combine certain components, or use different components. For example, the device may also include network access devices, buses, etc., and processor 1301, memory 1302, input / output device 1303, and network access devices are connected via a bus.

[0201] The processor 1301 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the vehicle information recognition model processing device, connecting various parts of the device through various interfaces and lines.

[0202] The memory 1302 can be used to store computer programs and / or modules. The processor 1301 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 1302 and by calling the data stored in the memory 1302. The memory 1302 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function, etc.; the data storage area may store data created by the use of the processing device based on the vehicle information recognition model, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0203] When processor 1301 executes a computer program stored in memory 1302, it can specifically perform the following functions:

[0204] Acquire vehicle images captured by cameras at preset locations and perform vehicle information recognition on the vehicle images. The preset locations are locations within the target area where vehicle management equipment is configured.

[0205] When there are abnormal vehicle images with abnormal vehicle information recognition, the abnormal vehicle images are uploaded to the server. After the server assigns the target vehicle information added by the user to the abnormal vehicle images, it trains an initial neural network model based on the abnormal vehicle images and uses the trained model as the vehicle information recognition model. The vehicle information recognition model is used to identify the vehicle information of the vehicle corresponding to the target site.

[0206] The vehicle information recognition model is received from the server.

[0207] Alternatively, when processor 1301 executes a computer program stored in memory 1302, it may also perform the following functions:

[0208] Receive abnormal vehicle images uploaded by vehicle management equipment. The vehicle management equipment is used to acquire vehicle images captured by cameras at preset locations and to identify vehicle information in the vehicle images. The preset location is a location in the target site where the vehicle management equipment is configured. Abnormal vehicle images are vehicle images with abnormal vehicle information identification.

[0209] After the abnormal vehicle image is assigned to the user with the target vehicle information, an initial neural network model is trained based on the abnormal vehicle image, and the trained model is used as the vehicle information recognition model. The vehicle information recognition model is used to identify the vehicle information of the vehicle corresponding to the target site.

[0210] The vehicle information recognition model is distributed to the vehicle management equipment.

[0211] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the vehicle information recognition model processing device, equipment, and its corresponding units described above can be found in, for example... Figures 1 to 10 The specific processing method of the vehicle information recognition model in any embodiment will not be repeated here.

[0212] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0213] Therefore, this application provides a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute the present application. Figures 1 to 10 For the steps in the vehicle information recognition model processing method corresponding to any embodiment, please refer to the following for specific operations: Figures 1 to 10 The processing method of the vehicle information recognition model in any embodiment will not be described again here.

[0214] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0215] Because of the instructions stored in the computer-readable storage medium, the present application can be executed as described above. Figures 1 to 10 The steps in the processing method of the vehicle information recognition model in any embodiment can be used to achieve the purpose of this application. Figures 1 to 10 For details on the beneficial effects that the vehicle information recognition model processing method can achieve in any embodiment, please refer to the preceding description, which will not be repeated here.

[0216] The processing method, apparatus, device, and computer-readable storage medium of the vehicle information recognition model provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for processing vehicle information recognition models, characterized in that, The method, applied to vehicle management equipment, includes: The system acquires vehicle images captured by a camera at a preset location and performs vehicle information recognition on the vehicle images, wherein the preset location is a location in the target site where the vehicle management device is configured; The vehicle information recognition process for the vehicle image further includes: acquiring vehicle status information sent by the vehicle, user terminal, or vehicle sensing device, wherein the vehicle status information is used to indicate the current motion state or preset motion state of the vehicle; tracking the vehicle in consecutive frames contained in the vehicle image and identifying the vehicle's motion trajectory; detecting whether the motion trajectory conforms to the current motion state or the preset motion state; if not, then confirming the vehicle image as an abnormal vehicle image with abnormal vehicle information recognition. When there is an abnormal vehicle image with abnormal vehicle information recognition, the abnormal vehicle image is uploaded to the server. After the server assigns the target vehicle information added by the user to the abnormal vehicle image, it trains an initial neural network model based on the abnormal vehicle image and uses the trained model as the vehicle information recognition model. The vehicle information recognition model is used to identify the vehicle information of the vehicle corresponding to the target site. There is a one-to-one correspondence between the vehicle information recognition model and the vehicle management device. Receive the vehicle information recognition model sent by the server.

2. The method according to claim 1, characterized in that, The vehicle information recognition of the vehicle image includes: When the preset location of the vehicle image is the exit location of the target site and the license plate recognition result of the vehicle image is normal, check whether there are still identical license plate recognition results among the license plate recognition results of historical vehicle images within a preset time period. If not, the vehicle image will be identified as an abnormal vehicle image with abnormal vehicle information recognition.

3. The method according to claim 1, characterized in that, The vehicle information recognition of the vehicle image includes: Detect whether the license plate in the vehicle image is complete; If the license plate in the vehicle image is missing, the vehicle image will be identified as an abnormal vehicle image with abnormal vehicle information recognition.

4. The method according to claim 1, characterized in that, The vehicle information recognition of the vehicle image includes: Identify the cargo compartment loading information in the vehicle image to obtain the cargo compartment loading information identification result; Determine whether the size of the detection box indicated in the carriage loading information is smaller than the preset detection box size; If so, the vehicle image is confirmed as an abnormal vehicle image with abnormal vehicle information recognition.

