Vehicle identification method and device, electronic equipment, storage medium and program product

Abstract

The invention relates to the field of information processing, and provides a vehicle identification method and device, electronic equipment, a storage medium and a program product. The method comprises the following steps: acquiring an identity label of an electronic tag carried by a target vehicle and an image feature of the target vehicle; determining a first identifier of the target vehicle according to the identity identifier, and determining a second identifier of the target vehicle according to the image features; if the first identifier is the same as the second identifier; obtaining a first confidence coefficient corresponding to the first identifier, a second confidence coefficient corresponding to the second identifier, a similarity between a historical image feature of a vehicle corresponding to the first identifier and an image feature of a target vehicle, and a co-travel probability of the vehicle corresponding to the first identifier and a vehicle corresponding to the second identifier; determining a third identifier of the target vehicle based on a random forest model; and updating the first confidence coefficient and the second confidence coefficient according to the information, and determining a real vehicle identifier of the target vehicle according to the updated information. According to the method, accurate identification of the electric vehicle can be realized.

Description

Technical Field

[0001] This application relates to the field of information processing, and more particularly to a vehicle identification method, device, electronic device, storage medium, and program product. Background Technology

[0002] With the rapid increase in the number of electric vehicles, the management challenges of electric vehicles in residential communities are becoming increasingly prominent. Existing management solutions have many shortcomings and are unable to meet the dual needs of community safety management and convenient use by residents. In recent years, fires caused by battery explosions of indoor electric vehicles have resulted in significant personal injury and property damage. Therefore, implementing safe and reliable management of electric vehicles is both urgent and essential.

[0003] Currently, the management technology for electric vehicles in residential communities mainly revolves around safety management, parking management, charging management, and entry warnings. Some management solutions deploy readers at community entrances and garage entrances to identify the tag information of electric vehicles to control the gate opening and closing. However, in scenarios with multiple vehicles driving side by side, metal obstructions, or electromagnetic interference, the RFID signal is prone to missed readings and misreadings. Some management solutions rely on high-definition smart cameras deployed in the community's public areas, using built-in artificial intelligence algorithms to achieve license plate recognition, vehicle type recognition, and behavior analysis, thereby issuing warnings for violations by electric vehicles. However, these solutions are greatly affected by factors such as light and weather, resulting in poor environmental adaptability. Other management solutions attempt to combine RFID technology with visual perception technology. After the reader reads the signal from the electronic tag, it triggers a nearby camera to take a picture, comparing the electronic tag ID with the license plate recognition result. If they match, entry is granted. However, these solutions lack an effective conflict arbitration mechanism, leading to a high conflict rate in scenarios during peak hours (multiple vehicles driving side by side, frequent obstructions), making accurate identification of electric vehicles impossible and resulting in poor quality management of electric vehicles. Summary of the Invention

[0004] This application provides a vehicle identification method, device, electronic device, storage medium, and program product to solve the problem that the prior art cannot accurately identify electric vehicles, resulting in poor management quality of electric vehicles.

[0005] In a first aspect, this application provides a vehicle identification method applied to an edge device, the method comprising: Upon receiving a command from the central server to initiate vehicle identification, the system acquires the identification of the electronic tag carried by the target vehicle and the image features of the target vehicle. The first identifier of the target vehicle is determined based on the identity identifier, and the second identifier of the target vehicle is determined based on the image features; If the first identifier and the second identifier are different, obtain the first confidence level corresponding to the first identifier, the second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier traveling together, and the third identifier of the target vehicle determined based on the random forest model. The first confidence level and the second confidence level are updated based on the similarity, the peer probability, and the third identifier. The true vehicle identifier of the target vehicle is determined based on the updated first and second confidence levels.

[0006] Secondly, this application provides a vehicle identification method applied to a central server, the method comprising: Send a command to the edge device to initiate vehicle recognition; The command to initiate vehicle identification instructs the edge device to: upon receiving the command from the central server, acquire the identity of the electronic tag carried by the target vehicle and the image features of the target vehicle; determine a first identifier of the target vehicle based on the identity, and determine a second identifier of the target vehicle based on the image features; if the first identifier and the second identifier are different, acquire a first confidence level corresponding to the first identifier, a second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of traveling together between the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier, and a third identifier of the target vehicle determined based on a random forest model; update the first confidence level and the second confidence level based on the similarity, the probability of traveling together, and the third identifier; and determine the true vehicle identifier of the target vehicle based on the updated first confidence level and second confidence level.

[0007] Thirdly, this application provides a vehicle recognition device for use in edge devices, the device comprising: The first acquisition module is used to acquire the identity of the electronic tag carried by the target vehicle and the image features of the target vehicle when it receives a command from the central server to start vehicle recognition. The first determining module is used to determine a first identifier of the target vehicle based on the identity identifier, and to determine a second identifier of the target vehicle based on the image features; The second acquisition module is used to acquire, when the first identifier and the second identifier are different, a first confidence level corresponding to the first identifier, a second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier traveling together, and a third identifier of the target vehicle determined based on a random forest model. The update module is used to update the first confidence level and the second confidence level based on the similarity, the peer probability, and the third identifier; The second determining module is used to determine the true vehicle identifier of the target vehicle based on the updated first confidence level and second confidence level.

[0008] Fourthly, this application provides a vehicle identification device applied to a central server, the device comprising: The sending module is used to send commands to edge devices to initiate vehicle recognition; The command to initiate vehicle identification instructs the edge device to: upon receiving the command from the central server, acquire the identity of the electronic tag carried by the target vehicle and the image features of the target vehicle; determine a first identifier of the target vehicle based on the identity, and determine a second identifier of the target vehicle based on the image features; if the first identifier and the second identifier are different, acquire a first confidence level corresponding to the first identifier, a second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of traveling together between the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier, and a third identifier of the target vehicle determined based on a random forest model; update the first confidence level and the second confidence level based on the similarity, the probability of traveling together, and the third identifier; and determine the true vehicle identifier of the target vehicle based on the updated first confidence level and second confidence level.

[0009] Fifthly, this application provides an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement the steps of a vehicle identification method as described in the first or second aspect.

[0010] In a sixth aspect, this application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the vehicle identification method described in the first or second aspect.

[0011] In a seventh aspect, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of a vehicle identification method as described in the first or second aspect.

[0012] The vehicle identification method of this application first obtains the identity identifier of the electronic tag carried by the target vehicle and the image features of the target vehicle; then, it determines the first identifier of the target vehicle based on the identity identifier and the second identifier of the target vehicle based on the image features. If the first identifier and the second identifier are different, it then obtains the first confidence level corresponding to the first identifier, the second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of traveling together between the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier, and the third identifier of the target vehicle determined based on the random forest model. Based on the similarity, the probability of traveling together, and the third identifier, it updates the first confidence level and the second confidence level; finally, based on the updated first confidence level and second confidence level, it determines the true vehicle identifier of the target vehicle. This application synchronously acquires and verifies the identity and image features of the electronic tag of the target vehicle on the edge device. In the event of a conflict between the first and second identifiers, it introduces historical image feature similarity, peer probability, and a third identifier determined based on a random forest model as multi-dimensional arbitration factors to dynamically correct the first and second confidence levels. This constructs a scientific conflict arbitration mechanism that can effectively solve the conflict problem of heterogeneous perception recognition results, achieve accurate identification of the real vehicle identifier of electric vehicles, and thus significantly improve the management quality of electric vehicles. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart illustrating a vehicle identification method according to an embodiment of this application; Figure 2 This is a schematic diagram illustrating a conflict arbitration process as shown in an embodiment of this application; Figure 3 This is a schematic diagram illustrating the working principle of a random forest model as shown in an embodiment of this application; Figure 4 This is a flowchart illustrating another vehicle identification method shown in the embodiments of this application; Figure 5 This is a flowchart illustrating the registration information of a complete electronic tag, as shown in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of the central server shown in an embodiment of this application; Figure 7 This is a structural block diagram of a vehicle identification device shown in an embodiment of this application; Figure 8 This is a structural block diagram of another vehicle identification device shown in the embodiments of this application; Figure 9 This is a schematic diagram of the physical structure of an electronic device shown in an embodiment of this application. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] The vehicle identification method provided in this application can be used to manage electric vehicles in any type of area, such as residential communities, parks, shopping malls, and hospitals. To better illustrate the method, subsequent embodiments will use the management of electric vehicles within a residential community as an example.

