Driver face recognition and vehicle real-time binding method based on ground induction coil triggering

Abstract

The present application relates to the technical field of artificial intelligence, in particular to a driver face recognition and vehicle real-time binding method based on ground inductor coil triggering, comprising the following steps: when a vehicle triggers a ground inductor coil buried under a lane, a trigger signal is sent to a face collection device; the face collection device responds to the trigger signal and continuously collects multiple frames of driver face images within a preset time window; one frame of driver face image meeting a preset quality condition is selected from the multiple frames of driver face images as a target face image; face feature extraction is performed on the target face image to generate a driver face feature code; vehicle information of a current vehicle is acquired, and the driver face feature code and the vehicle information are used as bidirectional index keys respectively to establish a bidirectional association index between the driver and the vehicle and generate a driver-vehicle binding record. The trigger signal is generated when the vehicle reaches the ground inductor coil, which ensures that the collection action is performed when the vehicle is in the best collection position.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method for real-time binding of driver facial recognition and vehicle based on inductive loop triggering. Background Technology

[0002] In the management of passenger transport lanes at ports, verifying the identities of passing vehicles and drivers and linking vehicles to their registration is a crucial aspect of safety control. Traditional methods rely primarily on manual verification, where staff visually inspect driver's documents and compare vehicle information. This is not only inefficient but also prone to errors and omissions due to fatigue or negligence. With the continuous increase in port traffic volume, manual verification methods can no longer meet the dual requirements of both clearance efficiency and safety control.

[0003] In the exploration of automated recognition solutions, facial recognition of the driver is required after a vehicle enters the lane. However, the driver is located inside the driver's cab, and the facial recognition conditions are complex due to factors such as reflections from the car window, changes in lighting conditions, and vehicle movement. Therefore, the accuracy of the timing of the capture is crucial. If the capture is too early, the vehicle has not yet reached the optimal capture position, and the driver's face is off-center from the field of view of the capture device. If the capture is too late, the vehicle has already passed the optimal capture position, resulting in the driver's face appearing too small or at too large an angle in the captured image.

[0004] Furthermore, in actual traffic scenarios, the same driver may drive different vehicles on different dates, and the same vehicle may be driven by different drivers. If only a temporary association is made between the driver and the vehicle for each passage, and each passage record is independent of the others, it is impossible to query which vehicles a driver has driven in the past from the driver's perspective, or which drivers have driven that vehicle from the vehicle's perspective. Moreover, it is impossible to determine the degree of binding between the driver and the vehicle based on the frequency of passage, thus failing to provide effective data support for port risk analysis and screening of key personnel.

[0005] In summary, existing technologies have the following problems in automating vehicle and driver verification: First, they lack a mechanism to precisely control the timing of data collection and acquire high-quality driver facial images at the optimal location; second, they lack a technical solution to establish a persistent, bidirectionally searchable, and dynamically updated vehicle-driver association index based on traffic behavior. Summary of the Invention

[0006] The purpose of this invention is to propose a method for real-time binding of driver face recognition and vehicle based on inductive loop triggering, which at least solves one of the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for real-time binding of driver facial recognition with vehicle based on inductive loop detectors includes the following steps:

[0009] When a vehicle triggers the inductive loop buried under the lane, it sends a trigger signal to the face recognition device.

[0010] The face acquisition device responds to the trigger signal and continuously acquires multiple frames of driver face images within a preset time window;

[0011] Select one frame from the multiple driver face images that meets the preset quality conditions as the target face image;

[0012] Facial features are extracted from the target face image to generate a driver's facial feature code;

[0013] Obtain the vehicle information of the current vehicle, and use the driver's facial feature code and the vehicle information as bidirectional index keys to establish a bidirectional association index between the driver and the vehicle, and generate a human-vehicle binding record.

[0014] A further improvement is that the preset time window is 100 milliseconds to 300 milliseconds; within the preset time window, the face acquisition device continuously acquires 3 to 10 frames of driver face images at a preset frame rate.

