A cloud edge terminal cooperative highway video grading processing and transmission optimization method

Abstract

This invention discloses a cloud-edge-device collaborative method for hierarchical processing and transmission optimization of highway videos. It relates to the field of highway video hierarchical processing and primarily addresses the coupling interference between adaptive adjustment of video acquisition parameters and the long-tail effect of vehicle target loss caused by parameter switching. This coupling interference results in inaccurate warnings when there is a lack of reliable trajectory deviation reference points. Specifically, it dynamically and adaptively adjusts the resolution, frame rate, and frame interval parameters of video acquisition to achieve fine-grained, high-precision acquisition in congested scenarios and coarse-grained, lightweight acquisition in smooth scenarios. It dynamically identifies the target long-tail effect caused by vehicle occlusion and avoids target loss and trajectory deviation problems through dynamic differentiation of contour points and dynamic parameter priority adjustment strategies. It also performs edge detection on highway surface areas and achieves accurate identification and cloud-based warnings for vehicle trajectory deviation, lane occupation risk, and lane departure risk in lane defect scenarios.

Description

Technical Field

[0001] This invention relates to the field of highway video classification processing, specifically a cloud-edge-device collaborative method for highway video classification processing and transmission optimization. Background Technology

[0002] Currently, the intelligent highway traffic monitoring system has fully implemented the entire process of video acquisition, data analysis, and traffic control based on the cloud-edge-device collaborative architecture. As the core data source for traffic status perception, vehicle target recognition, and traffic risk warning, the processing accuracy and transmission efficiency of highway video directly determine the real-time performance and accuracy of traffic control.

[0003] In existing technologies, there is a coupling interference between the adaptive adjustment of video acquisition parameters and the long-tail effect of vehicle target loss caused by parameter switching, resulting in inaccurate warnings when there is a lack of reliable trajectory deviation reference. Specifically, the congestion-intensifying phase requires fine-grained, high-precision adjustments, while the smoothness-increasing phase requires coarse-grained, low-precision adjustments. However, parameter switching itself is precisely the cause of the long-tail effect. Conventional methods are no longer able to prioritize and simultaneously complete target differentiation and parameter adjustment when the two conflict, leading to target tracking interruption. Moreover, when a vehicle enters an area with missing or faded lane markings, the lack of a reliable trajectory deviation reference means the system cannot construct an adaptive distance range based on historical oncoming distances to distinguish between the risk of inner lane occupation and the risk of deviating from the lane boundary, further exacerbating the inaccuracy of warnings.

[0004] To address the aforementioned technical shortcomings, this paper proposes a cloud-edge-device collaborative method for hierarchical processing and transmission optimization of highway videos. This method relies entirely on a layered collaborative processing logic, with the edge device responsible for real-time video acquisition, traffic flow detection, parameter adjustment, and target recognition, while the cloud device handles data aggregation, risk warning, and rule iteration. Through multi-step, multi-threshold dynamic adaptation and optimization, it improves the accuracy of vehicle target recognition under complex road conditions and enhances the ability to warn of abnormal traffic risks. This method is applicable to real-time traffic monitoring and intelligent management scenarios on various urban highways, intercity trunk roads, and expressways. Summary of the Invention

[0005] The purpose of this invention is to solve the problems mentioned above by proposing a cloud-edge-device collaborative method for hierarchical processing and transmission optimization of highway video.

[0006] The objective of this invention can be achieved through the following technical solution: a cloud-edge-device collaborative method for hierarchical processing and transmission optimization of highway video, comprising the following steps: Step 1: Adaptive adjustment of video grading granularity: Based on the video capture hardware specifications, determine the frame rate, resolution, and frame interval, and capture video to obtain the deviation of traffic flow density on the same road segment at different time periods. The density is the ratio of the traffic flow per unit time to the total traffic volume. If the deviation exceeds the maximum threshold, it is marked as a traffic congestion stage; if it does not exceed the minimum threshold, it is marked as a smooth traffic stage; if the deviation is within the threshold range, it is marked as a congestion aggravation stage or a smooth traffic increase stage. When the parameters of the two stages are consistent, the accuracy deviation of the vehicle target movement parameter is obtained. The movement parameter is the vehicle speed or the vehicle trajectory offset distance. If the accuracy deviation shows an increasing trend, the parameters are adjusted.

