Intelligent driving vehicle risk early warning method and system based on highway checkpoint

Abstract

The invention discloses an intelligent driving vehicle risk early warning method and system based on highway checkpoints. The method comprises the following steps: S1, acquiring traffic event information and geographic position coordinates on a highway; s2, based on the geographic position coordinates, determining two upstream continuous highway checkpoint devices affected by the traffic event; s3, candidate vehicles with the intelligent driving ability are screened out from the passing vehicles; s4, executing an intelligent driving state judgment process on the candidate vehicles; s5, for the vehicle judged to be in the intelligent driving state, executing a graded early warning reaching process; and S6, through a guidance screen in front of the traffic incident point, synchronously issuing directional text guidance information containing the license plate number of the vehicle, and reminding a driver to turn off the intelligent driving mode. The accident risk caused by insufficient response of the intelligent driving system or untimely taking over of the driver is effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a method and system for risk warning of intelligent driving vehicles based on highway checkpoints. Background Technology

[0002] With the rapid development of intelligent driving technology, vehicles equipped with assisted driving or autonomous driving systems (hereinafter collectively referred to as "intelligent driving systems") are increasingly being used in highway scenarios. Intelligent driving systems rely on sensors such as LiDAR and machine vision to perceive the road environment and achieve a certain degree of autonomous vehicle control.

[0003] However, in high-speed, enclosed driving environments like highways, the time window between the system detecting the risk, issuing a warning, and instructing the driver to take over the vehicle is extremely short, typically only a few seconds, when sudden traffic events such as construction, temporary traffic control, accidents, severe congestion, or inclement weather occur ahead. Within this short time, drivers often struggle to fully understand the risk, assess the situation, and execute effective evasive maneuvers. This significantly increases the risk of secondary accidents such as rear-end collisions and other collisions, and may further lead to serious consequences such as battery fires, posing a serious threat to road safety.

[0004] Currently, existing highway safety early warning technologies mostly focus on detecting traffic events themselves and issuing broadcast warnings. For example, Chinese invention patent CN120913386A discloses a monitoring system for autonomous driving equipment on highways, which uses various sensors to detect events on highways and issue multi-level warnings. While such solutions can replace manual inspections and improve event detection efficiency, their warning information is usually broadcast and not targeted at specific vehicles or driving states. For vehicles operating in intelligent driving mode, broadcast warnings have the following shortcomings: First, it cannot accurately identify specific vehicles that are using intelligent driving functions.

[0005] Secondly, it is impossible to determine whether the driver of the vehicle received the warning information in a timely manner and was prepared to take over.

[0006] Third, the warning information lacks mandatory and interactive features, making it difficult to ensure that drivers can promptly and proactively exit the intelligent driving mode and revert to fully manual driving at critical road sections.

[0007] It is evident that existing technical solutions have significant limitations in directly ensuring the safety of intelligent driving vehicles when passing through high-risk areas, and cannot effectively solve the problem of safety accidents caused by untimely takeover.

[0008] Therefore, there is an urgent need for a method that can utilize existing road infrastructure to accurately identify vehicles in intelligent driving mode and provide proactive and targeted risk warnings and interventions to compensate for the shortcomings of current warning technologies in protecting intelligent driving vehicles and reduce the risk of traffic accidents caused by them. Summary of the Invention