[0015] A further improvement is that the specific method for selecting a frame that meets the preset quality conditions from the multiple frames of driver face images as the target face image includes: performing face detection on the multiple frames of driver face images respectively, locating the face region in each frame image and calculating the area ratio of the face region in the whole frame image, filtering out image frames in which no face region is detected or the area ratio of the face region is lower than a preset threshold; calculating the clarity score of each frame image remaining after filtering, and selecting the frame with the highest clarity score as the target face image.

[0016] A further improvement is that the face detection employs a face detection algorithm based on a multi-task cascaded convolutional neural network. This algorithm processes the image sequentially through three cascaded neural network stages: the first stage, a fast proposal network, rapidly scans the image, generating a large number of candidate face windows; the second stage, a fine-tuning network, refines these candidate face windows, filtering out non-face windows; and the third stage, an output network, makes the final determination on the remaining windows, outputting the bounding box coordinates of the face region and the positions of facial feature points. Based on the bounding box coordinates, the area ratio of the face region in the entire frame is calculated, and image frames with a face region area ratio lower than a preset threshold are filtered out.

[0017] A further improvement is that obtaining the vehicle information of the current vehicle specifically includes: collecting the license plate number of the current vehicle through a license plate recognition system, and using the license plate number as the vehicle information of the current vehicle.

[0018] A further improvement is that the specific method for establishing a bidirectional association index between the driver and the vehicle, using the driver's facial feature code and the vehicle information as bidirectional index keys respectively, includes:

[0019] Create a first index table in the database with the driver's facial feature code as the first index key. The first index table stores all vehicle information bound to the driver.

[0020] Create a second index table in the database with vehicle information as the second index key. The second index table stores the facial feature codes of all drivers associated with the vehicle.

[0021] The first index table and the second index table enable bidirectional retrieval between the driver and the vehicle.

[0022] A further improvement is that the specific method for establishing a bidirectional association index between the driver and the vehicle, using the driver's facial feature code and the vehicle information as bidirectional index keys respectively, also includes dynamically updating the weights of the bidirectional association index:

[0023] An association weight value is set for the binding relationship between the driver and the vehicle. The association weight value is used to characterize the tightness of the binding relationship between the driver and the vehicle. The larger the weight value, the tighter the binding relationship.

[0024] Whenever the same driver and the same vehicle are successfully bound together, the association weight value of the binding relationship is incremented by 1.

[0025] The binding strength between the driver and the vehicle is divided into several preset levels based on the association weight value.

[0026] A further improvement is that the specific method for establishing a bidirectional association index between the driver and the vehicle using the driver's facial feature code and the vehicle information as bidirectional index keys also includes a step of time decay of the association weight value: recording the time of the most recent successful binding of each binding relationship; when the current time is more than the time of the most recent successful binding exceeds a preset time decay period, the association weight value decreases according to a preset decay magnitude.

[0027] A further improvement is that the method further includes: in response to a retrieval request indexed by the driver's facial feature code, querying and outputting all vehicle information associated with the driver through the bidirectional association index; and in response to a retrieval request indexed by vehicle information, querying and outputting all drivers associated with the vehicle through the bidirectional association index.

[0028] A further improvement is that the method further includes: receiving a deployment instruction, the deployment instruction including the facial feature code of the target driver; loading the facial feature code of the target driver into the deployment list of the checkpoint system; when the facial feature code extracted from the collected driver's facial image successfully matches the target facial feature code in the deployment list, generating an interception instruction and sending it to the checkpoint gate control system to trigger the gate to perform an interception operation.

[0029] The beneficial effects of this invention are as follows:

[0030] This invention achieves precise sensing of a vehicle's arrival at the optimal sampling location by embedding an inductive loop at the optimal position within the lane. It utilizes the principle that the change in magnetic flux through the loop caused by a vehicle passing over it alters the oscillation frequency, thus enabling accurate detection of the vehicle's arrival at the optimal sampling location. When the vehicle reaches the sampling location, the inductive loop immediately generates a trigger signal, and the face recognition device begins sampling under the excitation of the trigger signal, ensuring that the sampling action is performed when the vehicle is in the optimal sampling position.