[0007] As a preferred option, decision video acquisition parameters are also graded based on highway scene analysis within the highway section, and the traffic density deviation is obtained by comparing the traffic density of the same road section at different time periods. When the stage type is the congestion worsening stage, the video acquisition parameter adjustment direction is set to a fine-grained adjustment direction to increase acquisition accuracy. When the stage type is the smooth increase stage, the video acquisition parameter adjustment direction is set to a coarse-grained adjustment direction that does not require the current high precision. During the adjustment phase, if congestion worsens and stops, the adjustment will continue in the current direction, with the actual recognition accuracy deviation used as the criterion for termination. If smoothness increases and continues, the adjustment will continue in the current direction, with the percentage of usable frames in the actual captured video used as the criterion for termination.

[0008] It divides traffic stages in real time (congestion stage, smooth traffic stage, worsening congestion stage, and increasing smoothness stage) based on traffic density deviation, and adaptively adjusts video acquisition parameters according to the recognition accuracy deviation. It can automatically improve acquisition accuracy to ensure target details when congestion worsens, and appropriately reduce accuracy to save transmission bandwidth when smoothness increases. It achieves optimal adjustment of video hierarchical granularity under cloud-edge-device collaboration, avoids invalid data uploads, and ensures high-quality monitoring during critical periods.

[0009] Step 2: Obstruction Compensation Transmission Optimization: Record the acquisition parameters, extract video segments based on road sections and ensure that the video segments are continuous and adjacent, record the movement parameters and driving trajectory of vehicle targets and update them continuously. If the deviation value of the movement parameters of non-single vehicle targets continues to decrease and the driving trajectory is horizontal, then perform dynamic recognition, record the preliminary outline, distinguish the outline points based on the speed adjustment delay or trajectory offset delay, establish a real-time outline and compare it with the preliminary outline. If they are consistent, adjust the parameters. If they are inconsistent, continue to identify and adjust synchronously. As a preferred approach, when the deviation value of the movement parameters of any non-single vehicle target in the video continues to decrease and the driving trajectory is horizontal, the current video segment is analyzed to determine the type of the current segment. If it is in the stage of increased congestion or increased smoothness, it is determined that there is a risk of inappropriate fine-grained adjustment or loss of target frames. Long-tail effect dynamic recognition is then performed to complete target differentiation. Based on the preliminary outline of the vehicle target record involved in the long tail effect, multiple vehicle targets are dynamically identified when the long tail effect is formed. Based on the dynamic differences of vehicle target driving according to the real-time road surface record, the dynamic differences are divided into speed adjustment delay and trajectory offset delay. When dynamic differences occur, the contour points of the vehicles involved in the long-tail effect are distinguished, and a real-time target contour is established based on the dynamic consistency of the contour points. After the real-time target contour is established and compared with the preliminary contour, if they are consistent, the distinction is completed, and the video acquisition parameters are adjusted. When the adjustment time conflicts with the adjustment time of the congestion aggravation stage or the smoothness increase stage, the video acquisition parameters of the current adjustment time are adjusted first, and the adjustment process of the current congestion aggravation stage or smoothness increase stage continues after the long-tail effect disappears. If they are inconsistent, the distinction is not completed, and the contour point identification and search are continued while the video acquisition parameters are adjusted synchronously.

[0010] To address target loss or long-tail effects caused by multiple vehicles occluding each other, the system uses dynamic contour point differentiation and real-time contour comparison to quickly restore target identity and trajectory when vehicles occlude each other, avoiding tracking interruptions caused by target loss. At the same time, when adjusting parameter conflicts, it prioritizes long-tail effect compensation to ensure the real-time performance and accuracy of transmission optimization, significantly improving the stability of cloud-edge-device collaborative management.

[0011] Step 3: Lane Adaptive Localization and Recognition For road edge detection, extract auxiliary line area images, and mark missing line areas, faded areas, or lined areas according to the consistency of image parameters. When entering a missing line or faded area, determine the known movement trajectory based on the adjacent lined area, obtain the reserved distance value for meeting oncoming traffic, and construct the distance value range. Upon entering the area, the current deviation trajectory is recorded starting from the historical oncoming distance. If the deviation trajectory continues to change and the reserved distance value leaves the range, an anomaly is identified and an alert is uploaded. Based on the trend of exceeding or falling below the range, lane-inside occupancy alert or lane boundary deviation alert are issued respectively.