[0017] This application divides the entire residential community into multiple sub-areas, such as the community gate area, unit door area, underground parking garage entrance area, elevator entrance area, and corridor area. Each sub-area is equipped with a Radio Frequency Identification (RFID) subsystem and a visual perception subsystem. The RFID subsystem includes a reader and an antenna. Electronic tags are installed on each electric vehicle within the community. The reader transmits radio frequency signals through its antenna. When an electric vehicle passes the reader, the electronic tag reflects its identification information back to the reader via the radio frequency signal. The visual perception subsystem includes an image acquisition device that automatically captures images of the electric vehicle while the reader reads the electronic tag's identification information. This image acquisition device can be a high-definition intelligent network camera (Internet Protocol Camera, IPC) with a built-in artificial intelligence algorithm chip, supporting functions such as license plate recognition, vehicle type recognition, and area intrusion analysis. In addition to being deployed in each sub-area, the image acquisition device can also be deployed in other public areas of the community, such as ground-level no-parking zones, to better manage electric vehicles within the community.

[0018] This application also provides edge devices for the RFID subsystem and visual perception subsystem in each sub-region. The edge devices are communicatively connected to the RFID and visual perception subsystems in their respective sub-regions. By analyzing the information collected by the RFID and visual perception subsystems, the edge devices achieve accurate identification of electric vehicles in their assigned sub-regions. The edge devices are also communicatively connected to a central server, which manages all RFID subsystems, visual perception subsystems, and edge devices within the cell.

[0019] The central server is responsible for receiving and integrating data collected by the radio frequency identification subsystem and the visual perception subsystem, assisting edge devices in local vehicle identification, and realizing electronic tag management, alarm push, and vehicle owner notification.

[0020] Therefore, this application constructs a three-layer architecture consisting of a radio frequency identification subsystem, a visual perception subsystem, and a central server, and achieves multimodal data fusion based on a deep collaborative algorithm. This effectively enables accurate identification and high-quality management of electric vehicles, thereby providing a guarantee for the security management of residential communities. The following section will describe in detail a vehicle identification method provided by this application. Figure 1 This is a flowchart illustrating a vehicle identification method according to an embodiment of this application. (Refer to...) Figure 1 The vehicle identification method of this application includes: Step S101: Upon receiving a command from the central server to initiate vehicle identification, acquire the identification of the electronic tag carried by the target vehicle and the image features of the target vehicle.

[0021] In this embodiment, when the central server needs to control an edge device to perform a vehicle recognition task, it sends a command to the edge device to start vehicle recognition; conversely, if it needs to stop the edge device from performing the vehicle recognition task, it sends a command to stop vehicle recognition. Therefore, when the edge device receives the command to start vehicle recognition, it enters vehicle recognition mode. In this mode, the edge device can identify electric vehicles in the target sub-area it is responsible for.

[0022] In this embodiment, when a target vehicle passes the reader in the target sub-region, the reader reads the identification mark of the electronic tag on the target vehicle using radio frequency identification (RFID) technology, and then sends the identification mark to the edge device responsible for the target sub-region. Simultaneously, while the reader reads the identification mark of the electronic tag on the target vehicle, the image acquisition device acquires an image of the target vehicle, performs preliminary feature analysis on the image to obtain image features, and then sends these image features to the edge device responsible for the target sub-region. The image needs to include a complete image of the target vehicle. Alternatively, the image acquisition device can also send the image of the target vehicle to the edge device, where the edge device can autonomously analyze the image to obtain image features.

[0023] Therefore, by executing step S101, the edge device can obtain the identity of the electronic tag and the image features of the target vehicle by having the reader actively send the electronic tag's identity identifier and the image acquisition device actively send the image or image features of the target vehicle.

[0024] In this embodiment, to avoid misjudgments caused by data asynchrony between multiple devices and to ensure accurate identification of the target vehicle, it is necessary to ensure time synchronization between the reader / writer and the image acquisition device as much as possible. Specifically, this embodiment can use Network Time Protocol (NTP) synchronization to ensure that the time when the reader / writer acquires the electronic tag's identification and the time when the image acquisition device acquires the image of the target vehicle are synchronized. NTP synchronization is a standard mechanism for clock synchronization between distributed computer systems using network protocols. In complex IoT sensing environments, its core function is to accurately measure network latency and local clock drift through round-trip communication between the client and the time server, thereby correcting the clock errors of heterogeneous devices in the system (such as the reader / writer and the image acquisition device in this embodiment) to a unified time base.

[0025] In this embodiment, the target vehicle can be any type of vehicle, including electric vehicles. For the sake of clarity in describing the method of this application, subsequent embodiments will be described using an electric vehicle as an example.

[0026] Step S102: Determine the first identifier of the target vehicle based on the identity identifier, and determine the second identifier of the target vehicle based on the image features.

[0027] In this embodiment, the edge device pre-stores a mapping table between the identity identifier of the electronic tag and the vehicle identifier of the target vehicle. Therefore, the edge device can directly use the vehicle identifier corresponding to the identity identifier of the electronic tag as the first identifier in the mapping table.

[0028] In this embodiment, the edge device can determine the second identifier of the target vehicle based on image features using any image recognition method. This embodiment does not impose specific limitations on the image recognition method.

[0029] The vehicle identification can be the vehicle's license plate number. The electronic tag's identification can be the electronic product code (EPC).

[0030] Step S103: If the first identifier and the second identifier are different, obtain the first confidence level corresponding to the first identifier, the second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier traveling together, and the third identifier of the target vehicle determined based on the random forest model.

[0031] In this embodiment, the difference between the first identifier and the second identifier indicates a conflict between the identification result of the vehicle identifier determined by the radio frequency identification method (first identifier) ​​and the identification result of the vehicle identifier determined by the image recognition method (second identifier). Steps S103-S105 are arbitration methods to resolve this conflict.

[0032] Here, the first confidence level represents the degree of confidence that the first identifier is the true vehicle identifier of the target vehicle. The second confidence level represents the degree of confidence that the second identifier is the true vehicle identifier of the target vehicle. The sum of the first and second confidence levels is the target value. This target value can be 1 or 100%.

[0033] In this embodiment, the image features of the target vehicle include, but are not limited to, vehicle model, color, and feature points. Feature points include, but are not limited to, added trunks, stickers, and wheel styles. The central server pre-stores a feature file for each electronic tag. This feature file includes the identity identifier of the associated electronic tag, the vehicle identifier of the bound vehicle, the visual feature information of the bound vehicle, and the owner's identity identifier. The visual feature information includes, but is not limited to, vehicle model, color, and feature points. Feature points include, but are not limited to, added trunks, stickers, and wheel styles. Therefore, the edge device can request the visual feature information associated with the vehicle corresponding to the first tag from the central server based on the identified electronic tag's identity identifier, and use this as the historical image feature of the vehicle corresponding to the first tag. Next, the edge device calculates the cosine similarity between the historical image feature of the vehicle corresponding to the first tag and the image feature of the target vehicle to obtain the similarity score.

[0034] In this embodiment, the central server also includes a vehicle travel history database to store multiple sets of vehicle identifiers that have traveled together (traveling together). The edge device can send the first identifier and the second identifier to the central server and receive at least one set of vehicle identifiers returned by the central server that simultaneously contains the first identifier and the second identifier. Then, it analyzes the probability of the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier traveling together based on the at least one set of vehicle identifiers.

[0035] In this embodiment, the edge device also includes a conflict arbitration model, which comprises a random forest model. The edge device constructs input features based on the necessary information required to obtain the true vehicle identifier of the target vehicle, and then inputs these input features into the conflict arbitration model. The conflict arbitration model then determines the third identifier of the target vehicle based on the random forest model. The third identifier can be the first identifier, the second identifier, or other identifiers indicating "requires manual review".

[0036] Step S104: Update the first confidence level and the second confidence level based on similarity, peer probability and third identifier.

[0037] In this embodiment, the edge device can update the first confidence level and the second confidence level based on similarity, peer probability, and a third identifier, for example, by increasing the first confidence level. When operating on one confidence level, an adaptive operation needs to be performed on the other confidence level to maintain the sum of the two as the target value.

[0038] Step S105: Determine the true vehicle identifier of the target vehicle based on the updated first confidence level and second confidence level.

[0039] In this embodiment, the true vehicle identifier of the target vehicle can be determined based on the maximum value of the updated first confidence level and the second confidence level.