[0031] This invention improves the success rate of face recognition by controlling a face acquisition device to continuously acquire multiple frames of driver face images within a preset time window after the ground induction coil is triggered, and performing face detection on the multiple frames to filter out invalid frames with no face or too low face proportion. Then, a clarity score is calculated based on the image gradient amplitude, and the best frame is selected as the target face image.

[0032] This invention establishes a bidirectional relational index structure in the database by using the driver's facial feature code and vehicle information as bidirectional index keys. This enables bidirectional retrieval capabilities, allowing the driver to query all vehicles they have ever driven and the vehicle to query all drivers who have driven that vehicle. This establishes a persistent association between each passage record and provides a foundation for the in-depth utilization of human-vehicle relationship data.

[0033] Furthermore, this invention introduces a dynamic weight update mechanism, automatically accumulating association weights based on the frequency of driver and vehicle traffic binding, thus achieving intelligent hierarchical management of human-vehicle relationships. Simultaneously, by introducing a time decay mechanism, the binding relationships can automatically fade and be cleared over time, ensuring that the system always reflects the latest and most active human-vehicle association status. Attached Figure Description

[0034] Figure 1This is a flowchart of a method for real-time binding of driver face recognition and vehicle based on inductive loop triggering according to the present invention. Detailed Implementation

[0035] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. The described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0036] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0037] Please refer to the attached document. Figure 1 This invention proposes a method for real-time driver face recognition and vehicle binding based on inductive loop triggering, which is applied to a port passenger lane management system. The system includes an inductive loop buried under the lane surface, a face acquisition device erected above or to the side of the lane, an edge computing unit connected to the inductive loop and the face acquisition device, and a checkpoint management system and a vehicle binding database deployed in the background.

[0038] The inductive loop is a vehicle detection sensor based on the principle of electromagnetic induction, embedded under the lane as part of an LC oscillation circuit. When a vehicle passes, the iron component at the bottom of the vehicle changes the magnetic flux of the coil, causing a change in the oscillation frequency, thereby detecting the vehicle's presence. In this embodiment, the inductive loop is embedded at the optimal acquisition position in the lane, specifically at the point where the optical axis of the face acquisition device intersects perpendicularly with the driver's normal sitting position. The face acquisition device is a high-definition network camera equipped with an autofocus lens and a supplementary lighting module, used to acquire a frontal facial image of the driver. The edge computing unit is a computing device deployed at the lane site, responsible for receiving the inductive signal from the inductive loop, controlling the face acquisition device, and performing preliminary processing for facial feature extraction and vehicle-person binding. The checkpoint management system is a backend management platform responsible for storing vehicle-person binding records, managing the control list, and providing query and retrieval services.

[0039] Specifically, the method includes the following steps:

[0040] Step S1: When the vehicle triggers the ground loop coil buried under the lane, a trigger signal is sent to the face recognition device.

[0041] Step S2: In response to the trigger signal, the face acquisition device continuously acquires multiple frames of driver face images within a preset time window.

[0042] Step S3: Select one frame from the multi-frame driver face images that meets the preset quality conditions as the target face image.

[0043] Step S4: Extract facial features from the target face image to generate a driver's facial feature code.

[0044] Step S5: Obtain the vehicle information of the current vehicle, and use the driver's facial feature code and the vehicle information as bidirectional index keys to establish a bidirectional association index between the driver and the vehicle, and generate a human-vehicle binding record.

[0045] The method steps of the present invention will be described in detail below:

[0046] Specifically, in step S1, the inductive loop is buried at the optimal collection location within the lane. In a port checkpoint scenario, lane directions are fixed; vehicles in the inbound lane can only enter, and vehicles in the outbound lane can only exit. Therefore, the direction of travel is determined by the physical properties of the lane and does not require device identification. When a vehicle reaches the collection location, the iron component at the bottom of the vehicle changes the magnetic flux of the inductive loop, causing a change in the oscillation frequency of the LC oscillation circuit. After detecting this frequency change, the edge computing unit determines that the vehicle has reached the optimal collection location. The edge computing unit then sends a trigger signal to the face recognition device via a signal line or network protocol. The trigger signal is a high-level pulse signal or a command message based on a network protocol, and includes a timestamp and lane number.