[0009] In a first aspect, to address the aforementioned technical problems, this invention provides a method for risk warning of intelligent driving vehicles based on highway checkpoints, comprising the following steps: S1. Obtain information on traffic incidents on highways and their geographical coordinates; S2. Based on the geographical coordinates of the traffic event and combined with road network data, determine the two consecutive upstream highway checkpoint devices affected by the traffic event through a path analysis algorithm. These devices are a remote checkpoint device and a near-end checkpoint device, respectively. S3. Based on the structured data and images of passing vehicles collected by the remote checkpoint device, the vehicle model in the structured data of passing vehicles is matched with the locally maintained intelligent driving model database; and / or, it is determined whether the length of the vehicle license plate number is the preset length of the intelligent driving license plate, so as to filter out candidate vehicles with intelligent driving capabilities from the passing vehicles. S4. For the candidate vehicles, execute the intelligent driving status judgment process: S41. Based on the YOLO series image recognition algorithm, identify the vehicle image collected by the remote checkpoint device to determine whether the vehicle has turned on the intelligent driving lights. S42. If the intelligent driving light is detected to be on, the vehicle is determined to be in intelligent driving mode; if the intelligent driving light is not detected to be on, the lane centering analysis sub-process is executed; wherein, the lane centering analysis sub-process includes: S42a. Based on the same vehicle image collected by the remote checkpoint device and the near checkpoint device respectively, and the lane calibration data pre-stored by each checkpoint, calculate the lane lateral position offset parameter of the vehicle at the two points of the remote checkpoint device and the near checkpoint device. S42b. Based on the lane lateral position offset parameter, determine whether the vehicle's driving trajectory between two consecutive points at the far end checkpoint device and the near end checkpoint device meets the preset intelligent driving lane centering stability condition. S43. If the lane centering stability condition is met, the vehicle is determined to be in intelligent driving mode. S5. For vehicles determined to be in the aforementioned intelligent driving state, execute the tiered warning notification process: S51. Prioritize facial recognition of the passenger in the front passenger seat based on the image collected by the remote checkpoint device, and compare it with the public security facial information database interface to obtain contact information; S52. If S51 fails, perform face recognition on the driver in the driver's seat based on the captured image and compare to obtain contact information. S53. If S52 fails, query the motor vehicle registration information database based on the license plate number in the passing vehicle structured data to obtain the contact information of the vehicle registrant. S54. Automatically make a voice call through the cloud whistle system to the contact information obtained in S53, broadcast the information of the traffic event ahead, and request the driver to turn off the intelligent driving mode. S6. Synchronously publish directional text induction information including the vehicle license plate number through the induction screen in front of the traffic event point to remind the driver to turn off the intelligent driving mode.

[0010] Further, the lane lateral position offset parameter in S42a is calculated by the following method: Denote two consecutive points of the far-end bayonet device and the near-end bayonet device as d1 and d2 respectively, and obtain the straight-line equation parameters of the two side lines of the lane in the image according to the bayonet lane calibration data. Based on the pixel coordinates (bx, by) of the midpoint of the lower edge of the vehicle detection frame obtained by image recognition. Calculate the horizontal pixel distances (dl1, dl2) from the midpoint to the left and right side lines of the lane. Calculate the spacing difference rate rate, which satisfies the calculation formula: .

[0011] Further, in S42b, the intelligent driving lane centering stability condition simultaneously satisfies the following two items: i) The lane spacing difference rates rate1 and rate2 calculated by the vehicle at points d1 and d2 respectively satisfy: |rate1| < thres; |rate2| < thres; where thres is a preset threshold. ii) The change amount of the lane spacing difference rate between points d1 and d2 of the vehicle satisfies: |rate2 - rate1| < thres.

[0012] Further, S41 specifically includes: S41a. Use the exclusive image detection model for intelligent driving lights to detect the front part picture of the vehicle; where the exclusive image detection model is trained based on an exclusive private dataset. S41b. For the intelligent driving light targets detected by the exclusive image detection model, calculate the coincidence degree IOA between its detection frame and at least one preset reference rectangular frame, and screen the targets according to the IOA and the detection confidence to obtain the待用目标集合O. S41c. For the targets in the target set O, calculate and verify the pixel value distribution within the detection box area according to the expected color corresponding to its detection category. S41d: A weighted score is calculated based on the confidence level, overlap ratio (IOA), and pixel distribution matching degree of the targets in the target set O. If the final score exceeds a preset threshold, the intelligent driving light is determined to be turned on.

[0013] Furthermore, in S41b, the preset reference rectangle is determined in the following way: Based on the position coordinates of all labeled intelligent driving light rectangles in the dedicated private dataset, the overlap ratio (IOA) between them is calculated and iteratively merged to generate a set of one or more baseline rectangle anchors.

[0014] Furthermore, the dedicated private dataset is constructed in the following manner: Obtain a set of vehicle area images D1 containing the vehicle area with the intelligent driving light on status labeled, and a set of vehicle area images D2 without the intelligent driving light on status label. By calculating the similarity of image hash values, the images that are most similar to and least similar to each image in the image set D1 are selected from the image set D2 and merged to generate a background comparison image set D12; The image set D1 and the image set D12 are merged to form the dedicated private dataset, which is used for model training.