[0040] The vehicle identification method of this application first obtains the identity identifier of the electronic tag carried by the target vehicle and the image features of the target vehicle; then, it determines the first identifier of the target vehicle based on the identity identifier and the second identifier of the target vehicle based on the image features. If the first identifier and the second identifier are different, it then obtains the first confidence level corresponding to the first identifier, the second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of traveling together between the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier, and the third identifier of the target vehicle determined based on the random forest model. Based on the similarity, the probability of traveling together, and the third identifier, it updates the first confidence level and the second confidence level; finally, based on the updated first confidence level and second confidence level, it determines the true vehicle identifier of the target vehicle. This application synchronously acquires and verifies the identity and image features of the electronic tag of the target vehicle on the edge device. In the event of a conflict between the first and second identifiers, it introduces historical image feature similarity, peer probability, and a third identifier determined based on a random forest model as multi-dimensional arbitration factors to dynamically correct the first and second confidence levels. This constructs a scientific conflict arbitration mechanism that can effectively solve the conflict problem of heterogeneous perception recognition results, achieve accurate identification of the real vehicle identifier of electric vehicles, and thus significantly improve the management quality of electric vehicles.

[0041] In conjunction with the above embodiments, in one implementation, step S104 may include: Step S1041: If the similarity is greater than the similarity threshold, or if the third identifier is the same as the first identifier, increase the first confidence level.

[0042] In this embodiment, if the similarity is greater than a similarity threshold, the first confidence level is increased by a first value, and the second confidence level is appropriately decreased by the first value. If the third identifier is the same as the first identifier, the first confidence level is increased by the first value, and the second confidence level is appropriately decreased by the first value. If both the similarity is greater than the similarity threshold and the third identifier is the same as the first identifier, the first confidence level is increased by twice the first value, and the second confidence level is appropriately decreased by twice the first value.

[0043] The similarity threshold can be set according to actual needs, for example, it can be 80%. The first value can also be set according to actual needs, for example, it can be 0.2.

[0044] Step S1042: If the probability of the same peer is greater than the probability threshold, or if the third identifier is the same as the second identifier, increase the second confidence level.

[0045] In this embodiment, if the peer probability is greater than the probability threshold, the second confidence level is increased by the first value, and the first confidence level is adaptively decreased by the first value. If the third identifier is the same as the second identifier, the second confidence level is increased by the first value, and the first confidence level is adaptively decreased by the first value. If both the peer probability is greater than the probability threshold and the third identifier is the same as the second identifier are satisfied simultaneously, the second confidence level is increased by twice the first value, and the first confidence level is adaptively decreased by twice the first value.

[0046] The probability threshold can be set according to actual needs, for example, it can be 60%.

[0047] Figure 2 This is a schematic diagram illustrating a conflict arbitration process according to an embodiment of this application. The implementation process of steps S1041-1043 described above can be referred to... Figure 2 As shown.

[0048] In this embodiment, by introducing historical image feature similarity, peer probability, and the prediction results of the random forest model (third identifier) ​​as the judgment criteria, the first confidence level and the second confidence level are adjusted, which can make the adjusted first confidence level and the second confidence level more accurate, thereby significantly improving the accuracy of vehicle recognition results.

[0049] In conjunction with the above embodiments, in one implementation, step S105 may include: The identifier corresponding to the maximum value between the updated first and second confidence levels is determined as the vehicle identifier of the target vehicle; or, If the maximum value is greater than the confidence threshold, the identifier corresponding to the maximum value is determined as the vehicle identifier of the target vehicle.

[0050] This embodiment provides two methods for determining the true vehicle identifier of a target vehicle: The first method involves determining the identifier corresponding to the maximum value between the updated first and second confidence levels as the vehicle identifier of the target vehicle.

[0051] For example, if the updated first confidence level and second confidence level are 0.70 and 0.30 respectively, then the first identifier corresponding to 0.70 will be taken as the real vehicle identifier.

[0052] The second approach is to determine the identifier corresponding to the maximum value as the vehicle identifier of the target vehicle when the maximum value is greater than the confidence threshold.

[0053] For example, if the updated first and second confidence levels are 0.92 and 0.08 respectively, and the confidence threshold is 0.90, then the first identifier corresponding to 0.92 will be taken as the real vehicle identifier.

[0054] In the second approach, if the maximum value is not greater than the confidence threshold, a manual review work order is generated, and staff determine the true vehicle identification based on the manual review work order.

[0055] In this embodiment, determining the real vehicle identifier based on the updated first confidence level and second confidence level can significantly improve the accuracy of the real vehicle identifier, thereby improving the management quality of electric vehicles.

[0056] In conjunction with the above embodiments, in one implementation, obtaining the identification of the electronic tag carried by the target vehicle and the image features of the target vehicle includes: The identification of the target vehicle is obtained from the electronic tag carried by the reader; and the image features of the target vehicle are obtained from the image acquisition device.

[0057] In this embodiment, key areas within the cell can be designated as sub-regions according to actual needs. A set of radio frequency identification (RFID) subsystems and visual perception subsystems are set up in each sub-region, and the set of RFID subsystems and visual perception subsystems serves as an edge node.

[0058] This application configures one edge device for each edge node. Each edge device locally deploys a conflict arbitration model, and the edge device trains the conflict arbitration model locally. Each edge device is used for vehicle identification in the sub-region it is responsible for.

[0059] Steps S101-S105 in this embodiment describe the working principle of an edge device. In actual implementation, all edge devices in the entire community can operate according to steps S101-S105, thereby achieving high-quality management of vehicles in the community.

[0060] In conjunction with the above embodiments, in one implementation, since the operation of the radio frequency identification subsystem and the image acquisition device has certain uncertainties, in order to better complete the identification of electric vehicles, this embodiment designs an environment-adaptive confidence calculation method to obtain a first confidence level and a second confidence level based on real-time environmental parameters and device status.

[0061] Specifically, step S103 may include: Step S1031: Obtain the signal strength of the electronic tag, the signal strength threshold of the electronic tag, the total number of electronic tags within the reader's recognition range, the number of interfering electronic tags within the reader's recognition range, the light intensity of the environment where the image acquisition device is located, and the occlusion rate of the target vehicle's markings.

[0062] In this embodiment, the interfering electronic tags within the reader's identification range include, but are not limited to, electronic tags of other electric vehicles besides the target vehicle, and false tags generated by reflections from metal objects.

[0063] Occlusion of signage includes, but is not limited to, being covered by dust or obscured by objects. The occlusion rate of signage can be determined by image acquisition equipment or by edge devices using image processing technology.

[0064] Step S1032: Determine the first confidence level based on signal strength, signal strength threshold, total number of electronic tags, number of interfering electronic tags, light intensity, and occlusion rate.

[0065] In this embodiment, the light intensity refers to the light intensity of the environment in which the image acquisition device is located.

[0066] Step S1033: Determine the difference between the target value and the first confidence level as the second confidence level.

[0067] In this embodiment, by analyzing multiple dimensions of features such as signal strength, total number of electronic tags, number of interfering electronic tags, light intensity, and occlusion rate, the first confidence level and the second confidence level are determined. This can effectively improve the accuracy of the obtained first confidence level and the second confidence level, thereby improving the accuracy of the recognition results of real vehicle identification.

[0068] In one implementation, in conjunction with the above embodiments, step S1032 may include: The initial first confidence level is determined based on the signal strength, signal strength threshold, total number of electronic tags, and number of interfering electronic tags; If the light intensity is less than the light intensity threshold, or the occlusion rate is greater than the occlusion rate threshold, the initial first confidence level is increased to obtain the first confidence level.

[0069] In this embodiment, the initial first confidence level can be calculated using the following formula: in, The initial first confidence level; The value is d (unit: dBm). This is the signal strength threshold (default -60dBm, which can be calibrated according to the cell environment). The number of interference electronic tags; This represents the total number of electronic tags.

[0070] Similarly, the difference between the target value and the initial first confidence level is determined as the initial second confidence level.

[0071] In this embodiment, the initial first confidence level or the initial second confidence level ranges from 0.20 to 0.80, ensuring that a single confidence level does not completely dominate the final decision.

[0072] After obtaining the initial first confidence level, if the light intensity is less than the light intensity threshold, the initial second confidence level can be multiplied by a number less than 1 (set according to actual needs, such as 0.8) to reduce the initial second confidence level. The difference between the target value and the reduced initial second confidence level is then used as the first confidence level. In this way, the initial first confidence level can be increased. The light intensity threshold can be set according to actual needs, for example, 200 lux.

