Adaptive visibility detection system for highways

Abstract

The present application relates to the technical field of image processing analysis, especially to the adaptive visibility detection system of expressway, the present application obtains the monitoring image, driving data of expressway through information acquisition module, and generates visibility description rule text for different specifications of vehicles in monitoring image; through image feature extraction module, the feature salience time domain is determined, different feature clustering sets are divided according to the text matching result; according to the driving data comparison result of feature salience time domain in different feature clustering sets, whether the current visibility broadcast information of monitoring section is searched is determined through the search comparison module; through the detection correction module, according to the comparison result of image extraction information and visibility broadcast information, whether the visibility broadcast information is corrected is determined. Further, the adaptive correction of visibility broadcast is realized according to the specification difference of vehicle using time domain dynamic information, and the demand of timely and safe management and control of expressway is met.

Description

Technical Field

[0001] This invention relates to the field of image processing and analysis technology, and in particular to an adaptive visibility detection system for highways. Background Technology

[0002] Highway visibility is a core indicator affecting highway traffic safety. In complex scenarios such as rain, road water accumulation, fog, and dust storms, a rapid decrease in visibility can easily lead to serious traffic accidents such as multi-vehicle collisions and rollovers. Real-time, accurate, and adaptive visibility detection has become an essential requirement for intelligent transportation and road safety management.

[0003] Currently, visibility detection on highways mainly relies on two types of technologies: dedicated meteorological equipment and traditional video detection. While dedicated visibility detection equipment, such as forward scattering meters and transmission-type visibility meters, offers high accuracy, they are expensive to deploy, difficult to maintain, and cannot achieve seamless coverage across the entire road section. They also have blind spots in monitoring sudden low visibility events such as patchy fog and localized water splash fog. Traditional visibility detection methods based on surveillance videos often rely on single metrics such as contrast, edge gradient, or average brightness for estimation, resulting in poor detection stability. With the advancement of smart highway construction, existing detection methods are no longer adequate for the needs of full-scene, full-coverage, and highly robust visibility monitoring.

[0004] For example, Chinese invention patent CN114359211A, published on April 15, 2022, discloses a visibility value detection method, apparatus, device, and storage medium. It can acquire an image to be detected; extract features from the image to obtain a characteristic quantity of the image; and determine the visibility value corresponding to the image to be detected based on a pre-calibrated mapping relationship between the characteristic quantity and the visibility value, wherein the characteristic quantity is used to characterize the magnitude of the visibility value of the scene contained in the image. In this way, the visibility value of the image to be detected can be determined without the need for a dedicated visibility measuring instrument.

[0005] Existing technologies do not consider the difference in visibility impact between large and small vehicles when passing through road surface water, and do not use time-domain dynamic information to adaptively correct visibility broadcasts based on vehicle size differences, making it difficult to meet the needs of highways for timely safety management. Summary of the Invention

[0006] To address this, the present invention provides an adaptive visibility detection system for highways, which overcomes the problem that existing technologies do not consider the difference in visibility impact between large and small vehicles when passing through road surface water, and cannot adaptively correct visibility broadcasts using time-domain dynamic information in conjunction with vehicle size differences, thus failing to meet the needs of highways for timely safety management.

[0007] To achieve the above objectives, the present invention provides an adaptive visibility detection system for highways, comprising: The information acquisition module includes a monitoring unit for acquiring monitoring images of highways, a driving data unit for acquiring driving data within the monitoring images, and a text generation unit for generating visibility description rule text containing at least two information dimensions for vehicles of different sizes within the monitoring images. The image feature extraction module, which is connected to the information acquisition module, is used to record the time domain corresponding to the visibility description rule text that meets the description screening conditions as the feature explicit time domain, obtain the visibility description rule text of the vehicle in each feature explicit time domain, and classify several feature explicit time domains into different feature cluster sets according to the text screening results. The retrieval and comparison module is connected to the information acquisition module and the image feature extraction module respectively. It is used to obtain the driving data of each feature cluster set and determine whether to retrieve the current visibility broadcast information of the monitored road section based on the comparison results of the driving data in the time domain of the feature explicitness in different feature cluster sets. The detection and correction module is connected to both the information acquisition module and the retrieval and comparison module. It is used to determine whether to correct the visibility broadcast information based on the comparison results between the image extraction information and the visibility broadcast information.