[0047] Specifically, in step S2, after receiving the trigger signal, the face acquisition device immediately activates the image sensor and continuously acquires multiple frames of driver face images at a preset frame rate within a preset time window. The preset time window is from 100 milliseconds to 300 milliseconds.

[0048] Understandably, if the time window is too short (less than 100 milliseconds), the number of image frames that can be acquired is insufficient (at 30 frames per second, only 3 frames can be acquired in 100 milliseconds), failing to effectively cover different facial postures of the driver. If the time window is too long (greater than 300 milliseconds), some acquired images may have deviated from the optimal acquisition position, increasing the burden of data storage and processing. In this embodiment, the frame rate is 30 frames per second, therefore 3 to 10 images can be acquired within a time window of 100 to 300 milliseconds. The start time of the time window is the time when the trigger signal is received, and the end time is 100 to 300 milliseconds after the start time. By continuously acquiring multiple frames of images, adverse factors such as motion blur, driver blinking, facial expression changes, and head turning that may exist in a single frame image can be effectively addressed.

[0049] During the data acquisition process, the supplementary lighting module of the face capture device automatically adjusts the supplementary lighting intensity based on the detection results of the ambient light sensor, ensuring sufficient image brightness even in low-light conditions. The supplementary lighting module uses a combination of infrared and visible light supplementary lighting, automatically switching to infrared supplementary lighting mode at night or in low light conditions to avoid glare interference from visible light supplementary lighting for the driver.

[0050] Specifically, in step S3, selecting a frame that meets a preset quality condition from the multiple frames of driver face images as the target face image includes the following sub-steps:

[0051] First, face detection is performed on the multiple frames of driver face images, the face region in each frame is located and the area ratio of the face region in the whole frame is calculated, and image frames in which no face region is detected or the area ratio of the face region is lower than a preset threshold are filtered out.

[0052] In a preferred embodiment, the face detection employs a face detection algorithm based on a multi-task cascaded convolutional neural network (MTCNN). This algorithm processes the image sequentially through three cascaded neural network stages. Specifically:

[0053] The first stage, the Fast Proposal Network (P-Net), scans each layer of the image pyramid, extracting features from each layer using a 3×3 sliding window. It outputs candidate face windows and their confidence scores through fully connected layers. Non-maximum suppression is applied to windows with confidence scores higher than a preset first threshold (e.g., 0.6). The coordinates of the filtered windows are mapped back to the original image size, outputting an initial set of candidate face windows. The image pyramid is constructed by continuously scaling the original image with a scaling factor of 0.709, resulting in a minimum image size of 20×20 pixels. Non-maximum suppression is used to merge highly overlapping candidate windows, with an Intersection over Union (IoU) threshold of 0.5. This means that when the ratio of the intersection area to the union area of ​​two candidate windows is greater than 0.5, the window with the higher confidence score is retained, while the window with the lower score is suppressed.

[0054] The second-stage fine network (R-Net) scales each window in the initial candidate face window set to 24×24 pixels and inputs it into the fine network. The fine network outputs the confidence score and bounding box regression offset of each window through a fully connected layer. Window is filtered according to the confidence score (e.g., confidence score is higher than 0.7), and non-maximum suppression is performed again (IoU threshold is 0.7), outputting a finely filtered candidate face window set.

[0055] The third-stage output network (O-Net) scales each window in the refined candidate face window set to 48×48 pixels and inputs it into the output network. The output network outputs the confidence score, bounding box regression offset, and the positions of five facial feature points (left eye, right eye, nose tip, left corner of mouth, and right corner of mouth) for each window. Windows are filtered based on their confidence scores (e.g., confidence scores higher than 0.9), and non-maximum suppression (IoU threshold of 0.7) is performed. The final bounding box coordinates and facial feature point positions for the face region are then output. The bounding box coordinates include the x-coordinate of the top-left corner, the y-coordinate of the top-left corner, the width, and the height.