[0073] After obtaining the initial first confidence level, if the occlusion rate is greater than the occlusion rate threshold, the initial second confidence level can be multiplied by a number less than 1 (set according to actual needs, such as 0.7) to reduce the initial second confidence level. The difference between the target value and the reduced initial second confidence level is then used as the first confidence level. In this way, the initial first confidence level can be increased. The occlusion rate threshold can be set according to actual needs, for example, 30%.

[0074] In this embodiment, the initial first confidence level is first determined based on the state of the radio frequency identification subsystem, and then the initial first confidence level is corrected based on the real-time environmental state (light intensity, occlusion rate of the sign) to obtain the first confidence level. In this way, the accuracy of the first confidence level can be effectively improved, thereby improving the accuracy of subsequent vehicle recognition results.

[0075] In conjunction with the above embodiments, in one implementation, the edge device can determine the third identifier through the following steps: Acquire the radio frequency link characteristics between the reader and the electronic tag, the visual perception characteristics of the image acquisition device, the spatiotemporal environmental characteristics of the target vehicle, and the prior characteristics of vehicle recognition. The radio frequency link features, visual perception features, spatiotemporal environment features, and vehicle recognition prior features are used as input features to input the random forest model to obtain the third identifier; The random forest model is trained based on multiple input feature samples and the corresponding electronic tags of each input feature sample. The random forest model includes multiple decision trees, each of which is used to output the identification prediction result of the target vehicle. The third identification is determined based on the identification prediction result with the most outputs from the decision trees.

[0076] In this embodiment, in order to achieve the best arbitration effect, three result labels are designed: label 0 (indicating that the first identifier is correct), label 1 (indicating that the second identifier is correct), and label 2 (indicating that manual review is required).

[0077] In this embodiment, each decision tree can obtain the probabilities of three outcome labels, namely: , , ,and Labels 0, 1, and 2 are represented using one-hot encoding as follows: For example, the probabilities of the three outcome labels obtained from a decision tree are: ,because The probability of the three possible labels from a decision tree is the highest, therefore the final output label of this decision tree is label 0. For example, the probability of a decision tree producing the three possible labels is: ,because The maximum value is given, therefore the final output label of this decision tree is label 1.

[0078] In this embodiment, the edge device counts all result labels, uses a majority voting mechanism to determine the final result label, and determines a third identifier based on the final result label: if label 0 has the most occurrences, the final result label is label 0, and the third identifier output by the conflict decision model is the first identifier; if label 1 has the most occurrences, the final result label is label 1, and the third identifier output by the conflict decision model is the second identifier. If the number of labels 0 and 1 is the same, or the number of labels 2 is the most, the final result label is label 2, and the third identifier output by the conflict decision model is the "requires manual review" identifier.

[0079] In this embodiment, the conflict decision model uses a random forest model to comprehensively analyze the radio frequency link characteristics between the reader and the electronic tag, the visual perception characteristics of the image acquisition device, the spatiotemporal environmental characteristics of the target vehicle, and the prior characteristics of vehicle identification. This enables the accurate determination of the third identifier, providing technical support for the subsequent accurate identification of the real vehicle identifier.

[0080] In one embodiment, based on the above embodiments, the radio frequency link characteristics include: the signal strength of the electronic tag (S_RF), the signal stability of the electronic tag (Sta_RF), and the number of interfering electronic tags within the reader's identification range (N_RF).

[0081] Visual perception features include: visual confidence level of the second sign (C_Vis), illumination intensity (L_Vis), and occlusion rate of the target vehicle's sign (O_Vis).

[0082] The visual confidence score of the second identifier refers to the confidence level of the image processing algorithm for the second identifier when the edge device analyzes the image features. In other words, the visual confidence score indicates the degree to which the image processing algorithm believes the second identifier to be a genuine vehicle identifier.

[0083] The spatiotemporal environmental features include: environment type (Env_Type), time period features (Time_Slot), weather features (Weather), and the time difference (ΔT) between the acquisition of the electronic tag's identification and the target vehicle's image features.

[0084] Environmental types could include, for example, the main gate of a residential complex, the entrance to an underground parking garage, an apartment building entrance, or an elevator entrance. Time-of-day characteristics could include, for example, morning rush hour, evening rush hour, off-peak hours, late at night, or early morning. Weather-specific features could include, for example, sunny days, rainy days, or snowy days.

[0085] Prior features for vehicle identification include: the historical matching rate between the electronic tag's identity identifier and the first identifier (Mat_His), and the probability of vehicles corresponding to the first identifier and vehicles corresponding to the second identifier traveling together (Co_His).

[0086] In this embodiment, taking into full account the influence of various factors, a 12-dimensional input feature is constructed. The third identifier is obtained through this 12-dimensional input feature, which can further improve the accuracy of the third identifier, thereby improving the accuracy of the vehicle recognition result.

[0087] In one implementation, after obtaining the 12-dimensional input features, the input features can be standardized, for example, by using the mean and standard deviation of the local training set to standardize the features, such as the signal strength of an electronic tag. The standardized formula can be: ,in The mean signal strength of the local training set. This represents the standard deviation of the signal strength in the local training set. This method significantly improves the quality of the input features, ensuring the accuracy of the obtained third identifier.

[0088] In one implementation, if the conflict decision model ultimately outputs a third identifier as "requires manual review," the edge device will generate a manual review work order. This work order may include feature importance analysis suggestions, such as "Due to light intensity = 50 lux (low light) and visual confidence level = 0.55, it is recommended to prioritize checking the radio frequency signal of the RFID subsystem." In this way, staff can quickly determine the third identifier.

[0089] After staff determine the review results based on the manual review work order, in addition to determining the third identifier based on the review results, the information in the review results (such as determining that the electronic tag is misread or that the license plate is obscured) can also be fed back to the federated model as samples, so that the federated model can periodically (e.g., monthly) iterate and optimize the random forest model, thereby improving the accuracy of the output results of the random forest model.

[0090] In one implementation, the maximum number of decision trees in the random forest model is 50, the maximum depth is 8 layers, and each decision tree selects 3 dimensions of features for splitting.

[0091] Figure 3 This is a schematic diagram illustrating the working principle of a random forest model as shown in an embodiment of this application. Figure 3 In model 'a', the random forest model includes 50 decision trees. The conflict arbitration model statistically analyzes the output labels of these 50 decision trees and finds that 38 decision trees are labeled 'label 0', 10 are labeled 'label 1', and 2 are labeled 'label 2'. Therefore, the conflict arbitration model ultimately outputs label 0 as the third label, indicating that the first label is correct.

[0092] Within each decision tree, three dimensions of features are selected for analysis to obtain the resulting label. Taking decision tree 1 as an example, assuming that the three selected dimensions of features are the electronic tag's signal strength (S_RF), light intensity (L_Vis), and time period feature (Time_Slot), as follows... Figure 3 As shown in b, the judgment process includes: First, the root node checks whether the standardized value (S_RF_norm) of the electronic tag's signal strength (S_RF) is less than or equal to 0.8. If so (indicating weak S_RF), it further checks whether the standardized value (L_Vis_norm) of the illumination intensity (L_Vis) is less than or equal to -0.3. If so, it checks whether the time slot feature (Time_Slot) is 1 (i.e., morning rush hour). If so, the leaf node outputs label 1 (i.e., visually correct / secondary identification correct, but there is multi-vehicle interference). If not, the leaf node outputs label 2 (requires manual review). If the standardized value (L_Vis_norm) of the illumination intensity (L_Vis) is greater than -0.3 (indicating normal illumination), the leaf node outputs label 1 (indicating visually correct / secondary identification correct). If the standardized value (S_RF_norm) of the signal strength (S_RF) in the root node is greater than 0.8 (indicating strong S_RF), the leaf node outputs label 0.

[0093] When training a random forest model locally on edge devices, due to limitations in computing power, the following cross-entropy loss function can be used: in, This represents the loss value for a single input feature sample. For result labels, including ; For the input feature sample to belong to the first The probability of class result labels, including , , .

[0094] In this way, edge devices can be trained based on a lightweight random forest model with a lightweight loss function, without consuming a lot of computing resources.

[0095] In conjunction with the above embodiments, in one implementation, the method of this application may further include: Receive model parameter update messages sent by the central server. The model parameter update messages include the target parameters of the random forest model. The target parameters are obtained by the central server using the federated averaging algorithm to process the model parameters of the random forest model reported by each edge device. Update the model parameters of the random forest model based on the target parameters.