[0015] Furthermore, in S54, the warning content of the voice call is dynamically generated based on the type of traffic event; wherein, the types of traffic events include construction, traffic control, accidents, congestion, and severe weather.

[0016] A second aspect of the present invention provides a risk warning system for intelligent driving vehicles based on highway checkpoints, using the method described above, comprising: The traffic incident acquisition module is used to acquire information on highway traffic incidents and their geographical coordinates. The checkpoint determination module is used to determine at least two consecutive highway checkpoint devices upstream of the affected road segment based on the geographical coordinates. The intelligent driving status recognition module is used to identify vehicles currently in intelligent driving status based on vehicle data collected by the high-speed checkpoint equipment. The warning delivery module is used to initiate proactive warnings to vehicles identified as being in the intelligent driving state through the warning system, prompting them to turn off the intelligent driving mode.

[0017] Furthermore, the intelligent driving state recognition module includes: The vehicle screening unit is used to filter candidate vehicles based on the vehicle data from the remote checkpoint, the intelligent vehicle model database, and license plate rules. The light recognition unit, based on an image recognition algorithm, determines the intelligent driving light status of the candidate vehicle; The trajectory analysis unit analyzes the vehicle's lane centering stability based on continuous checkpoint images when the intelligent driving light recognition is not turned on.

[0018] Furthermore, the trajectory analysis unit calculates the lane lateral position offset parameters and rate of change of the vehicle at continuous checkpoint positions by calling preset lane calibration data, and makes a judgment based on a preset stability threshold.

[0019] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: This invention significantly improves the accuracy and reliability of intelligent driving status recognition by reusing existing highway checkpoint equipment and its collected vehicle data and images, and by integrating multi-dimensional judgment methods such as vehicle type screening, light recognition, and continuous checkpoint lane centering stability analysis. It achieves precise and efficient identification of the driving status of intelligent driving vehicles, and establishes a hierarchical proactive warning and intervention mechanism on this basis. This ensures that risk warnings can reach the driver of the target vehicle in a timely and effective manner, thereby prompting the driver to take over the vehicle in advance in the face of complex road conditions. This effectively reduces the risk of secondary accidents caused by insufficient response of the intelligent driving system or untimely driver takeover, and improves the overall traffic safety of highways. At the same time, it does not require large-scale modification or addition of roadside facilities, which greatly reduces the system deployment and promotion costs. Attached Figure Description

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

[0021] Figure 1 This is the overall flowchart disclosed in this invention; Figure 2 This is a flowchart of the intelligent driving vehicle determination process disclosed in this invention; Figure 3 This is a schematic diagram of lane centering detection disclosed in this invention. Detailed Implementation

[0022] 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. Obviously, the described embodiments are only 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 are within the scope of protection of the present invention.

[0023] This invention aims to provide a method and system for risk warning and proactive prevention of intelligent driving vehicles based on highway checkpoints. Its core lies in reusing existing highway checkpoint equipment, accurately identifying vehicles in intelligent driving mode through multi-dimensional data analysis, and proactively and targetedly issuing warnings and interventions, thereby reducing the risk of traffic accidents caused by system limitations or untimely driver takeover. Please refer to [link / reference]. Figure 1 It mainly includes the following steps: S1. Obtain traffic incident information and their geographical coordinates on the highway. The system acquires real-time information on various traffic incidents on highways, including but not limited to construction, temporary traffic control, traffic accidents, severe congestion, and severe weather, through interface integration or manual input. Each incident information must include its precise geographical coordinates.

[0024] S2. Based on the geographical coordinates of the traffic incident and combined with high-precision road network data, the path analysis algorithm determines two consecutive key highway checkpoint devices in the upstream direction affected by the traffic incident: the far-end checkpoint device (denoted as d1, which is far from the location of the traffic incident) and the near-end checkpoint device (denoted as d2, which is close to the location of the traffic incident and is usually located at the last checkpoint upstream of the traffic incident point).

[0025] In this scheme, d1 is used to initially identify and assess potential risky vehicles, and d2 is used for auxiliary verification.