[0012] Preferably, the image parameter is a grayscale value. If the grayscale value of the auxiliary line area image is the same as that of the surrounding road surface area image, the current auxiliary line area image is marked as a missing line area. If the grayscale value of the auxiliary line area image and the surrounding road surface area image deviates from the image parameter deviation threshold, the current auxiliary line area is marked as a faded area. Areas other than those mentioned above are marked as wired areas. When entering a missing or faded area, the reference standard for the vehicle target trajectory offset in the video is missing. At this time, the known movement trajectory of the vehicle target is determined based on the adjacent wired areas, and the reserved workshop distance value when the vehicle target meets another vehicle is obtained by combining the known movement trajectory passage process. The distance value range is constructed based on the identified reserved workshop distance value. At the start of entering a missing or faded area, the reserved distance value of the historical adjacent passing time is used as the starting value to record the current vehicle target's offset trajectory. The reserved distance value is collected at subsequent adjacent passing times. If the current vehicle target's offset trajectory continues to change and the reserved distance value fluctuates in the direction of leaving the distance value range, the abnormal vehicle target offset in the highway video is identified and uploaded to the cloud edge for early warning.

[0013] When lane auxiliary lines are missing (missing line area) or faded (faded area), a dynamic reference system is constructed using the historical trajectory of adjacent lined areas and the pre-arranged distance for passing vehicles. This accurately determines whether a vehicle has deviated abnormally and distinguishes between trends exceeding the distance value range (risk of lane encroachment inside the lane) and trends below the distance value range (risk of deviating from the lane boundary). Thus, lane adaptive positioning is achieved even in the absence of clear lane markings, improving the traffic feedback and traffic detection capabilities of highway videos in complex road environments.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. A fine-grained adjustment strategy is adopted for the stage of increased congestion. By improving resolution and optimizing frame rate, the accuracy of video acquisition is improved. This is to solve the problem of insufficient recognition accuracy of moving parameters such as vehicle speed and trajectory deviation in traffic congestion and dense vehicle scenes. It avoids the defects of blurred target recognition and excessive data deviation in congested scenes, and ensures the accuracy of traffic data detection in complex congested road conditions. Secondly, a coarse-grained fine-tuning strategy is adopted for the smoothness enhancement stage. Without affecting the basic video acquisition quality and traffic detection requirements, the acquisition parameters are simplified, invalid video frame acquisition is reduced, the video data volume is significantly reduced, the storage pressure of edge devices is effectively alleviated, the bandwidth occupation of cloud-edge-device data transmission is reduced, and the problems of data redundancy and resource waste during off-peak hours of traditional technologies are eliminated.

[0015] 2. By extracting continuous video clips of adjacent road segments and iteratively updating vehicle movement parameters and driving trajectories in real time, it is possible to predict the risk of vehicle target loss and long-tail effect in advance, and realize problem identification in advance. For the target loss problem caused by the mismatch of granular adjustment during the congestion transition stage, through the initial vehicle contour recording, dynamic point recognition, and driving dynamic difference differentiation, overlapping and occluded vehicle targets can be accurately distinguished. The target reconstruction is completed by relying on the dynamic consistency comparison of the contour, effectively eliminating the recognition deviation caused by the long-tail effect. A parameter adjustment priority mechanism is set up so that when there is a conflict between granularity adjustment and long-tail effect compensation adjustment, target compensation optimization is prioritized to avoid target loss. After the interference is eliminated, the original granularity adjustment process is resumed. This ensures both the continuity of adaptive optimization of video acquisition parameters and the integrity and accuracy of vehicle target data in complex occlusion scenarios.

[0016] 3. Addressing the challenge of lacking a fixed trajectory reference standard in lane defect scenarios, this system innovatively constructs a standardized distance range based on the historical driving trajectories of vehicles in adjacent wired areas and the distance between oncoming vehicles. This provides a precise reference for vehicle deviation detection in defective road conditions, breaking through the scenario limitations of traditional algorithms. It can accurately distinguish the risk of lane encroachment inside the lane and the risk of deviating from the lane boundary based on the changing trend of vehicle deviation trajectory and the direction of fluctuation of the distance between vehicles. It also uploads the data to the cloud, edge, and terminal in real time to complete the early warning, realizing accurate identification and classification of abnormal vehicle traffic behavior in complex road conditions, and effectively reducing the probability of missed or false detections in traffic. Attached Figure Description

[0017] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0018] Figure 1 This is a block diagram illustrating the principle of the method of the present invention; Figure 2 This is a flowchart of step one of the present invention. Detailed Implementation

[0019] 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.