[0056] The area percentage of the face region in the entire image frame is calculated based on the bounding box coordinates, and image frames with a face region area percentage below 10% are filtered out. Specifically, the area percentage is calculated by dividing the area of ​​the face bounding box by the area of ​​the entire image frame: Area percentage = (bounding box width × bounding box height) / (image width × image height). Image frames with a face region area percentage below 10% indicate that the driver's face is too small or off-center in the image, making effective facial feature extraction impossible.

[0057] Then, the sharpness score of each remaining frame after filtering is calculated.

[0058] In a preferred embodiment of this invention, the sharpness score is obtained in the following manner:

[0059] The image is converted to grayscale, and a second-order derivative operation is performed on the grayscale image using the Laplacian operator to obtain a response map. The variance of the response map is calculated as the sharpness score of the frame image. The Laplacian operator is a second-order derivative operator used to detect regions of rapid change in grayscale values ​​in an image, and its expression is:

[0060]

[0061] in, The grayscale function represents the grayscale value of the image at coordinates (x, y); grayscale function The second-order partial derivative in the horizontal direction (x direction) is used to detect changes in the rate of grayscale change in the horizontal direction; grayscale function The second-order partial derivative in the vertical direction (y-direction) is used to detect changes in the rate of grayscale change in the vertical direction; Represents the Laplace operator for grayscale functions The combined effect reflects the degree of drastic local grayscale changes in the image in two-dimensional space. In actual calculations, a discrete approximation is used, i.e., a 3×3 Laplacian kernel K=[0, 1, 0; 1, -4, 1; 0, 1, 0] is used to convolve the grayscale image. The convolution operation slides the Laplacian kernel pixel by pixel on the image, and the response value at each pixel position is the weighted sum of the kernel and the corresponding pixel value in the image region. After convolution, a Laplacian response map with the same size as the original image is obtained. Positive values ​​in the response map indicate that the grayscale value of the point is higher than the neighborhood average (peak), and negative values ​​indicate that it is lower than the neighborhood average (valley). The variance is calculated for all pixel values ​​in the response map. The variance is the average of the squares of the differences between each pixel response value and the mean of the response map, and its calculation formula is: ,in Total number of pixels For the first The response value of each pixel. The variance is the mean of the response image. A larger variance indicates sharper edges and richer details, meaning higher clarity.

[0062] Finally, the frame with the highest clarity score is selected as the target face image. This frame image is then used as input for subsequent face feature extraction.

[0063] It is understandable that this invention improves the success rate of single acquisition by continuously acquiring multiple frames of images after triggering, first filtering out invalid frames through face detection, and then automatically selecting the frame with the best clarity.

[0064] Specifically, in step S4, the edge computing unit or the backend server performs facial feature extraction on the target face image obtained in step S3.

[0065] First, the target face image is preprocessed, including size normalization, grayscale conversion, and histogram equalization, to eliminate the influence of illumination variations and acquisition angle on feature extraction. The preprocessed image is then input into a face feature extraction model to extract a fixed-dimensional face feature vector. This face feature extraction model employs a deep convolutional neural network-based face recognition model, such as ArcFace or CosFace, which maps the input face image into a fixed-length feature vector. In this embodiment, the fixed-length feature vector is a 512-dimensional floating-point vector.

[0066] After extraction, the 512-dimensional floating-point vector is used as the driver's facial feature code and stored in the cache of the edge computing unit in association with metadata such as collection time, collection port, and lane number.

[0067] It should be understood that the facial feature code has the following characteristics: facial images of the same driver collected at different times and under different lighting conditions have a high cosine similarity between the feature vectors obtained after calculation by the facial feature extraction algorithm; while the feature vectors corresponding to facial images of different drivers have a low cosine similarity.