[0026] S3. Based on the structured data and images of passing vehicles collected by the remote checkpoint equipment, the vehicle model in the structured data is matched with the locally maintained intelligent driving vehicle model database; and / or, the length of the vehicle license plate number is determined to be the preset length of the intelligent driving license plate, so as to filter out candidate vehicles with intelligent driving capabilities from the passing vehicles.

[0027] In this solution, this step efficiently filters out most traditional vehicles that undoubtedly possess intelligent driving capabilities. Specifically, the system retrieves vehicle passage data from the remote checkpoint device d1 over a period of time, including structured data (such as license plate number, vehicle type, and passage time) and captured images. To reduce the computational load of subsequent image analysis, the vehicle passage data is first preliminarily screened to identify candidate vehicles with intelligent driving capabilities. The screening methods include the following two, which can be used individually or in combination: a) Vehicle Model Database Matching: The system maintains a local database of intelligent driving vehicles, which is continuously updated manually with known vehicle models possessing intelligent driving capabilities. The "Vehicle Model" field in the structured data of vehicle d1 is matched against this database; vehicles that successfully match are listed as candidate vehicles.

[0028] b) License Plate Rule Judgment: In certain regions or during the testing phase, autonomous driving vehicles may use specific license plates (such as 8-digit license plates for new energy vehicles or test vehicles). The system judges the length of the license plate number in the vehicle data. If it is a preset 8-digit number, it is assumed that the vehicle has autonomous driving capabilities and is listed as a candidate vehicle.

[0029] S4. For the initially selected candidate vehicles, the system needs to further determine whether they were actually in autonomous driving mode when passing through the checkpoint. The determination process follows... Figure 2 The logic shown employs a two-level decision method: S41. First-level judgment: Intelligent driving light (ADS light) recognition.

[0030] The system uses a YOLOv8-based deep learning image recognition model to analyze the front images of candidate vehicles captured by the d1 checkpoint, focusing on whether the vehicle's autonomous driving status indicator light (ADS light) is on. If the model recognizes the ADS light as on with high confidence, it directly determines that the vehicle is currently in intelligent driving mode and proceeds to the warning process (step S5). If the ADS light is not recognized as on, it proceeds to the second-level judgment.

[0031] Specifically, the ADS light recognition process includes model building and multimodal fusion judgment, as described below: S41a, Construction and Training of Dedicated Detection Model The high-precision image detection model for ADS (Autonomous Driving Status) indicator lights involved in this invention is a deep learning image detection model specifically designed for detecting and recognizing ADS indicator lights in intelligent driving vehicles. The construction of this model first relies on the preparation of a high-quality, proprietary private dataset, specifically including the following steps: First, a large number of images of vehicles passing through highway checkpoints under different times, devices, weather conditions, and other scenarios were collected in advance. Based on the actual needs of this invention, the ADS light status in the images was precisely labeled, with the tags categorized as {1: ADS, 2: ADS_yellow, 3: ADS_other}. Here, "1: ADS" refers to the standard cyan-green light indicating the intelligent driving status of the ADS light; "2: ADS_yellow" refers to the yellow light indicating special states such as turning; and "3: ADS_other" refers to other colors (neither black nor white).

[0032] Next, considering that the proportion of positive samples (including those with ADS lights on) is extremely small, directly using the full set of images for training could easily lead to missed detections or overfitting. Therefore, the following method is used to construct a balanced dataset: a. Using the vehicle detection model, small images of local areas of the vehicle are cropped out. After manual inspection and labeling, image set D1 with labels and image set D2 without labels are obtained.

[0033] b. Calculate the image hash values ​​of all images in image sets D1 and D2, and sort them by hash value.

[0034] c. Traverse D1. For each image in D1, find the image in D2 that is most similar to and least similar to its hash value, remove it from D2 and add it to the background comparison image set D12.

[0035] d. Finally, merge D1 and D2 (keeping the number ratio of approximately 1:2), and generate blank annotation files for the images in D12, together forming a dedicated private dataset for model training.

[0036] Finally, based on the YOLO deep learning network framework, and using the aforementioned dedicated private dataset for training, an image detection model specifically designed for high-precision detection of ADS sign lights was obtained.