[0020] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] Please see Figures 1-2 As shown, a cloud-edge-device collaborative method for hierarchical processing and transmission optimization of highway video includes the following steps: Step 1: Adaptive adjustment of video grading granularity; Based on real-time highway traffic flow detection, video classification rules are set to complete video acquisition. After the classification rules are set, the video is divided into stages and transmitted to the cloud, edge, and device. Step 2: Obstruction compensation transmission optimization; Based on the highway identification stored in the cloud, edge, and terminal, occlusion compensation and transmission optimization are performed, and independent detection and trajectory determination of vehicle targets in the video are further carried out to ensure the efficiency of cloud-edge-terminal collaborative management. Step 3: Lane adaptive positioning and recognition; After completing the vehicle target detection in the video, the vehicle target is fuzzy identified based on the road environment where the vehicle target is located, and adaptive positioning is performed after the fuzzy identification is determined, so as to facilitate traffic feedback and traffic detection through highway video. The adaptive adjustment process of video grading granularity in step one is as follows: Based on the specifications of the video acquisition hardware devices for cloud-edge-device collaboration, the video acquisition parameters are determined. The specific video acquisition parameters are frame rate, resolution, and frame interval. Based on the current video acquisition parameters, video acquisition is performed on the current highway segment. At the same time, the video acquisition parameters are classified according to the analysis of the highway scene within the highway segment. The deviation of traffic density of the same road segment at different time periods during the video acquisition phase of the highway segment is obtained. The traffic density is expressed as the ratio of the traffic flow through the corresponding road segment per unit time to the total traffic flow of the road segment. When the traffic flow is smaller and the total traffic flow is larger, the traffic density is greater. If the traffic density deviation of the same road segment at different times exceeds the maximum value of the density deviation threshold range, it is inferred that the current time period has experienced increased congestion and is marked as a congested traffic phase. Conversely, if the traffic density deviation of the same road segment at different times does not exceed the minimum value of the density deviation threshold range, it is inferred that the current time period has not experienced increased congestion and is marked as a smooth traffic phase. Furthermore, based on the density deviation value being within the density deviation threshold range, the time period between the traffic congestion stage and the smooth traffic stage is marked as the congestion aggravation stage and the smooth traffic increase stage. That is, the stage between the traffic congestion stage and the smooth traffic stage is the smooth traffic increase stage, and conversely, the stage between the smooth traffic stage and the congestion stage is the congestion aggravation stage. Based on the smooth traffic flow stage and the congested traffic flow stage, when the video acquisition parameters in the corresponding stage are consistent, the discrepancy in the recognition accuracy of vehicle target movement parameters in the two stages of video acquisition is obtained. The movement parameters are represented by vehicle speed or vehicle trajectory offset distance, and the recognition accuracy is represented by the recognition accuracy of vehicle speed or offset distance, such as one meter per minute or one meter per half minute. If the accuracy deviation of vehicle target movement parameter recognition shows an increasing trend, it is inferred that video acquisition parameters need to be adjusted under the current road environment; if the accuracy deviation of vehicle target movement parameter recognition does not show an increasing trend, it is inferred that video acquisition parameters do not need to be adjusted under the current road environment. Adjust video capture parameters: Based on the currently identified smooth traffic phase and congested traffic phase, determine the phase type between the phases, such as the congestion worsening phase or the smoothness increasing phase. When the phase type is the congestion worsening phase, set the video acquisition parameter adjustment direction to a fine-grained adjustment direction, specifically to increase acquisition accuracy, such as increasing resolution. When the stage type is the smooth increase stage, the video acquisition parameter adjustment direction is set to coarse-grained adjustment direction, which means fine-tuning without requiring the current high precision and without affecting the quality of the acquired video. If congestion worsens during the adjustment phase, the adjustment will continue in the current direction and a termination point will be determined. The actual recognition accuracy deviation will be used as the termination criterion. If smoothness increases, the adjustment will continue in the current direction and a termination point will be determined. The percentage of usable frames in the actual video will be used as the criterion. After completing the adaptive adjustment, proceed to step two; The occlusion compensation transmission optimization process in step two is as follows: The system continuously collects highway video, records video acquisition parameters at each moment, analyzes the collected highway video images, and extracts highway video