[0008] Furthermore, the text generation unit is used to generate visibility description rule text, wherein, The text generation unit pre-classifies vehicles in the monitored images into two categories, large vehicles and small vehicles, based on vehicle size parameters. The visibility description rule text generated for each target vehicle includes at least two information dimensions: whether the whole vehicle is fully visible and whether the parts are visible.

[0009] Furthermore, whether the whole is fully visible in the two information dimensions refers to whether the overall outline of the vehicle appears completely, and whether the part is visible refers to whether there are any identifiable pixel features in the wheel assembly area. If there are identifiable pixel features in the wheel assembly area, it is determined to be partially visible.

[0010] Furthermore, the image feature extraction module is used to determine the feature dominance time domain, wherein, The described filtering condition is that within the same continuous time domain segment, the monitoring images simultaneously meet either the first filtering condition or the second filtering condition. The first filtering condition is that the visibility description rule text for large vehicles is either not fully visible or partially visible, and the visibility description rule text for small vehicles is either fully visible or partially visible; the second filtering condition is that the visibility description rule text for large vehicles is either not fully visible or partially invisible, and the visibility description rule text for small vehicles is either not fully visible or partially visible.

[0011] Furthermore, the image feature extraction module is used to classify several explicit temporal features into different feature cluster sets, wherein, The image feature extraction module assigns the dominant temporal domain features that meet the first screening criteria to the first feature cluster set, and assigns the dominant temporal domain features that meet the second screening criteria to the second feature cluster set.

[0012] Furthermore, the retrieval and comparison module is used to obtain driving data for each feature cluster set, wherein, The driving data unit is used to collect the driving speed of all vehicles passing through the corresponding monitored road segment within each feature explicit time domain; and to calculate the average driving speed of all vehicles in the first feature cluster set corresponding to the explicit time domain as the first driving data, and to calculate the average driving speed of all vehicles in the second feature cluster set corresponding to the explicit time domain as the second driving data.

[0013] Furthermore, the retrieval and comparison module is used to determine whether to retrieve the current visibility broadcast information for the monitored road segment, wherein, If the comparison result shows that the first driving data corresponding to the first feature cluster set is greater than the second driving data corresponding to the second feature cluster set, then the retrieval comparison module retrieves the visibility broadcast information currently published for the monitored road segment.

[0014] Furthermore, the detection and correction module is used to obtain image extraction information, wherein the image extraction information includes the vertical installation distance between the monitoring unit at the corresponding monitoring point and the road surface.

[0015] Furthermore, the detection and correction module is used to determine whether to correct the visibility broadcast information, wherein, If the broadcast value in the current visibility broadcast information of the monitored road segment is greater than the value in the image extraction information of the corresponding monitoring point, the detection and correction module determines to correct the current visibility broadcast information.

[0016] Furthermore, the detection and correction module is used to correct the visibility broadcast information, wherein, The detection and correction module uses the values ​​in the image extraction information as the correction reference value, and determines the corrected visibility broadcast value by multiplying the correction reference value and the correction coefficient. The correction coefficient is negatively correlated with the absolute value of the difference between the first and second driving data.

[0017] The beneficial effects of the technical solution presented in this application include: acquiring highway monitoring images and driving data within the monitoring images through an information acquisition module, and generating visibility description rule text for vehicles of different sizes within the monitoring images; determining the dominant temporal domain of features through an image feature extraction module, and dividing different feature cluster sets based on text matching results; determining whether to retrieve the current visibility broadcast information for the monitored road segment based on the comparison results of driving data in the dominant temporal domain within different feature cluster sets through a retrieval and comparison module; and determining whether to correct the visibility broadcast information based on the comparison results between the image extraction information and the visibility broadcast information through a detection and correction module. Furthermore, this achieves adaptive correction of visibility broadcasts using temporal dynamic information based on vehicle size differences, meeting the needs of highways for timely and safe management.

[0018] Furthermore, this invention defines whether the overall visibility is complete as whether the overall outline of the vehicle appears completely within the effective recognition area of ​​the monitoring image, which can intuitively reflect the transparency of the global field of view and characterize the overall visibility level caused by water accumulation. It defines whether the local visibility is complete as whether there are any identifiable pixel features in the vehicle wheel area. As long as there are identifiable pixels, it is determined to be locally visible, which can accurately capture the small-scale visibility attenuation phenomenon of local splashing water and provide an objective basis for the visibility judgment of vehicle models.