[0037] S41b, Target screening based on location overlap (IOA) After the aforementioned dedicated model detects vehicle images, it may output multiple ADS light targets (including category cls, detection box coordinates xyxy, and confidence score conf). To improve accuracy, location verification and filtering are required, specifically including the following steps: Anchor generation: Based on the coordinates of all labeled ADS light bounding boxes in the training dataset D1, the coordinates are iteratively merged by calculating the Intersection over Adaptive Area (IOA) to generate one or more preset sets of anchor bounding boxes representing common ADS light locations. All coordinates have been normalized to improve the algorithm's adaptability to images of different resolutions.

[0038] IOA Calculation and Selection: For each target detected by the model, calculate the IOA (Intersection over Adjacency) between its detection bounding box and each preset baseline rectangle anchor. Set a threshold. and If the target's IOA is ≥ Or satisfy conf+IOA≥ If a target is identified as a valid target, it is added to the target set O. This step filters out false positives that are obviously misplaced or have too low confidence.

[0039] S41c, Pixel Color Distribution-Based Verification For each target in set O, the color distribution of pixels within its detection box is further verified based on its detection category cls to confirm whether the light color meets expectations. This includes the following steps: First, based on the target category (e.g., cls=1 corresponds to cyan), set the RGB reference value and dynamic error threshold for the ideal color of that category.

[0040] Next, the pixels within the target detection box are traversed, and the pixels that simultaneously satisfy the condition that the green and blue channel pixel values ​​are higher than the dynamic threshold and the red channel pixel value is lower than another dynamic threshold are counted to form the effective pixel set P.

[0041] Finally, a series of spatial distribution checks are performed on set P, including the proportion of effective pixels, the distribution range of pixels in the vertical and horizontal directions, and the linearity of the pixel distribution (excluding scattered noise) by fitting a straight line. Only targets that pass all spatial distribution checks are retained in set O, and their pixel matching degree pr is recorded.

[0042] S41d, Multimodal Fusion Scoring and State Determination Iterate through the final set of candidate targets O after filtering. For each target element, calculate a comprehensive score by combining its confidence (conf), position overlap (IOA), and pixel matching degree (pr) through weighted summation. It satisfies the expression:

[0043] In the formula, , , These are the preset weight coefficients for the three judgment conditions, and ; Represents ADS light element Confidence level in image detection output; Represents element The degree of overlap between the detection frame position and the preset reference frame; Represents element The degree of matching in pixel value distribution.

[0044] In a specific example, if If the value is greater than 0.95, then with high confidence, it is determined that the vehicle has turned on the ADS light, indicating that it is currently in intelligent driving mode, and the process directly proceeds to the warning procedure (i.e., step S5); if If the threshold is not reached, or the target set O is empty, it is determined that no reliable ADS light has been identified and the process proceeds to the second-level judgment (i.e., step S42).

[0045] S42, Second-level judgment: Lane centering stability analysis based on continuous checkpoints.

[0046] For candidate vehicles whose ADS lights are not activated, the system analyzes their driving trajectories at two consecutive checkpoints, d1 and d2, to determine whether they conform to the lane centering stability characteristics typically exhibited by intelligent driving systems on high-speed straight sections. For example... Figure 3 As shown, the lane centering analysis sub-process includes: S42a. Based on the same vehicle image collected by the remote checkpoint device and the near checkpoint device respectively, and the lane calibration data pre-stored at each checkpoint, calculate the lane lateral position offset parameter of the vehicle at the two points of the remote checkpoint device and the near checkpoint device.

[0047] In this scheme, the lane lateral position offset parameter is calculated for each checkpoint (taking d1 as an example) in the following way: First, based on the lane calibration data, obtain the pixel coordinates of the four vertices of the lane rectangle. Calculate the equation (slope k1, intercept b1) of the left lane line l1 using the top left (x1, y1) and bottom left (x2, y2) points; calculate the equation (slope k2, intercept b2) of the right lane line l2 using the bottom right (x3, y3) and top right (x4, y4) points.

[0048] Then, the YOLOv8 model is used to detect the vehicle bounding box from the vehicle image, and the pixel coordinates (bx, by) of the midpoint of the lower edge of the bounding box are taken.