[0068] Specifically, in step S5, obtaining the vehicle information of the current vehicle includes: acquiring the license plate number of the current vehicle through a license plate recognition system. The license plate recognition system captures an image of the front of the vehicle using a camera mounted above the lane, and uses an optical character recognition algorithm to locate and recognize the license plate area in the image, outputting the license plate number. This license plate number is then used as the vehicle information of the current vehicle.

[0069] The step of establishing a bidirectional association index between the driver and the vehicle, using the driver's facial feature code and the vehicle information as bidirectional index keys respectively, specifically includes: creating a first index table in the database with the driver's facial feature code as the first index key. This first index table stores all vehicle information bound to the driver, as well as the association weight value, first binding time, and most recent binding time for each binding relationship. A second index table in the database is created with the vehicle information as the second index key. This second index table stores all driver facial feature codes bound to the vehicle, as well as the binding strength level for each binding relationship. Bidirectional retrieval between the driver and the vehicle is achieved through the first and second index tables.

[0070] Specifically, the bidirectional association index structure supports queries in two dimensions: from the driver's perspective, one facial feature code can be associated with multiple vehicle information, i.e., the scenario of "one person linked to multiple vehicles"; from the vehicle's perspective, one vehicle information can be associated with multiple facial feature codes, i.e., the scenario of "one vehicle linked to multiple people". The bidirectional association index is stored in the person-vehicle binding database in the form of a cross-tab in a relational database.

[0071] The bidirectional association index is stored in the vehicle-person binding database to generate vehicle-person binding records containing binding time, binding port, and direction of travel. The direction of travel is determined by the lane's physical attributes; inbound lanes are recorded as inbound, and outbound lanes are recorded as outbound.

[0072] In a preferred embodiment of the present invention, the method further includes the step of dynamically updating the weights of the bidirectional association index:

[0073] Each time the same driver and the same vehicle are successfully bound together, the association weight value of the binding relationship is increased. The association weight value is a cumulative numerical indicator used to quantify the tightness of the binding between the driver and a specific vehicle. The more frequently binding occurs, the more stable and reliable the association between the driver and the vehicle. Specifically, each successful binding increases the association weight value by one basic unit, meaning the weight value is incremented by 1 from its original value.

[0074] For newly established binding relationships, i.e., when a driver and vehicle are associated for the first time, their association weight value accumulates from an initial value of 0. When the driver drives the same vehicle a second time and successfully verifies via facial recognition, the system triggers the binding operation again. This time, the system detects that a binding record for the same driver and vehicle already exists in the database, so instead of creating a new record, it updates the association weight value from 1 to 2 based on the existing record. This process continues, with each repeated binding incrementing the weight value by 1. Therefore, the association weight value can be understood as "the total number of times the same driver and the same vehicle have been successfully bound in historical passage records."

[0075] In a preferred embodiment of this invention, the binding strength between the driver and the vehicle is divided into several preset levels based on the association weight value.

[0076] When the association weight value exceeds a preset first threshold (e.g., 10 times), the binding strength is determined to be at the first level, and the driver is marked as the primary driver of the vehicle. It is understood that "primary driver" indicates that the driver is the main user of the vehicle.

[0077] When the association weight value is between a preset second threshold (e.g., 3 times) and the first threshold, the binding strength is determined to be at the second level, and the driver is marked as a frequent driver of the vehicle. It is understood that a frequent driver indicates that the driver is a regular user of the vehicle, and their binding relationship is stable but has not yet reached the closeness of a primary driver.

[0078] When the association weight value is lower than a preset second threshold but exceeds a preset third threshold (e.g., once), the binding strength is determined to be at the third level, and the driver is marked as a temporary driver of the vehicle. It is understood that a temporary driver means the driver uses the vehicle only occasionally.

[0079] When the association weight value is not higher than the preset third threshold, that is, only one binding has occurred, the binding strength is determined to be of the occasional level, the driver is not specially marked, and the binding relationship is only retained as a historical record.