[0019] Furthermore, by setting a first screening condition and a second screening condition within the same continuous time domain segment, this invention can accurately match the objective law that large vehicles have a significantly greater impact on visibility than small vehicles in road flooding scenarios, corresponding to two typical states of mild and severe visibility attenuation, respectively. Thus, it can screen out the real periods of abnormal visibility caused by the road environment, providing real and effective parameter samples for cluster analysis.

[0020] Furthermore, the first and second screening conditions of the present invention correspond to two typical states of visibility attenuation: mild and severe attenuation, respectively, under the scenario of road surface water accumulation. Using these as the basis for clustering, the time-domain samples within the same set can maintain a high degree of consistency in terms of visibility attenuation, interference type, and scene characteristics. This can effectively distinguish between different intensities of abnormal visibility conditions and improve the reliability of system detection.

[0021] Furthermore, in this invention, when the average vehicle speed of the first feature cluster set corresponding to slight visibility decay is higher than the average vehicle speed of the second feature cluster set corresponding to severe visibility decay, the difference in driving speed can be directly ruled out as the dominant factor causing the slight or severe visibility decay. It can be determined that the difference in visibility level is due to the continuous deterioration of objective environmental factors such as weather and road surface water. At this time, triggering visibility broadcast information retrieval is necessary for timely safety management of highways.

[0022] Furthermore, this invention uses the vertical installation distance between the monitoring unit and the road surface as the correction benchmark value, and dynamically determines the correction coefficient by combining the difference between the average driving speed of the first feature cluster set and the second feature cluster set. It can be understood that the installation height of the monitoring equipment provides a physical benchmark for the road surface visibility distance, while the vehicle speed difference can objectively reflect the actual impact intensity of water splash fog on visibility on the road surface, so that the corrected visibility value conforms to the real visibility conditions on site, realizes accurate calibration of broadcast information, and significantly improves the reliability of highway visibility detection results and the safety of traffic guidance. Attached Figure Description

[0023] Figure 1 This is a system block diagram of an adaptive visibility detection system for highways according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the logic for determining whether the entire system is fully visible in an embodiment of the present invention. Figure 3 This is a flowchart illustrating the logic for determining whether a local area is visible in an embodiment of the present invention. Figure 4 This is a flowchart illustrating the logic of determining whether to retrieve the current visibility broadcast information for the monitored road segment in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0025] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0026] It should be noted that in the description of this invention, the terms "upper," "lower," "inner," "outer," etc., which indicate the direction or positional relationship, are based on the direction or positional relationship shown in the drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0027] It should be understood that although the terms "first," "second," etc., may be used in this invention to describe various types of information, these information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this invention, first information may also be referred to as second information, and similarly, second information may also be referred to as first information.

[0028] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation" and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0029] Please see Figure 1 The diagram shown is a system block diagram of an adaptive visibility detection system for highways according to an embodiment of the present invention. The adaptive visibility detection system for highways according to the present invention includes: The information acquisition module includes a monitoring unit for acquiring monitoring images of highways, a driving data unit for acquiring driving data within the monitoring images, and a text generation unit for generating visibility description rule text containing at least two information dimensions for vehicles of different sizes within the monitoring images. The monitoring unit in this invention uses a bullet-type high-definition camera that is commonly deployed on highways and is installed on highway gantry or roadside poles.

[0030] In this invention, the vehicle data unit can access speed measurement data from highway gantry checkpoint radar, and can also calculate the real-time vehicle speed by using the background difference method combined with the vehicle wheelbase calibration value through the video frame sequence of the monitoring unit.

[0031] The vehicle recognition, size extraction, and visibility determination models relied upon by the text generation unit in this invention adopt the YOLO instance segmentation model as the base model, adapted for highway scenarios. The sample set covers 10 common highway scenarios, including sunny, rainy, flooded, foggy, backlit, and nighttime, with ≥200 frames per scenario and a total of ≥2000 frames. The sample annotations include the vehicle's bounding rectangle, vehicle size, vehicle type, wheel area, and visibility label. The samples are divided into training, validation, and test sets in a 7:2:1 ratio, using a two-stage training mode of pre-training and fine-tuning. The pre-training stage has 30 iterations, and the fine-tuning stage optimizes highway vehicle features with 15 iterations. The trained model is converted to ONNX format and deployed on a roadside edge computing terminal. The text generation unit can be a pre-trained model. Models that generate text information output from image recognition are widely used in scene-based image recognition, which will not be elaborated here.