[0096] In this embodiment, since the data collected by each edge device has the same feature dimensions but different samples (identifying different vehicles), a horizontal federated learning architecture is adopted. Without transmitting privacy data such as vehicle identification or electronic tag identity, the edge devices upload the model parameters of the conflict arbitration model (including the model parameters of the random forest model) to the central server. The central server processes the model parameters reported by each edge device to obtain the target parameters of the conflict arbitration model (including the target parameters of the random forest model), and then sends the target parameters back to each edge device, allowing the edge devices to update their model parameters based on the target parameters. This approach not only allows the conflict arbitration model of each edge device to adapt to its own environmental characteristics (e.g., large variations in gate lighting and significant radio frequency signal interference in the garage), but also improves the accuracy of the global conflict arbitration results of the conflict arbitration models in each edge device through parameter collaboration.

[0097] In this embodiment, since the sample size and data quality of each edge node are different (for example, the daily average number of conflict samples in the main entrance area is 100, while that in the corridor area is only 30), the central server can use a federated averaging algorithm to obtain the target parameters of the conflict arbitration model based on the sample size of each sub-region. The specific formula is as follows: in, This represents the total number of edge nodes (edge ​​devices) within the cell; For the first The number of samples for each edge node; For the first Model parameters of a conflict decision-making model within an edge device.

[0098] Finally, the central server can set the target parameters. The data is distributed to various edge devices, which then update the model parameters of the conflict arbitration model based on the target parameters, thereby optimizing the conflict arbitration capability and enabling each edge device to handle conflicts in the identification results more accurately and independently.

[0099] In this embodiment, the central server obtains the target parameters using a federated averaging algorithm based on the model parameters of the conflict arbitration model reported by each edge device, and then distributes the target parameters to each edge device. This effectively improves the conflict arbitration capability of the conflict arbitration model without exposing the data privacy of each edge device, providing technical support for accurate vehicle identification in the future.

[0100] In conjunction with the above embodiments, in one implementation, the method of this application may further include: If the first identifier and the second identifier are the same, the first identifier or the second identifier shall be identified as the vehicle identifier of the target vehicle.

[0101] In this embodiment, if the edge device determines that the first identifier and the second identifier are the same, then either identifier is directly used as the real vehicle identifier of the target vehicle, thereby improving the recognition efficiency of the vehicle identifier.

[0102] This application also provides a vehicle identification method applied to a central server. Figure 4 This is a flowchart illustrating another vehicle identification method according to an embodiment of this application. For example... Figure 4 As shown, the method of this application includes: Step S201: Send a command to the edge device to start vehicle recognition; The command to initiate vehicle recognition instructs the edge device to: upon receiving the command from the central server, acquire the identification of the electronic tag carried by the target vehicle and the image features of the target vehicle; determine the first identifier of the target vehicle based on the identification and the second identifier based on the image features; if the first identifier and the second identifier are different, acquire the first confidence level corresponding to the first identifier, the second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of traveling together between the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier, and the third identifier of the target vehicle determined based on a random forest model; update the first confidence level and the second confidence level based on the similarity, the probability of traveling together, and the third identifier; and determine the vehicle identifier of the target vehicle based on the updated first confidence level and the second confidence level.

[0103] The vehicle identification method of this application first obtains the identity identifier of the electronic tag carried by the target vehicle and the image features of the target vehicle; then, it determines the first identifier of the target vehicle based on the identity identifier and the second identifier of the target vehicle based on the image features. If the first identifier and the second identifier are different, it then obtains the first confidence level corresponding to the first identifier, the second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of traveling together between the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier, and the third identifier of the target vehicle determined based on the random forest model. The first confidence level and the second confidence level are updated based on the similarity, the probability of traveling together, and the third identifier; finally, the true vehicle identifier of the target vehicle is determined based on the updated first confidence level and the second confidence level. This application synchronously acquires and verifies the identity and image features of the electronic tag of the target vehicle on the edge device. In the event of a conflict between the first and second identifiers, it introduces historical image feature similarity, peer probability, and a third identifier determined based on a random forest model as multi-dimensional arbitration factors to dynamically correct the first and second confidence levels. This constructs a scientific conflict arbitration mechanism that can effectively solve the conflict problem of heterogeneous perception recognition results, achieve accurate identification of the real vehicle identifier of electric vehicles, and thus significantly improve the management quality of electric vehicles.

[0104] In conjunction with the above embodiments, in one implementation, the method of this application may further include: Identify the first electronic tag for the vehicle identification system that needs improvement. Obtain the image features of the first vehicle to which the first electronic tag belongs; The vehicle identifier of the first vehicle is determined based on its image features, and the first electronic tag is associated with and stored with the vehicle identifier of the first vehicle; or, Based on the image features of the first vehicle, obtain the multi-dimensional body features of the first vehicle, generate a virtual vehicle identifier based on the multi-dimensional body features, and associate and store the first electronic tag with the virtual vehicle identifier.

[0105] In this embodiment, each electronic tag corresponds to a feature profile. If the feature profile of an electronic tag lacks a vehicle identifier, then that electronic tag is designated as the first electronic tag.

[0106] In practice, the central server can acquire image features of the first vehicle from the image acquisition device via edge devices. Based on these image features, it determines the vehicle identifier and associates the first electronic tag with the vehicle identifier, storing it in the feature file corresponding to the first electronic tag. If the vehicle identifier cannot be determined from the image features, multiple dimensions of the vehicle body features (i.e., visual feature information of the vehicle), such as model, color, and feature points, are obtained. A virtual vehicle identifier is generated based on these features, and the first electronic tag is associated with the virtual vehicle identifier and stored in the feature file corresponding to the first electronic tag. When the vehicle identifier can be determined later, the virtual vehicle identifier is replaced with the original vehicle identifier.

[0107] In this embodiment, a ResNet50 neural network can be used to extract vehicle body features. Vehicle color recognition can employ the HSV color space model, dividing colors into 12 primary hue ranges (e.g., red: H=0-10° / 350-360°, S=40%-100%, V=40%-100%) to ensure the stability of color features under different lighting conditions. Feature points are extracted using the Scale Invariant Feature Transform (SIFT) algorithm, such as the shape of the trunk and the position of stickers; at least 10 stable feature points are extracted for each vehicle.

[0108] In this embodiment, when an electric vehicle X enters the residential area for the first time, the community assigns an electronic tag Y to the electric vehicle X. At this time, the feature file of the electronic tag Y consists of: the identity identifier of the electronic tag Y, the vehicle identifier of the electric vehicle X (empty), the visual feature information of the electric vehicle X (empty), and the owner's identity identifier (empty). By executing the method in this embodiment, the vehicle identifier of the electric vehicle X can be completed in the feature file of the electronic tag Y. This process can be referred to... Figure 5 As shown. Figure 5 This is a flowchart illustrating the registration information of a complete electronic tag, as shown in an embodiment of this application. (Refer to...) Figure 5 After electric vehicle X enters the residential area, the image acquisition device first captures an image of electric vehicle X, then obtains its visual feature information based on the image characteristics, and saves this visual feature information to the visual feature information field of electric vehicle X in the feature file. Next, if the vehicle identification of electric vehicle X can be clearly determined, it is saved to the vehicle identification field of electric vehicle X in the feature file. If the vehicle identification of electric vehicle X cannot be clearly determined, a virtual vehicle identification is generated based on the visual feature information of electric vehicle X, and then the virtual vehicle identification is saved to the vehicle identification field of electric vehicle X in the feature file.

[0109] The method in this embodiment can be applied to the management of newly registered electronic tags. This method can be used to manage electric vehicles entering the community for the first time, significantly improving the management quality of electric vehicles.

[0110] In conjunction with the above embodiments, in one implementation, the method of this application may further include: A second electronic tag for identifying the vehicle owner, which needs to be improved; The identity of the candidate vehicle owner is determined based on the time when the second vehicle, to which the second electronic tag belongs, enters the target area and the vehicle owner's passage records at the entrance and exit of the target area; Send vehicle binding requests to each candidate vehicle owner; Upon receiving a response message agreeing to the vehicle binding request, the identity of the candidate vehicle owner who sent the response message agreeing to the vehicle binding request will be associated and stored with the identity of the second electronic tag.

[0111] In this embodiment, the vehicle owner's identity identifier refers to the unique identifier of the vehicle owner in the central server, through which the vehicle owner's basic information can be determined.

[0112] In this embodiment, if the feature profile of an electronic tag lacks the vehicle owner's identity information, then the electronic tag is a second electronic tag.