[0049] Next, at the by height, calculate the horizontal pixel distance (dl1, dl2) from the vehicle's center point to the left and right lane lines: Distance from the left side line:

[0050] Distance from the right side line:

[0051] Finally, the distance difference rate (based on the right-hand distance) is calculated, which satisfies the following formula:

[0052] When the rate value is negative, it indicates that the distance between the right lanes is large and the vehicle is veering to the left; when the rate value is positive, it indicates that the distance between the left lanes is large and the vehicle is veering to the right.

[0053] S42b. Based on the lane lateral position offset parameters, determine whether the vehicle's driving trajectory between two consecutive points at the remote checkpoint and the near checkpoint meets the preset intelligent driving lane centering stability conditions.

[0054] It should be noted that the lane centering stability condition for intelligent driving must simultaneously meet the following two conditions: i) The lane spacing difference rates rate1 and rate2 calculated for the vehicle at points d1 and d2 respectively satisfy: |rate1| <thres;|rate2|<thres; Where thres is a preset threshold; ii) The change in the lane spacing difference rate between points d1 and d2 satisfies: |rate2-rate1| <thres。

[0055] When both of the above conditions are met, it is determined that the vehicle is in a stable lane-centered driving state between two consecutive points, and thus it is presumed that it is currently in intelligent driving state.

[0056] S43. If the lane centering stability condition is met, the vehicle is determined to be in intelligent driving mode. S5. For vehicles identified as being in autonomous driving mode, a tiered warning notification process will be implemented to improve reach: S51. Prioritize querying the co-driver's contact person: Prioritize facial recognition of the co-driver based on images collected by remote checkpoint equipment, and compare them with the public security facial information database interface to obtain contact information.

[0057] S52, Secondary query for driver's seat contact person: If S51 is unsuccessful, then perform facial recognition of the driver's seat occupant based on the collected image and compare to obtain contact information.

[0058] S53. Backup vehicle registrant query: If S52 is unsuccessful, the vehicle registration information database will be queried based on the license plate number in the vehicle structured data to obtain the vehicle registrant's contact information; S54. After obtaining valid contact information, the system initiates a call to the obtained mobile phone number by connecting to "Cloud Sentinel" or other automated voice outbound calling systems. Once the call is connected, a pre-synthesized voice warning is played, including the type of event ahead, its approximate location, and a clear instruction to the driver to "immediately deactivate intelligent driving mode and switch to manual driving to navigate complex road sections." The types of traffic events include construction, traffic control, accidents, congestion, and severe weather.

[0059] S6. Simultaneously display targeted text guidance information including vehicle license plate numbers on the guidance screens ahead of traffic incident locations (e.g., "Intelligent driving vehicle with license plate number XXXXXX, please take over the vehicle immediately and drive with caution") to remind drivers to turn off intelligent driving mode. This precise prompt with license plate number can provide a strong warning to the driver of the target vehicle.

[0060] This solution, through the aforementioned three-dimensional early warning method of "telephone + guidance screen," greatly increases the probability of early warning information reaching the driver and prompting them to take action. This allows for the switch from intelligent driving to manual driving to be completed before the vehicle reaches a high-risk event area, thus achieving proactive risk prevention and control.

[0061] This invention also provides a risk warning system for intelligent driving vehicles based on highway checkpoints, applying the above-described method. It mainly includes a traffic event acquisition module, a checkpoint determination module, an intelligent driving state recognition module, and a warning delivery module. Specifically, the traffic event acquisition module acquires highway traffic event information and its geographical coordinates; the checkpoint determination module determines at least two consecutive highway checkpoint devices upstream of the affected road segment based on the geographical coordinates; the intelligent driving state recognition module identifies vehicles currently in intelligent driving mode based on vehicle data collected by the highway checkpoint devices; and the warning delivery module initiates proactive warnings to the identified vehicles in intelligent driving mode through the warning system, prompting them to deactivate the intelligent driving mode.

[0062] In a further embodiment, the intelligent driving status recognition module includes a vehicle initial screening unit, a light recognition unit, and a trajectory analysis unit. The vehicle initial screening unit filters candidate vehicles based on remote checkpoint data, a smart vehicle model database, and license plate rules. The light recognition unit uses image recognition algorithms to determine the smart driving light status of candidate vehicles. The trajectory analysis unit is configured to analyze the vehicle's lane centering stability based on continuous checkpoint images when the smart driving light recognition is not activated.