segments based on road sections, ensuring that the videos are continuous and adjacent. Record the identified vehicle targets at the beginning of the video segment capture, and record the movement parameters and driving trajectory of each vehicle target. Based on the continuous adjacent road segment videos, continuously update the vehicle target movement parameters and driving trajectory. If the deviation value of the movement parameters of any non-single vehicle target in the video continues to decrease, and the driving trajectory is horizontal, it is inferred that there is a risk of vehicle target loss forming a long-tail effect. The current video segment is analyzed to determine the corresponding stage type. If it is in the stage of increased congestion or increased smoothness, there is a risk of inappropriate fine-grained adjustment or target loss frames. At this time, long-tail effect dynamic recognition is performed to complete target differentiation. Based on the vehicle targets involved in the long tail effect, a preliminary outline is recorded, and multiple vehicle targets are dynamically identified when the long tail effect is formed. The differences in the driving dynamics of vehicle targets are recorded based on the real-time road surface, specifically divided into speed adjustment delay, trajectory offset delay, etc. When dynamic differences occur, the vehicle targets involved in the long-tail effect will be distinguished by contour points. Based on the dynamic consistency of the contour points, a real-time target contour will be established and compared with the preliminary contour. If they match, the distinction is completed. At this time, the video acquisition parameters will be adjusted to avoid target loss. If the adjustment time conflicts with the adjustment time of the congestion aggravation stage or the smoothness increase stage, the video acquisition parameters of the current adjustment time will be adjusted first. After the long-tail effect disappears, the adjustment process of the current congestion aggravation stage or the smoothness increase stage will continue. If there is a discrepancy, the distinction is completed, and the contour point identification and search are carried out simultaneously while adjusting the video acquisition parameters to improve the efficiency of contour point search. If the deviation value of the movement parameters of any non-single vehicle target in the video does not continue to decrease, or the driving trajectory is not horizontal, it is inferred that there is no risk of vehicle target loss forming a long tail effect. The current video capture stage type is analyzed. If it is in the stage of increased congestion or increased smoothness, the video acquisition parameter adjustment strategy of the current stage is continued to be adjusted. After avoiding interference from the long-tail effect, proceed to step three; The lane adaptive localization and recognition process in step three is as follows: Edge detection is performed on each road surface area based on highway video. Images of the road surface area are acquired and images of the road surface auxiliary line area are extracted. If the corresponding image parameters of the auxiliary line area image are consistent with those of the surrounding road surface area images, such as grayscale values, the current auxiliary line area image is marked as a missing line area. If the corresponding image parameters of the auxiliary line area image and the surrounding road surface area images deviate from the image parameter deviation threshold, the current auxiliary line area is marked as a faded area. Areas other than those in the above cases are marked as lined areas. When entering a defective or faded area, the vehicle target trajectory in the video deviates from the reference standard and is missing. The known movement trajectory of the vehicle target is determined based on the adjacent wired area, and the reserved inter-vehicle distance value when the vehicle target meets another vehicle is obtained by combining the known movement trajectory passage process. The distance value range is constructed based on the identified reserved inter-vehicle distance value. At the start of entering a defective or faded area, the reserved inter-vehicle distance value from previous adjacent oncoming traffic is used as the starting value to record the current vehicle's offset trajectory. Reserved inter-vehicle distance values ​​are collected at subsequent adjacent oncoming traffic moments. If the current vehicle's offset trajectory continues to change, and the reserved inter-vehicle distance value fluctuates in a direction away from the specified distance range, an abnormal vehicle offset is identified in the highway video, and an alert is sent to the cloud-edge system. Simultaneously, if the trend exceeds the specified distance range, a lane-inside risk warning is issued for the current vehicle; if the trend falls below the specified distance range, a lane-bound risk warning is issued for the current vehicle. It should be explained that the deviation trajectory does not change with the change of the road trajectory, but rather the actual deviation exceeds the change of the road trajectory. For example, when the road trajectory is a straight line, the vehicle target deviation trajectory is continuously generated. The method for obtaining the threshold in the above scheme is hereby disclosed: Traffic density deviation threshold (maximum value, minimum value) The core judgment threshold for the adaptive adjustment step of video grading granularity is obtained by combining historical traffic big data of the target highway segment, the designed traffic capacity of the segment, and the traffic flow fluctuation pattern at different times through statistical fitting and engineering measurement calibration.