[0032] The image feature extraction module, which is connected to the information acquisition module, is used to record the time domain corresponding to the visibility description rule text that meets the description screening conditions as the feature explicit time domain, obtain the visibility description rule text of the vehicle in each feature explicit time domain, and classify several feature explicit time domains into different feature cluster sets according to the text screening results. The retrieval and comparison module is connected to the information acquisition module and the image feature extraction module respectively. It is used to obtain the driving data of each feature cluster set and determine whether to retrieve the current visibility broadcast information of the monitored road section based on the comparison results of the driving data in the time domain of the feature explicitness in different feature cluster sets. The detection and correction module is connected to both the information acquisition module and the retrieval and comparison module. It is used to determine whether to correct the visibility broadcast information based on the comparison results between the image extraction information and the visibility broadcast information.

[0033] The present invention does not limit the specific structure of the image feature extraction module, the retrieval and comparison module, and the detection and correction module. The modules themselves or their functional units can be constructed using logic components. The logic components can be field-programmable logic components, microprocessors, processors used in computers, etc., which will not be elaborated here.

[0034] The image feature extraction module, retrieval and comparison module, detection and correction module, and text generation unit of this invention all run on a computing terminal, establish communication with the monitoring unit and the driving data unit, and connect to the traffic meteorological platform to obtain the visibility broadcast information officially released by the monitored road section.

[0035] Specifically, the text generation unit is used to generate visibility description rule text, wherein, The text generation unit pre-classifies vehicles in the monitored images into two categories, large vehicles and small vehicles, based on vehicle size parameters. The visibility description rule text generated for each target vehicle includes at least two information dimensions: whether the whole vehicle is fully visible and whether the parts are visible.

[0036] In this invention, the vehicle body length in the surveillance image is used as the dividing criterion. The bounding rectangle of the vehicle is extracted by a pre-trained segmentation model, and the actual size of the vehicle body is calculated. Among them, vehicles with a body length ≥ 6m are classified as large vehicles, and vehicles with a body length < 6m are classified as small vehicles.

[0037] For example, the text generation unit generates standardized structured descriptive text for each target vehicle within the effective identification area of ​​the surveillance image. The text format is uniformly: vehicle ID + vehicle specifications + whether the whole is fully visible + whether the part is visible, for example: "ID001 + large vehicle + not fully visible + partially visible".

[0038] The criteria for determining whether the vehicle is fully visible are that the front, body, and rear of the vehicle are completely visible in the image. The criteria for determining whether a part is visible are that there are identifiable tire edge pixel features or tire-road boundary pixel features in any sub-region of the image.

[0039] Understandably, this invention uses vehicle body size parameters as the classification basis, which can accurately distinguish between large and small vehicles, ensuring the consistency and reliability of vehicle specification classification. For each target vehicle, visibility description rule text is generated in two dimensions: overall visibility and local visibility. Overall visibility characterizes the global visibility level, while local wheel group visibility characterizes the degree of local visibility attenuation. These two aspects corroborate and complement each other, accurately reflecting the differentiated impact of different vehicle types on visibility in road flooding scenarios, meeting the actual needs of visibility detection in complex highway scenarios.

[0040] Specifically, whether the whole is fully visible in the two information dimensions refers to whether the overall outline of the vehicle appears completely, and whether the local area is visible refers to whether there are any identifiable pixel features in the wheel assembly area. If there are identifiable pixel features in the wheel assembly area, it is determined to be locally visible.

[0041] Please see Figure 2 As shown, it is a logic flowchart for determining whether the whole is completely visible in an embodiment of the present invention. In the implementation of the present invention, if the front, body and rear of the vehicle appear completely in the picture without any image edge truncation, it can be determined that the whole is completely visible; if the front / rear of the vehicle is truncated by the image, it is determined that the whole is not completely visible.

[0042] Please see Figure 3 The diagram shown is a flowchart illustrating the logic of determining whether a part is visible in an embodiment of the present invention. In this embodiment, the wheel area of ​​the vehicle is used as the determination object. The wheel area is the area where the tires are located at the bottom of the vehicle body. For the wheel area, the Canny edge detection algorithm is used to extract pixel features. If there are identifiable tire edge pixel features or tire-road boundary pixel features in any sub-region, it is determined to be locally visible. Only when there are no identifiable tire-related pixel features in any sub-region is it determined to be locally invisible.