[0080] In a preferred embodiment of the present invention, the method further includes a step of time decaying the associated weight value:

[0081] The system records the time of the most recent successful binding for each relationship. When the current time is more than a preset time decay period (30 days in this embodiment), the association weight value decreases according to a preset decay rate (1 unit every 30 days in this embodiment). It is understood that if a driver does not drive a vehicle for a long time, the tightness of the driver-vehicle binding relationship should naturally weaken over time.

[0082] When the association weight value decreases to no higher than a preset third threshold, the corresponding binding relationship is marked as a historical relationship and will no longer be used as an active binding in retrieval. Historical relationships are archived in a history table, which users can view through a separate "History Query" function.

[0083] It is understood that the present invention enables the relationship between people and vehicles to evolve dynamically with the accumulation of traffic behavior through the dynamic weight update mechanism and time decay mechanism. The system automatically identifies the main driver and the temporary driver, while ensuring that the system always reflects the latest and most active relationship between people and vehicles.

[0084] In a preferred embodiment of the present invention, the method further includes the step of performing a reverse retrieval based on the person-vehicle binding record:

[0085] In response to a retrieval request indexed by the driver's facial feature code, the system queries and outputs all vehicle information associated with the driver through the bidirectional association index. For example, when a supervisor enters a driver's facial feature code, the system returns a list of all vehicles the driver has historically driven, including license plate numbers, number of times the vehicle has been bound, the most recent binding time, and the binding strength level.

[0086] In response to a retrieval request indexed by vehicle information, the system queries and outputs the facial feature codes of all drivers associated with the vehicle and their corresponding binding strength levels through the bidirectional association index. For example, when a supervisor enters a license plate number, the system returns the facial feature codes of all drivers who have previously driven the vehicle and their corresponding binding strength levels.

[0087] In a preferred embodiment of the present invention, the method further includes the steps of face detection and coordinated interception:

[0088] The system receives a control command, which includes the facial feature code of the target driver. The control command can be manually issued by the port management department through the back-end management platform, or it can be automatically pushed by the superior system through the API interface.

[0089] The facial feature code of the target driver is loaded into the control list of the checkpoint system. The control list is stored in the cache of the edge computing unit.

[0090] When the facial feature code extracted from the captured driver's facial image successfully matches the target facial feature code in the control list, an interception command is generated and sent to the checkpoint gate control system, triggering the gate to execute the interception operation.

[0091] Specifically, after extracting facial features, the edge computing unit first compares the extracted facial feature codes with the feature codes of each target in the control list. The comparison uses cosine similarity calculation. When the similarity exceeds a preset matching threshold, it is considered a successful match. After a successful match, the edge computing unit immediately sends an interception command to the checkpoint gate control system via a signal line or network protocol. The gate control system responds to the interception command by closing the gate to prevent vehicles from passing through, and at the same time triggers an audible and visual alarm device to alert on-site personnel.

[0092] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. 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; and these 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 the present invention.

Claims

1. A method for real-time binding of driver face recognition and vehicle based on inductive loop triggering, characterized in that, Includes the following steps: When a vehicle triggers the inductive loop buried under the lane, it sends a trigger signal to the face recognition device. The face acquisition device responds to the trigger signal and continuously acquires multiple frames of driver face images within a preset time window; Select one frame from the multiple driver face images that meets the preset quality conditions as the target face image; Facial features are extracted from the target face image to generate a driver's facial feature code; Obtain the vehicle information of the current vehicle, and use the driver's facial feature code and the vehicle information as bidirectional index keys to establish a bidirectional association index between the driver and the vehicle, and generate a human-vehicle binding record.

2. The method for real-time binding of driver face recognition and vehicle based on inductive loop triggering according to claim 1, characterized in that, The preset time window is 100 milliseconds to 300 milliseconds; within the preset time window, the face acquisition device continuously acquires 3 to 10 frames of driver face images at a preset frame rate.