[0022] The specific acquisition method is as follows: First, collect 24 / 7 traffic data for the target highway segment for more than 30 consecutive days, statistically analyze the traffic flow and maximum total traffic volume of the segment at different time periods within a unit of time, and calculate the traffic density and density deviation values ​​for each time period. Second, through big data clustering analysis, select the minimum density deviation under normal smooth traffic conditions and the maximum density deviation under critical congestion conditions to form an initial threshold range. Finally, through on-site road testing and calibration, eliminate abnormal data from special scenarios such as extreme weather and sudden accidents, correct the accuracy of the threshold range, and finally determine the maximum and minimum density deviation threshold values ​​suitable for the road segment. This is used to accurately classify the four traffic stages: smooth traffic, increased congestion, congested traffic, and increased smoothness. This threshold is a segment-specific dynamic threshold that can be continuously updated via the cloud to adapt to the long-term traffic flow patterns of the road segment.

[0023] Vehicle target movement parameter recognition accuracy deviation threshold; This threshold is used to determine whether video acquisition parameter adjustments need to be performed. It is derived from a combination of the general standard for vehicle recognition accuracy in the transportation industry and the scene-based measured accuracy error threshold.

[0024] The data acquisition method is as follows: Based on the accuracy indicators of the national standard for highway traffic vehicle detection, the basic allowable error range for vehicle speed recognition and trajectory offset distance recognition is determined. Under different traffic stages (smooth flow, congestion), multi-scenario tests are conducted using standard test vehicles, and basic data on recognition accuracy deviation under fixed acquisition parameters are recorded. The critical value at which the deviation shows an increasing trend is statistically analyzed and set as the recognition accuracy deviation judgment threshold. When the measured accuracy deviation exceeds this threshold and continues to increase, it is determined that the acquisition parameters need to be adjusted; otherwise, no adjustment is required. This threshold takes into account both industry standards and the characteristics of real-world scenarios, ensuring the rationality and accuracy of parameter adjustments.

[0025] Threshold for the percentage of valid frames captured in video; This threshold is the core threshold for terminating video granularity fine-tuning during the smooth enhancement phase. It is derived from the balance threshold between the need for effective video recognition and lightweight data transmission. It is obtained by statistically analyzing the minimum effective frame rate percentage required for highway traffic target recognition, combined with the cloud-edge-device transmission bandwidth capacity and the upper limit of edge device processing power, conducting multiple sets of comparative experiments. While ensuring that core functions such as vehicle trajectory recognition, traffic flow statistics, and anomaly detection are not affected, the video processing effect and transmission efficiency under different frame rate percentages are tested to determine the critical frame rate percentage that can balance recognition accuracy and low-redundancy transmission. This value is set as the effective frame rate percentage threshold, serving as the basis for terminating parameter fine-tuning during the smooth enhancement phase.

[0026] Image parameter deviation threshold (grayscale value deviation threshold); This threshold is the core judgment threshold for lane area classification and recognition, used to distinguish between faded lane areas and lined lane areas. It is derived from the quantification standard of pixel feature differences in highway pavement images. The acquisition method is as follows: A large number of real-world images of normal lined lanes, slightly faded lanes, severely faded lanes, and lanes without lines are collected. Core parameters such as grayscale values ​​and pixel contrast of each area are extracted. The baseline value of parameter deviation between normal lane areas and the surrounding pavement, as well as the minimum deviation threshold between faded lane areas and the surrounding pavement, are statistically analyzed. This threshold value is set as the image parameter deviation threshold. When the parameter deviation between the auxiliary line area and the surrounding pavement exceeds this threshold, it is determined to be a faded area; otherwise, it is determined to be a lane without lines, thus accurately achieving lane area classification.