[0063] The trajectory analysis unit calculates the lateral position offset parameters and rate of change of the vehicle at continuous checkpoints by calling preset lane calibration data, and makes a judgment based on preset stability thresholds.

[0064] In existing technologies, checkpoints are mainly used for traffic violation capture and traffic flow monitoring. This solution applies them to the accurate identification and proactive intervention of intelligent driving states, and designs specific data processing and judgment logic. On the one hand, it reduces the system deployment and promotion costs, eliminating the need for large-scale modifications or new roadside facilities; on the other hand, by integrating multi-dimensional judgment methods such as vehicle type screening, light recognition, and lane centering stability analysis at continuous checkpoints, it significantly improves the accuracy and reliability of intelligent driving state identification. This effectively reduces the risk of secondary accidents caused by insufficient response from the intelligent driving system or untimely driver intervention, thus improving the overall traffic safety of highways.

[0065] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for risk warning of intelligent driving vehicles based on highway checkpoints, characterized in that, Includes the following steps: S1. Obtain information on traffic incidents on highways and their geographical coordinates; S2. Based on the geographical coordinates of the traffic event and combined with road network data, determine the two consecutive upstream highway checkpoint devices affected by the traffic event through a path analysis algorithm. These devices are a remote checkpoint device and a near-end checkpoint device, respectively. S3. Based on the structured data and images of passing vehicles collected by the remote checkpoint device, the vehicle model in the structured data of passing vehicles is matched with the locally maintained intelligent driving model database; and / or, it is determined whether the length of the vehicle license plate number is the preset length of the intelligent driving license plate, so as to filter out candidate vehicles with intelligent driving capabilities from the passing vehicles. S4. For the candidate vehicles, execute the intelligent driving status judgment process: S41. Based on the YOLO series image recognition algorithm, identify the vehicle image collected by the remote checkpoint device to determine whether the vehicle has turned on the intelligent driving lights. S42. If the intelligent driving light is detected to be turned on, it is determined that the vehicle is currently in intelligent driving mode; If the intelligent driving light is not detected to be on, then the lane centering analysis sub-process is executed; wherein, the lane centering analysis sub-process includes: S42a. Based on the same vehicle image collected by the remote checkpoint device and the near checkpoint device respectively, and the lane calibration data pre-stored by each checkpoint, calculate the lane lateral position offset parameter of the vehicle at the two points of the remote checkpoint device and the near checkpoint device. S42b. Based on the lane lateral position offset parameter, determine whether the vehicle's driving trajectory between two consecutive points at the far end checkpoint device and the near end checkpoint device meets the preset intelligent driving lane centering stability condition. S43. If the lane centering stability condition is met, the vehicle is determined to be in intelligent driving mode. S5. For vehicles determined to be in the aforementioned intelligent driving state, execute the tiered warning notification process: S51. Prioritize facial recognition of the passenger in the front passenger seat based on the image collected by the remote checkpoint device, and compare it with the public security facial information database interface to obtain contact information; S52. If S51 is unsuccessful, then perform facial recognition of the driver's seat occupant based on the collected images and obtain contact information by comparison. S53. If S52 is unsuccessful, then based on the license plate number in the vehicle structured data, query the motor vehicle registration information database to obtain the contact information of the vehicle registrant; S54. Automatically dial a voice call through the cloud whistle system to the contact information obtained in S53, broadcast the traffic incident information ahead and ask the driver to turn off the intelligent driving mode; S6. Simultaneously publish targeted text guidance information containing vehicle license plate numbers through the guidance screen in front of the traffic incident point to remind the driver to turn off the intelligent driving mode.

2. The method for risk warning of intelligent driving vehicles based on highway checkpoints according to claim 1, characterized in that, The lane lateral position offset parameter in S42a is calculated in the following manner: Let d1 and d2 be two consecutive points of the remote checkpoint device and the near checkpoint device, respectively. Based on the lane calibration data of the checkpoint, obtain the straight line equation parameters of the lane side lines in the image. Based on the pixel coordinates (bx, by) of the midpoint of the lower edge of the vehicle detection box obtained from image recognition; Calculate the horizontal pixel distances (dl1, dl2) of the midpoint from the left and right side lines of the lane. The interval difference rate (rate) is calculated according to the following formula: .