[0113] In this embodiment, the central server can determine the identity of the candidate vehicle owner by analyzing the time the second vehicle enters the target area and the vehicle owner's passage records at the entrances and exits of the target area. Next, it can determine the contact information of the candidate vehicle owner based on the vehicle owner's identity and then send a vehicle binding request to each candidate vehicle owner using that contact information. If a response message agreeing to the vehicle binding request is received, the identity of the candidate vehicle owner who sent the response message is associated with the identity of the second electronic tag and stored in the feature file of the second electronic tag.

[0114] The second vehicle may be the same as or different from the first vehicle. Assuming the second vehicle is still an electric vehicle X, then... Figure 5 For example, the identity of a candidate vehicle owner can be determined based on the time electric vehicle X enters the community and the vehicle owner's passage records at the community's entrances and exits. Next, the central server sends a vehicle binding request to each candidate vehicle owner. This request may include an image of electric vehicle X and the text "Authorize registration?". If a response message agreeing to the vehicle binding request is received from a candidate vehicle owner, that candidate vehicle owner's identity is saved to the vehicle owner's identity field in the feature file for electric vehicle X. The method in this embodiment can be applied to the management of newly registered electronic tags. This method enables the management of electric vehicles entering the community for the first time, significantly improving the quality of electric vehicle management.

[0115] In conjunction with the above embodiments, in one implementation, the method of this application may further include: A command to start heartbeat monitoring is sent to each reader in the target area. The command to start heartbeat monitoring is used to instruct the reader to send heartbeat messages to the electronic tag and to determine whether the electronic tag is in normal working condition based on the response message of the heartbeat message.

[0116] The target area is the residential community.

[0117] In this embodiment, the command to initiate heartbeat monitoring sent by the central server to each reader carries a heartbeat monitoring strategy, such as the frequency of sending heartbeat messages. In this embodiment, the central server can set the heartbeat frequency for readers in each sub-area based on the coverage density of electronic tags in each sub-area (for example, setting the heartbeat message sending frequency of reader 1 located in the garage area to 1 hour, and the heartbeat message sending frequency of reader 2 located in the corridor area to 2 hours).

[0118] In this embodiment, if the reader determines that the electronic tag is in normal working condition based on the heartbeat message response, it replies to the central server with a message indicating that the electronic tag is in normal working condition. If the reader determines that the electronic tag is not in normal working condition based on the heartbeat message response, it replies to the central server with a message indicating that the electronic tag is not in normal working condition. The central server marks all electronic tags that are not in normal working condition as tags that are not "suspected to be invalid".

[0119] In this embodiment, the central server can manage the status of all electronic tags within the cell, ensuring that all electronic tags work normally, thereby guaranteeing the stable execution of the vehicle identification task.

[0120] In conjunction with the above embodiments, in one implementation, the method of this application may further include: Receive the identity information of a third electronic tag sent by a reader in the target area to verify whether it is invalid; Obtain vehicle identification or image features associated with the identity of the third electronic tag; If, within a preset time period, a message is received from an edge device within the target area indicating that a vehicle matching the vehicle identifier or image features has been identified, the third electronic tag is determined to be invalid.

[0121] In this embodiment, the central server determines that the "suspected invalid" electronic tag is a third electronic tag to be verified as invalid.

[0122] Next, the central server initiates a visual reverse verification process: it obtains the vehicle identifier (or image feature) M associated with the identity of the third electronic tag, and sends a search command to each edge device. Within a preset time period, each edge device, using its RFID subsystem and image acquisition equipment, determines whether it can find a vehicle matching the vehicle identifier (or image feature) M. If a match is found, the edge device returns a message to the central server indicating that a vehicle matching the vehicle identifier (or image feature) M has been identified, allowing the central server to determine that the third electronic tag is invalid.

[0123] If none of the edge devices find a vehicle matching the vehicle identifier (or image feature) M within the preset time period, they return a message to the central server indicating that no vehicle matching the vehicle identifier (or image feature) M has been identified. The central server then marks the third electronic tag as "to be observed" and continues to send search commands to each edge device. This allows each edge device to determine, within a set extended time period, whether it can find a vehicle matching the vehicle identifier (or image feature) M using its RFID subsystem and image acquisition equipment. If no vehicle is found, the third tag is deemed invalid.

[0124] Through this embodiment, the central server can promptly determine whether each electronic tag has become invalid, so as to promptly remind staff to replace the electronic tags, thereby ensuring that all electronic tags can work normally and preventing any impact on the management of electric vehicles.

[0125] Figure 6 This is a schematic diagram of the structure of the central server shown in an embodiment of this application. Figure 6In the core layer, the perception layer uses readers and antennas deployed in various sub-regions to acquire the identification of electronic tags. Combined with high-definition intelligent network cameras with built-in AI algorithm chips, it automatically collects image features and license plate numbers of electric vehicles and performs intrusion analysis. Simultaneously, it works with charging pile sensors and building entrance sensors to achieve comprehensive perception of the community environment. The data storage layer manages feature profiles containing electronic tag identification, vehicle identification, and visual characteristic information through an electric bicycle database. A real-time data storage module ensures high-speed exchange of perceived data, while a file storage module is responsible for the long-term retention of images, videos, and system logs. The core service layer serves as the decision-making core. Its registration service drives the seamless and automated registration of electronic tags and the improvement of feature profiles. The verification service performs multi-modal data fusion verification based on a deep collaborative algorithm. The positioning and tracking service enables all-time and all-space trajectory monitoring of electric vehicles. The building entry monitoring service accurately identifies and intercepts unauthorized entry into elevators or corridors. The charging management service ensures energy utilization safety. The algorithm analysis service is responsible for running a random forest model for conflict arbitration and using a federated average algorithm to aggregate and update model parameters. The business application layer displays vehicle identification and attribute details through a vehicle management panel, aggregates the overall operational status of the area on a real-time monitoring screen, the building entry monitoring system is responsible for closed-loop early warning and handling of violations, the charging management system realizes safe control of charging equipment, and the reporting and analysis modules provide multi-dimensional data support for management decisions. The user interface layer provides property administrators with a manual review and management entry point for work orders, vehicle owners with an authorization interface for vehicle binding requests, security personnel with a platform for real-time alarm push notifications, and system administrators with a management channel for global configuration and equipment maintenance. Through the central server in this embodiment, electric vehicles within the community can be effectively managed, significantly improving the quality of electric vehicle management.

[0126] In summary, the solution presented in this application has at least the following technical effects: First, this application constructs an environment-adaptive confidence calculation model and conflict arbitration mechanism, and introduces a random forest model and a horizontal federated learning architecture. This can effectively solve the conflict of heterogeneous perception results in complex environments such as multi-vehicle parallel operation, metal occlusion, and poor lighting, ensuring high accuracy of electric vehicle identification and meeting the needs of precise management in complex scenarios.

[0127] Second, this application is based on the deep integration of the radio frequency identification subsystem and the visual perception subsystem, which deeply associates the identity of the electronic tag with the image features, behavior analysis and dwell status of the target vehicle. This enables accurate identification of violations (such as unauthorized entry into buildings) and rapid location of the vehicle owner's identity, greatly improving the pertinence and efficiency of alarm processing.

[0128] Third, by utilizing ResNet50 feature extraction and virtual tag generation technology, seamless and automated registration is achieved. Coupled with electronic heartbeat monitoring and visual reverse verification mechanisms, the central server can detect the damage or removal of electronic tags in real time, effectively solving problems such as the cumbersome traditional manual registration and vehicles leaving management.

[0129] Fourth, relying on the three-layer architecture of radio frequency identification subsystem + visual perception subsystem + central management platform, it can realize dynamic monitoring of electric vehicles from entry, parking, violation detection, charging to departure, fill the gaps in the management process of existing solutions, and comprehensively enhance the scientific and proactive nature of community security management.

[0130] The vehicle identification device provided in the embodiments of this application is described below. The vehicle identification device described below and the vehicle identification method described above can be referred to each other.

[0131] This application provides a vehicle recognition device applied to edge devices. Figure 7 This is a structural block diagram of a vehicle identification device illustrated in an embodiment of this application. Figure 7 As shown, the vehicle identification device of this application may include: The first acquisition module 701 is used to acquire the identity of the electronic tag carried by the target vehicle and the image features of the target vehicle when it receives a command to start vehicle recognition sent by the central server. The first determining module 702 is used to determine the first identifier of the target vehicle based on the identity identifier, and to determine the second identifier of the target vehicle based on the image features; The second acquisition module 703 is used to acquire, when the first identifier and the second identifier are different, a first confidence level corresponding to the first identifier, a second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier traveling together, and a third identifier of the target vehicle determined based on a random forest model. Update module 704 is used to update the first confidence level and the second confidence level based on the similarity, the peer probability and the third identifier; The second determining module 705 is used to determine the vehicle identifier of the target vehicle based on the updated first confidence level and second confidence level.