5. The method according to claim 1, characterized in that, The vehicle information recognition of the vehicle image includes: The license plate, vehicle status, and cargo loading information in the vehicle image are identified to obtain license plate recognition results, vehicle status recognition results, and cargo loading information recognition results. Determine whether the license plate recognition result, the vehicle status recognition result, and the cargo compartment loading information recognition result match; If so, the vehicle image is confirmed as an abnormal vehicle image with abnormal vehicle information recognition.

6. The method according to claim 1, characterized in that, The vehicle information recognition of the vehicle image includes: The confidence level of the vehicle information recognition result is obtained, wherein the confidence level is used to indicate the degree of credibility of the vehicle information recognition result; If the confidence level is lower than the preset confidence threshold, the vehicle image is identified as an abnormal vehicle image with abnormal vehicle information recognition.

7. The method according to claim 1, characterized in that, The vehicle information recognition of the vehicle image includes: Obtain the blur detection value of the vehicle image, wherein the blur detection value is used to indicate the degree of blur of the vehicle image; If the blur detection value is higher than the preset blur threshold, the vehicle image is identified as an abnormal vehicle image with abnormal vehicle information recognition.

8. A method for processing vehicle information recognition models, characterized in that, The method includes: The system receives abnormal vehicle images uploaded by a vehicle management device. The vehicle management device acquires vehicle images captured by a camera at a preset location and performs vehicle information recognition on the images. This vehicle information recognition further includes: acquiring vehicle status information sent by a vehicle, a user terminal, or a vehicle sensing device, wherein the vehicle status information indicates the vehicle's current motion state or a preset motion state; tracking the vehicle within consecutive frames of the vehicle image and identifying its motion trajectory; detecting whether the motion trajectory matches the current motion state or the preset motion state; and if not, confirming the vehicle image as an abnormal vehicle image with abnormal vehicle information recognition. After the abnormal vehicle image is assigned to the user with the target vehicle information, an initial neural network model is trained based on the abnormal vehicle image, and the trained model is used as the vehicle information recognition model. The vehicle information recognition model is used to identify the vehicle information of the vehicle corresponding to the target site, and there is a one-to-one correspondence between the vehicle information recognition model and the vehicle management device. The vehicle information recognition model is sent to the vehicle management equipment.

9. A processing device for a vehicle information recognition model, characterized in that, Applications to vehicle management equipment, the device comprising: The acquisition unit is used to acquire vehicle images captured by the camera at a preset location, wherein the preset location is a location in the target site where the vehicle management device is configured; The identification unit is used to identify vehicle information in the vehicle image. The identification of vehicle information in the vehicle image further includes: acquiring vehicle status information sent by a vehicle, a user terminal, or a vehicle sensing device, wherein the vehicle status information is used to indicate the current motion state or a preset motion state of the vehicle; tracking the vehicle in consecutive frames contained in the vehicle image and identifying the vehicle's motion trajectory; detecting whether the motion trajectory conforms to the current motion state or the preset motion state; if not, confirming the vehicle image as an abnormal vehicle image with abnormal vehicle information identification. The uploading unit is used to upload the abnormal vehicle image to the server when there is an abnormal vehicle image with abnormal vehicle information recognition. After the server assigns the target vehicle information added by the user to the abnormal vehicle image, it trains an initial neural network model based on the abnormal vehicle image and uses the trained model as the vehicle information recognition model. The vehicle information recognition model is used to identify the vehicle information of the vehicle corresponding to the target site. There is a one-to-one correspondence between the vehicle information recognition model and the vehicle management device. The receiving unit is used to receive the vehicle information recognition model sent by the server.

10. A processing device for a vehicle information recognition model, characterized in that, The device includes: A receiving unit is configured to receive abnormal vehicle images uploaded by a vehicle management device. The vehicle management device acquires vehicle images captured by a camera at a preset location and performs vehicle information recognition on the vehicle images. The vehicle information recognition also includes: acquiring vehicle status information sent by a vehicle, a user terminal, or a vehicle sensing device, wherein the vehicle status information indicates the vehicle's current motion state or a preset motion state; tracking the vehicle within consecutive frames of the vehicle image and identifying the vehicle's motion trajectory; detecting whether the motion trajectory matches the current motion state or the preset motion state; and if not, confirming the vehicle image as an abnormal vehicle image with abnormal vehicle information recognition. The training unit is used to train an initial neural network model based on the abnormal vehicle image after the user adds target vehicle information to the abnormal vehicle image, and to use the trained model as a vehicle information recognition model. The vehicle information recognition model is used to identify the vehicle information of the vehicle corresponding to the target site, and there is a one-to-one correspondence between the vehicle information recognition model and the vehicle management device. The distributing unit is used to distribute the vehicle information recognition model to the vehicle management equipment.

11. A processing device for a vehicle information recognition model, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, it executes the processing method of the vehicle information recognition model as described in any one of claims 1 to 8.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to execute the processing method of the vehicle information recognition model according to any one of claims 1 to 8.

Images