[0043] It is understood that the present invention defines whether the overall visibility is complete as whether the overall outline of the vehicle appears completely within the effective recognition area of ​​the monitoring image, which can intuitively reflect the transparency of the global field of view and characterize the overall visibility level caused by water accumulation. The definition of whether the local visibility is complete is whether there are any identifiable pixel features in the vehicle wheel area. As long as there are identifiable pixels, it is determined to be locally visible. This can accurately capture the small-scale visibility attenuation phenomenon of local water splashes and provide an objective basis for the visibility judgment of vehicle models.

[0044] Specifically, the image feature extraction module is used to determine the dominant time domain of features, denoted as the dominant time domain of features, wherein... The described filtering condition is that within the same continuous time domain segment, the monitoring images simultaneously meet either the first filtering condition or the second filtering condition. The first filtering condition is that the visibility description rule text for large vehicles is either not fully visible or partially visible, and the visibility description rule text for small vehicles is either fully visible or partially visible; the second filtering condition is that the visibility description rule text for large vehicles is either not fully visible or partially invisible, and the visibility description rule text for small vehicles is either not fully visible or partially visible.

[0045] In this invention, the surveillance images are divided into time domains, and the shortest duration of a single time domain segment is 5 seconds.

[0046] Within the same continuous time domain segment, all large vehicles identified in the surveillance image have visibility description rule text of "overall not fully visible, partially visible," and all small vehicles identified in the same time period have visibility description rule text of "overall fully visible, partially visible." Continuous time domain segments that meet this condition meet the first screening condition. Similarly, all large vehicles identified in the surveillance image have visibility description rule text of "overall not fully visible, partially invisible," and all small vehicles identified in the same time period have visibility description rule text of "overall not fully visible, partially visible." Continuous time domain segments that meet this condition meet the second screening condition. Within the same continuous time domain segment, if either the first or second screening condition is met, it is determined to be a feature-dominant time domain segment.

[0047] This invention sets up a first and a second screening condition within the same continuous time segment. This allows for precise matching of the objective law that large vehicles significantly impact visibility more than small vehicles in road flooding scenarios. The second condition corresponds to two typical states of visibility attenuation: mild attenuation (large vehicles are generally affected but partially still identifiable, while small vehicles are almost unaffected); and severe attenuation (large vehicles are obscured both entirely and partially, while small vehicles are only generally affected). This allows for the identification of genuine periods of abnormal visibility caused by road conditions, providing accurate and effective parameter samples for cluster analysis.

[0048] Specifically, the image feature extraction module is used to classify several features in the temporal domain into different feature cluster sets, wherein, The image feature extraction module assigns the dominant temporal domain features that meet the first screening criteria to the first feature cluster set, and assigns the dominant temporal domain features that meet the second screening criteria to the second feature cluster set.

[0049] The first and second screening conditions of this invention correspond to two typical states of visibility attenuation: mild and severe attenuation, respectively, under road surface water conditions. Using these conditions as the basis for clustering, the time-domain samples within the same set can maintain a high degree of consistency in terms of visibility attenuation, interference type, and scene characteristics. This can effectively distinguish between different intensities of abnormal visibility conditions and improve the reliability of system detection.

[0050] Specifically, the retrieval and comparison module is used to obtain driving data for each feature cluster set, wherein, The driving data unit is used to collect the driving speed of all vehicles passing through the corresponding monitored road segment within each feature explicit time domain; and to calculate the average driving speed of all vehicles in the first feature cluster set corresponding to the explicit time domain as the first driving data, and to calculate the average driving speed of all vehicles in the second feature cluster set corresponding to the explicit time domain as the second driving data.

[0051] In the implementation of this invention, for a single feature explicit time domain, the real-time driving speed of all passing vehicles in the effectively identified area of ​​the monitored road segment is collected within that time domain segment, and the driving speed of a single vehicle is taken as the average speed within that time domain segment.

[0052] This invention uses the average speed of all vehicles in the time domain as the statistical basis to objectively reflect the true level of the overall driving speed of the road segment under the corresponding visibility attenuation state; it independently calculates the two cluster sets to form two sets of directly comparable quantitative indicators, which accurately reflect the correlation between different degrees of visibility attenuation and driving speed.