[0132] According to the vehicle recognition device 700 provided in this application, the update module 704 is used to: increase the first confidence level when the similarity is greater than the similarity threshold, or when the third identifier is the same as the first identifier; and increase the second confidence level when the probability of being on the same side is greater than the probability threshold, or when the third identifier is the same as the second identifier.

[0133] According to the vehicle identification device 700 provided in this application, the second determining module 705 is used to: determine the identifier corresponding to the maximum value of the updated first confidence level and the second confidence level as the vehicle identifier of the target vehicle; or, if the maximum value is greater than the confidence level threshold, determine the identifier corresponding to the maximum value as the vehicle identifier of the target vehicle.

[0134] According to the vehicle identification device 700 provided in this application, the first acquisition module 701 is used to: receive the identity identifier of the electronic tag carried by the target vehicle sent by the reader / writer; and the image features of the target vehicle sent by the image acquisition device.

[0135] According to the vehicle identification device 700 provided in this application, the second acquisition module 703 is used to: acquire the signal strength of the electronic tag, the signal strength threshold of the electronic tag, the total number of electronic tags within the recognition range of the reader, the number of interfering electronic tags within the recognition range of the reader, the illumination intensity of the environment where the image acquisition device is located, and the occlusion rate of the target vehicle's identification mark; determine the first confidence level based on the signal strength, the signal strength threshold, the total number of electronic tags, the number of interfering electronic tags, the illumination intensity, and the occlusion rate; and determine the difference between the target value and the first confidence level as the second confidence level.

[0136] According to the vehicle identification device 700 provided in this application, the second acquisition module 703 is specifically used to: determine an initial first confidence level based on the signal strength, the signal strength threshold, the total number of electronic tags, and the number of interfering electronic tags; and increase the initial first confidence level to obtain the first confidence level when the illumination intensity is less than the light intensity threshold or the occlusion rate is greater than the occlusion rate threshold.

[0137] According to the vehicle identification device 700 provided in this application, the radio frequency link characteristics between the reader and the electronic tag, the visual perception characteristics of the image acquisition device, the spatiotemporal environment characteristics of the target vehicle, and the vehicle identification prior characteristics are obtained; the radio frequency link characteristics, the visual perception characteristics, the spatiotemporal environment characteristics, and the vehicle identification prior characteristics are input into a random forest model as input features to obtain the third identifier; wherein, the random forest model is trained based on multiple input feature samples and the corresponding electronic tags of each input feature sample, the random forest model includes multiple decision trees, each decision tree is used to output the identifier prediction result of the target vehicle, and the third identifier is determined based on the identifier prediction result with the most outputs from the decision trees.

[0138] According to the vehicle identification device 700 provided in this application, the radio frequency link features include: the signal strength of the electronic tag, the signal stability of the electronic tag, and the number of interfering electronic tags within the identification range of the reader / writer; the visual perception features include: the visual confidence level of the second identifier, the illumination intensity, and the occlusion rate of the target vehicle's identifier; the spatiotemporal environment features include: environment type, time period features, weather features, and the acquisition time difference between the identification of the electronic tag and the image features of the target vehicle; the vehicle identification prior features include: the historical matching rate between the identification of the electronic tag and the first identifier, and the probability of vehicles traveling together with vehicles corresponding to the first identifier and the second identifier.

[0139] The vehicle recognition device 700 provided in this application further includes: a model update module, configured to: receive a model parameter update message sent by the central server, wherein the model parameter update message includes the target parameters of the random forest model, wherein the target parameters are obtained by the central server processing the model parameters of the random forest model reported by each edge device using a federated averaging algorithm; and update the model parameters of the random forest model according to the target parameters.

[0140] The vehicle identification device 700 provided in this application further includes: a third determining module, used to: determine the first identifier or the second identifier as the vehicle identifier of the target vehicle when the first identifier and the second identifier are the same.

[0141] This application also provides a vehicle identification device applied to a central server. Figure 8 This is a structural block diagram of another vehicle identification device shown in an embodiment of this application. For example... Figure 8 As shown, the vehicle identification device of this application may include: The sending module 801 is used to send a command to the edge device to start vehicle recognition; The command to initiate vehicle identification instructs the edge device to: upon receiving the command from the central server, acquire the identity of the electronic tag carried by the target vehicle and the image features of the target vehicle; determine a first identifier of the target vehicle based on the identity, and determine a second identifier of the target vehicle based on the image features; if the first identifier and the second identifier are different, acquire a first confidence level corresponding to the first identifier, a second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of traveling together between the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier, and a third identifier of the target vehicle determined based on a random forest model; update the first confidence level and the second confidence level based on the similarity, the probability of traveling together, and the third identifier; and determine the true vehicle identifier of the target vehicle based on the updated first confidence level and second confidence level.

[0142] The vehicle identification device 800 provided in this application further includes: a first registration module, configured to: determine a first electronic tag for a vehicle identification to be improved; obtain image features of a first vehicle to which the first electronic tag belongs; determine the vehicle identification of the first vehicle based on the image features of the first vehicle; and associate and store the first electronic tag with the vehicle identification of the first vehicle; or, obtain multiple dimensions of the vehicle body features of the first vehicle based on the image features of the first vehicle; generate a virtual vehicle identification based on the multiple dimensions of the vehicle body features; and associate and store the first electronic tag with the virtual vehicle identification.

[0143] The vehicle identification device 800 provided in this application further includes: a second registration module, used for: determining a second electronic tag for identifying the identity of the vehicle owner to be identified; determining the identity of a candidate vehicle owner based on the time when the second vehicle to which the second electronic tag belongs enters the target area and the vehicle owner's passage records at the entrance and exit of the target area; sending a vehicle binding request to each of the candidate vehicle owners; and, upon receiving a response message agreeing to the vehicle binding request, associating and storing the identity of the candidate vehicle owner who sent the response message agreeing to the vehicle binding request with the identity of the second electronic tag.

[0144] The vehicle identification device 800 provided in this application further includes: a monitoring module, configured to: send a command to each reader in the target area to initiate heartbeat monitoring, wherein the command to initiate heartbeat monitoring is used to instruct the reader to send a heartbeat message to the electronic tag, and to determine whether the electronic tag is in normal working condition based on the response message of the heartbeat message.

[0145] The vehicle identification device 800 provided in this application further includes: a verification module, configured to: receive the identity identifier of a third electronic tag to be verified as invalid from a reader in the target area; obtain a vehicle identifier or image feature associated with the identity identifier of the third electronic tag; and if, within a preset time period, a message is received from an edge device in the target area indicating that a vehicle matching the vehicle identifier or image feature has been identified, determine that the third electronic tag is invalid.

[0146] Figure 9 This is a schematic diagram of the physical structure of an electronic device shown in an embodiment of this application, such as... Figure 9 As shown, the electronic device may include a processor 910, a communication interface 920, a memory 930, and a communication bus 940, wherein the processor 910, the communication interface 920, and the memory 930 communicate with each other via the communication bus 940. The processor 910 can call a computer program in the memory 930 to execute the steps of a vehicle identification method.

[0147] Furthermore, the logical instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0148] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the steps of a vehicle identification method provided in the above embodiments.

[0149] On the other hand, embodiments of this application also provide a processor-readable storage medium storing a computer program for causing a processor to execute the steps of a vehicle identification method provided in the above embodiments.

[0150] The processor-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0151] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0152] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A vehicle identification method characterized by, Applied to edge devices, the method includes: Upon receiving a command from the central server to initiate vehicle identification, the system acquires the identification of the electronic tag carried by the target vehicle and the image features of the target vehicle. The first identifier of the target vehicle is determined based on the identity identifier, and the second identifier of the target vehicle is determined based on the image features; If the first identifier and the second identifier are different, obtain the first confidence level corresponding to the first identifier, the second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier traveling together, and the third identifier of the target vehicle determined based on the random forest model. The first confidence level and the second confidence level are updated based on the similarity, the peer probability, and the third identifier. The true vehicle identifier of the target vehicle is determined based on the updated first and second confidence levels.