3. The method for real-time driver face recognition and vehicle binding based on inductive loop triggering according to claim 1, characterized in that, The specific method for selecting a frame that meets the preset quality conditions from the multiple frames of driver face images as the target face image includes: performing face detection on the multiple frames of driver face images respectively, locating the face region in each frame image and calculating the area ratio of the face region in the whole frame image, filtering out image frames in which no face region is detected or the area ratio of the face region is lower than a preset threshold; calculating the clarity score of each frame image remaining after filtering, and selecting the frame with the highest clarity score as the target face image.

4. The method for real-time driver face recognition and vehicle binding based on inductive loop triggering according to claim 3, characterized in that, The face detection employs a face detection algorithm based on a multi-task cascaded convolutional neural network. This algorithm processes the image sequentially through three cascaded neural network stages: the first stage, a fast proposal network, rapidly scans the image, generating a large number of candidate face windows; the second stage, a fine-tuning network, refines these candidate face windows, filtering out non-face windows; and the third stage, an output network, makes the final determination of the remaining windows, outputting the bounding box coordinates of the face region and the positions of facial feature points. Based on the bounding box coordinates, the area ratio of the face region in the entire frame is calculated, and image frames with a face region area ratio lower than a preset threshold are filtered out.

5. The method for real-time driver face recognition and vehicle binding based on inductive loop triggering according to claim 1, characterized in that, The process of obtaining the vehicle information of the current vehicle specifically includes: collecting the license plate number of the current vehicle through a license plate recognition system, and using the license plate number as the vehicle information of the current vehicle.

6. The method for real-time driver face recognition and vehicle binding based on inductive loop triggering according to claim 1, characterized in that, The specific method for establishing a bidirectional association index between the driver and the vehicle, using the driver's facial feature code and the vehicle information as bidirectional index keys respectively, includes: Create a first index table in the database with the driver's facial feature code as the first index key. The first index table stores all vehicle information bound to the driver. Create a second index table in the database with vehicle information as the second index key. The second index table stores the facial feature codes of all drivers associated with the vehicle. The first index table and the second index table enable bidirectional retrieval between the driver and the vehicle.

7. The method for real-time driver face recognition and vehicle binding based on inductive loop triggering according to claim 6, characterized in that, The specific method for establishing a bidirectional association index between the driver and the vehicle, using the driver's facial feature code and the vehicle information as bidirectional index keys respectively, further includes dynamically updating the weights of the bidirectional association index: An association weight value is set for the binding relationship between the driver and the vehicle. The association weight value is used to characterize the tightness of the binding relationship between the driver and the vehicle. The larger the weight value, the tighter the binding relationship. Whenever the same driver and the same vehicle are successfully bound together, the association weight value of the binding relationship is incremented by 1. The binding strength between the driver and the vehicle is divided into several preset levels based on the association weight value.

8. The method for real-time binding of driver face recognition and vehicle based on inductive loop triggering according to claim 7, characterized in that, The specific method for establishing a bidirectional association index between the driver and the vehicle using the driver's facial feature code and the vehicle information as bidirectional index keys also includes a step of time decay of the association weight value: recording the time of the most recent successful binding of each binding relationship; when the current time is more than the time of the most recent successful binding exceeds a preset time decay period, the association weight value decreases according to a preset decay range.

9. The method for real-time driver face recognition and vehicle binding based on inductive loop triggering according to claim 1, characterized in that, The method further includes: in response to a retrieval request indexed by the driver's facial feature code, querying and outputting all vehicle information associated with the driver through the bidirectional association index; and in response to a retrieval request indexed by vehicle information, querying and outputting all drivers associated with the vehicle through the bidirectional association index.

10. A method for real-time driver face recognition and vehicle binding based on inductive loop triggering according to claim 1, characterized in that, The method further includes: receiving a deployment instruction, the deployment instruction including the facial feature code of the target driver; loading the facial feature code of the target driver into the deployment list of the checkpoint system; when the facial feature code extracted from the collected driver's facial image successfully matches the target facial feature code in the deployment list, generating an interception instruction and sending it to the checkpoint gate control system to trigger the gate to perform an interception operation.

Images