[0053] Specifically, please refer to Figure 4 The diagram shown illustrates the logic flowchart for determining whether to retrieve the current visibility broadcast information for the monitored road segment according to an embodiment of the present invention. The retrieval and comparison module is used to determine whether to retrieve the current visibility broadcast information for the monitored road segment. If the comparison result matches the first driving data corresponding to the first feature cluster set being greater than the second driving data corresponding to the second feature cluster set, then the retrieval comparison module retrieves the visibility broadcast information currently published for the monitored road segment. If the comparison result does not match the first vehicle data corresponding to the first feature cluster set being greater than the second vehicle data corresponding to the second feature cluster set, the retrieval comparison module determines that the retrieval process will not be triggered.

[0054] In the implementation of this invention, after the retrieval is triggered, the retrieval and comparison module retrieves the latest visibility broadcast information of the road section where the monitoring point is located through the interface of the provincial traffic meteorological platform, extracts the visibility value in the broadcast information, and the unit is uniformly in meters. If the broadcast information is a visibility interval value, the maximum value of the interval is taken as the broadcast value for this comparison.

[0055] Understandably, under conditions of road flooding, higher vehicle speeds result in a larger splash area and a stronger impact on local visibility. When the average vehicle speed of the first feature cluster corresponding to slight visibility reduction is higher than the average vehicle speed of the second feature cluster corresponding to severe visibility reduction, it can be directly ruled out that the difference in driving speed is the dominant factor causing the slight or severe reduction in visibility. It can be determined that the difference in visibility levels originates from the continuous deterioration of objective environmental factors such as weather and road flooding. At this point, triggering visibility broadcast information retrieval is necessary for timely safety management of highways.

[0056] Specifically, the detection and correction module is used to obtain image extraction information, wherein the image extraction information includes the vertical installation distance between the monitoring unit at the corresponding monitoring point and the road surface.

[0057] Specifically, the detection and correction module is used to determine whether to correct the visibility broadcast information, wherein, If the broadcast value in the current visibility broadcast information of the monitored road segment is greater than the value in the image extraction information of the corresponding monitoring point, the detection and correction module determines to correct the current visibility broadcast information. If the visibility report value is less than or equal to the vertical installation distance calibration value of the monitoring unit, the original report value will be maintained.

[0058] This invention uses the vertical installation distance between the monitoring unit and the road surface as the core basis for image extraction information, and uses a broadcast value greater than this vertical installation distance as a correction trigger condition. The vertical installation distance of the monitoring equipment is the distance at which road targets can be clearly imaged. When the visibility broadcast value is greater than this limit distance, it indicates that the weather broadcast data is higher than the actual visible distance of the road surface and there is distortion. At this time, the broadcast value is calibrated to a reliable value that conforms to the actual imaging conditions on site, so as to avoid traffic control errors caused by falsely high visibility broadcasts and improve the safety of control.

[0059] Specifically, the detection and correction module is used to correct the visibility broadcast information, wherein... The detection and correction module uses the values ​​in the image extraction information as the correction reference value, and determines the corrected visibility broadcast value L by multiplying the correction reference value and the correction coefficient. The correction coefficient is negatively correlated with the absolute value of the difference between the first and second driving data.

[0060] For example, when the average difference in driving speed ΔV < 5 km / h, the correction factor is 0.8; When the average difference in driving speed ΔV is within the range of 5km / h≤ΔV<10km / h, the correction factor is 0.6; When the average difference in driving speed ΔV is greater than or equal to 10 km / h, the correction factor is 0.5.

[0061] The corrected visibility forecast value is L = H × k, where k is the correction coefficient, H is the correction reference value, and the unit of the visibility forecast value is meters.

[0062] This invention uses the vertical installation distance between the monitoring unit and the road surface as the correction benchmark value, and dynamically determines the correction coefficient by combining the difference between the average driving speed of the first feature cluster set and the second feature cluster set. It can be understood that the installation height of the monitoring equipment provides a physical benchmark for the road surface visibility distance, while the vehicle speed difference can objectively reflect the actual impact of water splash fog on visibility on the road surface. This makes the corrected visibility value fit the actual visibility conditions on site, achieves accurate calibration of the broadcast information, and improves the reliability of highway visibility detection results and the safety of traffic guidance.