3. The method for risk warning of intelligent driving vehicles based on highway checkpoints according to claim 2, characterized in that, In the S42b, the intelligent driving lane centering stability condition simultaneously satisfies the following two items: i) The lane spacing difference rates rate1 and rate2 calculated by the vehicle at two points d1 and d2 satisfy: |rate1| < thres; |rate2| < thres; where thres is a preset threshold. ii) The change amount of the lane spacing difference rate between two points d1 and d2 of the vehicle satisfies: |rate2 - rate1| < thres.

4. The method for risk warning of intelligent driving vehicles based on highway checkpoints according to claim 1, characterized in that, The S41 specifically includes: S41a. Use the exclusive image detection model for intelligent driving lights to detect the front image of the vehicle; wherein, the exclusive image detection model is trained based on an exclusive private dataset. S41b. For the intelligent driving light targets detected by the exclusive image detection model, calculate the coincidence degree IOA between its detection frame and at least one preset reference rectangular frame, and screen the targets according to the IOA and detection confidence to obtain a set of待用 targets O. S41c. For the targets in the set of待用 targets O, calculate and verify the pixel value distribution within the detection frame area according to the expected color corresponding to their detection categories. S41d. Perform weighted scoring based on the confidence, coincidence degree IOA, and pixel distribution matching degree of the targets in the set of待用 targets O. If the final score exceeds the preset threshold, it is determined that the intelligent driving lights are turned on.

5. The intelligent driving vehicle risk warning method based on highway checkpoints according to claim 4, characterized in that, In the S4lb, the preset reference rectangular frame is determined by the following method: Based on the position coordinates of all labeled intelligent driving light rectangular frames in the exclusive private dataset, generate a set of one or more reference rectangular frames anchor by iterative merging through calculating the coincidence degree IOA between each other.

6. The intelligent driving vehicle risk warning method based on highway checkpoints according to claim 4, characterized in that, The exclusive private dataset is constructed by the following method: Obtain a set of vehicle area pictures D1 containing the labeled intelligent driving light on states, and a set of vehicle area pictures D2 not containing the intelligent driving light on state labels. By calculating the similarity of image hash values, screen out the pictures in the picture set D2 that are most similar and least similar to each picture in the picture set D1, and merge them to generate a background comparison picture set D12. Merge the picture set D1 and the picture set D12 to form the exclusive private dataset for model training.

7. The intelligent driving vehicle risk warning method based on highway checkpoints according to claim 1, characterized in that, In the S54, the warning content of the voice call is dynamically generated according to the traffic event type; wherein, the traffic event type includes construction, control, accident, congestion, and bad weather.

8. A risk warning system for intelligent driving vehicles based on highway checkpoints, applying the method as described in any one of claims 1-7, characterized in that, It includes: A traffic event acquisition module for acquiring highway traffic event information and its geographical location coordinates. A checkpoint determination module for determining at least two consecutive highway checkpoint devices upstream of the affected section based on the geographical location coordinates. An intelligent driving state recognition module for identifying the vehicles currently in the intelligent driving state based on the vehicle data collected by the highway checkpoint devices. A warning reach module for initiating an active warning to the identified vehicles in the intelligent driving state through a warning system to prompt to turn off the intelligent driving mode.

9. The intelligent driving vehicle risk warning system based on highway checkpoints according to claim 8, characterized in that, The intelligent driving state recognition module includes: The vehicle screening unit is used to filter candidate vehicles based on the vehicle data from the remote checkpoint, the intelligent vehicle model database, and license plate rules. The light recognition unit, based on an image recognition algorithm, determines the intelligent driving light status of the candidate vehicle; The trajectory analysis unit analyzes the vehicle's lane centering stability based on continuous checkpoint images when the intelligent driving light recognition is not turned on.

10. The intelligent driving vehicle risk warning system based on highway checkpoints according to claim 9, characterized in that, The trajectory analysis unit calculates the lateral position offset parameters and rate of change of the vehicle at continuous checkpoints by calling preset lane calibration data, and makes a judgment based on preset stability thresholds.

Images