2. The vehicle identification method according to claim 1, characterized in that, The step of updating the first confidence level and the second confidence level based on the matching score, the peer probability, and the third identifier includes: If the similarity is greater than the similarity threshold, or if the third identifier is the same as the first identifier, the first confidence level is increased. If the probability of being in the same category is greater than the probability threshold, or if the third identifier is the same as the second identifier, the second confidence level is increased.

3. The vehicle identification method according to claim 1, characterized in that, The step of determining the true vehicle identifier of the target vehicle based on the updated first confidence level and second confidence level includes: The identifier corresponding to the maximum value of the updated first confidence level and second confidence level is determined as the true vehicle identifier of the target vehicle; or... If the maximum value is greater than the confidence threshold, the identifier corresponding to the maximum value is determined as the real vehicle identifier of the target vehicle.

4. The vehicle identification method according to any one of claims 1-3, characterized in that, Obtain the identification of the electronic tag carried by the target vehicle and the image features of the target vehicle, including: The identification of the target vehicle is obtained from the electronic tag carried by the reader; and the image features of the target vehicle are obtained from the image acquisition device.

5. The vehicle identification method according to claim 4, characterized in that, The step of obtaining the first confidence level corresponding to the first identifier and the second confidence level corresponding to the second identifier includes: The system acquires the signal strength of the electronic tag, the signal strength threshold of the electronic tag, the total number of electronic tags within the recognition range of the reader, the number of interfering electronic tags within the recognition range of the reader, the light intensity of the environment where the image acquisition device is located, and the occlusion rate of the target vehicle's markings. The first confidence level is determined based on the signal strength, the signal strength threshold, the total number of electronic tags, the number of interfering electronic tags, the light intensity, and the occlusion rate. The difference between the target value and the first confidence level is determined as the second confidence level.

6. The vehicle identification method according to claim 5, characterized in that, Determining the first confidence level based on the signal strength, the signal strength threshold, the total number of electronic tags, the number of interfering electronic tags, the light intensity, and the occlusion rate includes: An initial first confidence level is determined based on the signal strength, the signal strength threshold, the total number of electronic tags, and the number of interfering electronic tags; If the light intensity is less than the light intensity threshold, or if the occlusion rate is greater than the occlusion rate threshold, the initial first confidence level is increased to obtain the first confidence level.

7. The vehicle identification method according to claim 4, characterized in that, The third identifier is determined through the following steps: The radio frequency link characteristics between the reader and the electronic tag, the visual perception characteristics of the image acquisition device, the spatiotemporal environmental characteristics of the target vehicle, and the prior vehicle recognition characteristics are obtained. The radio frequency link features, the visual perception features, the spatiotemporal environment features, and the vehicle identification prior features are used as input features to input the random forest model to obtain the third identifier; The random forest model is trained based on multiple input feature samples and corresponding electronic tags for each input feature sample. The random forest model includes multiple decision trees, each of which outputs a prediction result for the identification of the target vehicle. The third identification is determined based on the identification prediction result with the most outputs from the decision trees.

8. The vehicle identification method according to claim 7, characterized in that, The radio frequency link features include: the signal strength of the electronic tag, the signal stability of the electronic tag, and the number of interfering electronic tags within the reader's recognition range; the visual perception features include: the visual confidence level of the second identifier, the illumination intensity, and the occlusion rate of the target vehicle's identifier; the spatiotemporal environment features include: environment type, time period features, weather features, and the time difference between the acquisition of the electronic tag's identity identifier and the image features of the target vehicle; the vehicle identification prior features include: the historical matching rate between the electronic tag's identity identifier and the first identifier, and the probability of vehicles traveling together with each other corresponding to the first identifier and the second identifier.

9. The vehicle identification method according to claim 4, characterized in that, Also includes: The system receives a model parameter update message sent by the central server. The model parameter update message includes the target parameters of the random forest model. The target parameters are obtained by the central server processing the model parameters of the random forest model reported by each edge device using a federated averaging algorithm. The model parameters of the random forest model are updated based on the target parameters.

10. A vehicle identification method, characterized in that, Applied to a central server, the method includes: Send a command to the edge device to initiate vehicle recognition; The command to initiate vehicle identification instructs the edge device to: upon receiving the command from the central server, acquire the identity of the electronic tag carried by the target vehicle and the image features of the target vehicle; determine a first identifier of the target vehicle based on the identity, and determine a second identifier of the target vehicle based on the image features; if the first identifier and the second identifier are different, acquire a first confidence level corresponding to the first identifier, a second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of traveling together between the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier, and a third identifier of the target vehicle determined based on a random forest model; update the first confidence level and the second confidence level based on the similarity, the probability of traveling together, and the third identifier; and determine the true vehicle identifier of the target vehicle based on the updated first confidence level and second confidence level.

11. The method according to claim 10, characterized in that, Also includes: Identify the first electronic tag for the vehicle identification system that needs improvement. Obtain the image features of the first vehicle to which the first electronic tag belongs; The vehicle identifier of the first vehicle is determined based on the image features of the first vehicle, and the first electronic tag is associated with and stored with the vehicle identifier of the first vehicle; or, Based on the image features of the first vehicle, obtain the body features of the first vehicle in multiple dimensions, generate a virtual vehicle identifier based on the body features in multiple dimensions, and associate and store the first electronic tag with the virtual vehicle identifier.

12. The method according to claim 10, characterized in that, Also includes: A second electronic tag for identifying the vehicle owner, which needs to be improved; The identity of the candidate vehicle owner is determined based on the time when the second vehicle to which the second electronic tag belongs enters the target area and the vehicle owner's passage records at the entrance and exit of the target area; Send vehicle binding requests to each of the aforementioned candidate vehicle owners; Upon receiving a response message agreeing to the vehicle binding request, the identity of the candidate vehicle owner who sent the response message agreeing to the vehicle binding request is associated with and stored with the identity of the second electronic tag.

13. The method according to claim 10, characterized in that, Also includes: A command to initiate heartbeat monitoring is sent to each reader in the target area. The command to initiate heartbeat monitoring is used to instruct the reader to send a heartbeat message to the electronic tag and to determine whether the electronic tag is in normal working condition based on the response message of the heartbeat message.

14. The method according to claim 10, characterized in that, Also includes: Receive the identity information of a third electronic tag sent by a reader in the target area to verify whether it is invalid; Obtain vehicle identification or image features associated with the identity of the third electronic tag; If, within a preset time period, a message is received from an edge device within the target area indicating that a vehicle matching the vehicle identifier or image features has been identified, the third electronic tag is determined to be invalid.

15. A vehicle identification device, characterized in that, Applied to edge devices, the device includes: The first acquisition module is used to acquire the identity of the electronic tag carried by the target vehicle and the image features of the target vehicle when it receives a command to start vehicle recognition sent by the central server. The first determining module is used to determine a first identifier of the target vehicle based on the identity identifier, and to determine a second identifier of the target vehicle based on the image features; The second acquisition module is used to acquire, when the first identifier and the second identifier are different, a first confidence level corresponding to the first identifier, a second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier traveling together, and a third identifier of the target vehicle determined based on a random forest model. The update module is used to update the first confidence level and the second confidence level based on the similarity, the peer probability, and the third identifier; The second determining module is used to determine the true vehicle identifier of the target vehicle based on the updated first confidence level and second confidence level.

16. A vehicle identification device, characterized in that, The device, applied to a central server, includes: The sending module is used to send commands to edge devices to initiate vehicle recognition; The command to initiate vehicle identification instructs the edge device to: upon receiving the command from the central server, acquire the identity of the electronic tag carried by the target vehicle and the image features of the target vehicle; determine a first identifier of the target vehicle based on the identity, and determine a second identifier of the target vehicle based on the image features; if the first identifier and the second identifier are different, acquire a first confidence level corresponding to the first identifier, a second confidence level corresponding to the second identifier, the similarity between the historical image features of the vehicle corresponding to the first identifier and the image features of the target vehicle, the probability of traveling together between the vehicle corresponding to the first identifier and the vehicle corresponding to the second identifier, and a third identifier of the target vehicle determined based on a random forest model; update the first confidence level and the second confidence level based on the similarity, the probability of traveling together, and the third identifier; and determine the true vehicle identifier of the target vehicle based on the updated first confidence level and second confidence level.

17. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the vehicle identification method according to any one of claims 1 to 9, or the steps of the vehicle identification method according to any one of claims 10 to 14.

18. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a vehicle identification method as described in any one of claims 1 to 9, or a vehicle identification method as described in any one of claims 10 to 14.

19. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the vehicle identification method according to any one of claims 1 to 9, or the steps of the vehicle identification method according to any one of claims 10 to 14.

Images