[0063] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0064] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive visibility detection system for highways, characterized in that, include: The information acquisition module includes a monitoring unit for acquiring monitoring images of highways, a driving data unit for acquiring driving data within the monitoring images, and a text generation unit for generating visibility description rule text containing at least two information dimensions for vehicles of different sizes within the monitoring images. The text generation unit pre-classifies vehicles in the monitoring images into two categories, large vehicles and small vehicles, based on vehicle size parameters. The visibility description rule text generated for each target vehicle includes at least two information dimensions: whether the whole vehicle is fully visible and whether the parts are visible. Whether the whole is completely visible in the two information dimensions refers to whether the overall outline of the vehicle appears completely, and whether the part is visible refers to whether there are any identifiable pixel features in the wheel area of ​​the vehicle. If there are identifiable pixel features in the wheel area, it is determined to be partially visible. The image feature extraction module, which is connected to the information acquisition module, is used to record the time domain corresponding to the visibility description rule text that meets the description screening conditions as the feature explicit time domain, obtain the visibility description rule text of the vehicle in each feature explicit time domain, and classify several feature explicit time domains into different feature cluster sets according to the text screening results. The retrieval and comparison module is connected to the information acquisition module and the image feature extraction module respectively. It is used to obtain the driving data of each feature cluster set and determine whether to retrieve the current visibility broadcast information of the monitored road section based on the comparison results of the driving data in the time domain of the feature explicitness in different feature cluster sets. The detection and correction module is connected to the information acquisition module and the retrieval and comparison module respectively, and is used to determine whether to correct the visibility broadcast information based on the comparison results between the image extraction information and the visibility broadcast information. The image extraction information includes the vertical installation distance between the monitoring unit at the corresponding monitoring point and the road surface. The detection and correction module is used to determine whether to correct the visibility broadcast information. If the broadcast value in the current visibility broadcast information of the monitored road segment is greater than the value in the image extraction information of the corresponding monitoring point, the detection and correction module determines to correct the current visibility broadcast information.

2. The adaptive visibility detection system for highways according to claim 1, characterized in that, The image feature extraction module is used to determine the dominant temporal domain of features, wherein... The described filtering condition is that within the same continuous time domain segment, the monitoring images simultaneously meet either the first filtering condition or the second filtering condition. The first filtering condition is that the visibility description rule text for large vehicles is either not fully visible or partially visible, and the visibility description rule text for small vehicles is either fully visible or partially visible; the second filtering condition is that the visibility description rule text for large vehicles is either not fully visible or partially invisible, and the visibility description rule text for small vehicles is either not fully visible or partially visible.

3. The adaptive visibility detection system for highways according to claim 2, characterized in that, The image feature extraction module is used to classify several explicit temporal features into different feature cluster sets, wherein... The image feature extraction module assigns the dominant temporal domain features that meet the first screening criteria to the first feature cluster set, and assigns the dominant temporal domain features that meet the second screening criteria to the second feature cluster set.

4. The adaptive visibility detection system for highways according to claim 3, characterized in that, The retrieval and comparison module is used to obtain driving data for each feature cluster set, wherein, The driving data unit is used to collect the driving speed of all vehicles passing through the corresponding monitored road segment within each feature explicit time domain; and to calculate the average driving speed of all vehicles in the first feature cluster set corresponding to the explicit time domain as the first driving data, and to calculate the average driving speed of all vehicles in the second feature cluster set corresponding to the explicit time domain as the second driving data.

5. The adaptive visibility detection system for highways according to claim 4, characterized in that, The retrieval and comparison module is used to determine whether to retrieve the current visibility broadcast information for the monitored road segment, wherein, If the comparison result shows that the first driving data corresponding to the first feature cluster set is greater than the second driving data corresponding to the second feature cluster set, then the retrieval comparison module retrieves the visibility broadcast information currently published for the monitored road segment.

6. The adaptive visibility detection system for highways according to claim 1, characterized in that, The detection and correction module is used to correct the visibility broadcast information, wherein... The detection and correction module uses the values ​​in the image extraction information as the correction reference value, and determines the corrected visibility broadcast value by multiplying the correction reference value and the correction coefficient. The correction coefficient is negatively correlated with the absolute value of the difference between the first and second driving data.

Images