[0027] Threshold range for vehicle passing distance; This threshold is used to determine the risks of abnormal vehicle deviation, lane occupation, and lane departure. It is derived from the fitting of highway safe passing distance standards and historical traffic data of the road segment. The acquisition method is as follows: First, the standard safe passing distance at different vehicle speeds is determined according to the "Highway Traffic Safety Standards"; second, historical data on vehicle passing distances under long-term normal traffic conditions of the target road segment are collected. Combined with the road segment's lane width, speed limit standards, and traffic flow characteristics, the normal passing distance fluctuation range of the road segment is fitted through big data statistics; this fluctuation range is set as the distance value range threshold. When the vehicle passing distance deviates from this range and the deviation trajectory continues to change abnormally, the abnormal vehicle traffic is accurately determined, and risk classification and early warning are achieved.

[0028] In summary, through a three-level collaborative process, a cloud-edge-device collaborative highway video hierarchical processing and transmission optimization system is constructed, which is adaptable to all scenarios, dynamic, and highly accurate. This system achieves a comprehensive technological upgrade in terms of intelligent video acquisition, efficient data transmission, accurate target recognition, and refined risk warning.

[0029] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A cloud-edge-device collaborative method for hierarchical processing and transmission optimization of highway video, characterized in that, Includes the following steps: Step 1: Determine the frame rate, resolution, and frame interval based on the video capture hardware specifications, and then capture the video. The deviation of traffic density on the same road segment at different time periods is obtained. The density is the ratio of the number of vehicles passing through per unit time to the total number of vehicles. If the deviation exceeds the maximum value of the threshold, it is marked as a traffic congestion stage. If it does not exceed the minimum value of the threshold, it is marked as a smooth traffic stage. If the deviation is within the threshold, it is marked as a stage of increased congestion or a stage of increased smoothness. When the parameters of the two stages are consistent, the vehicle target movement parameter recognition accuracy deviation is obtained. The movement parameter is the vehicle speed or the vehicle trajectory offset distance. If the accuracy deviation shows an increasing trend, the parameters are adjusted. Step 2: Record the acquisition parameters and extract consecutive adjacent segments from the video based on road sections: Record the movement parameters and driving trajectory of vehicle targets and update them continuously. If the deviation value of the movement parameters of non-single vehicle targets continues to decrease and the driving trajectory is horizontal, then perform dynamic identification, record the preliminary outline, distinguish the outline points according to the speed adjustment delay or trajectory offset delay, establish a real-time outline and compare it with the preliminary outline. If they are consistent, adjust the parameters. If they are inconsistent, continue to identify and adjust synchronously. Step 3: Detect the road surface edges, extract the auxiliary line area image, and mark the missing line areas, faded areas, or lined areas based on the consistency of image parameters: When entering a missing or faded area, the known movement trajectory is determined based on the adjacent wired area, the reserved distance value for meeting oncoming traffic is obtained, and a distance value range is constructed. After entering the area, the current deviation trajectory is recorded with the historical reserved distance value for meeting oncoming traffic as the starting value. If the deviation trajectory continues to change and the reserved distance value leaves the range, an anomaly is identified and an early warning is uploaded. Based on the trend of exceeding or falling below the range, a lane-inside lane occupation warning or a lane boundary deviation warning is issued respectively.

2. The cloud-edge-device collaborative highway video hierarchical processing and transmission optimization method according to claim 1, characterized in that, In step one, decision-making video acquisition parameters are classified based on the analysis of the highway scene within the highway segment. The traffic density deviation is obtained by comparing the traffic density of the same road segment at different time periods.

3. The cloud-edge-device collaborative highway video hierarchical processing and transmission optimization method according to claim 1, characterized in that, In step one, when the stage type is the congestion worsening stage, the video acquisition parameter adjustment direction is set to a fine-grained adjustment direction to increase acquisition accuracy. When the stage type is the smoothness improvement stage, the video acquisition parameter adjustment direction is set to a coarse-grained adjustment direction that does not require the current high precision.

4. The cloud-edge-device collaborative highway video hierarchical processing and transmission optimization method according to claim 1, characterized in that, In step one, if congestion worsens during the adjustment phase, the adjustment continues in the current direction, with the actual recognition accuracy deviation used as the criterion for termination. If smoothness increases, the adjustment continues in the current direction, with the percentage of usable frames in the actual captured video used as the criterion for termination.

5. The cloud-edge-device collaborative highway video hierarchical processing and transmission optimization method according to claim 1, characterized in that, In step two, when the deviation value of the motion parameters of any non-single vehicle target in the video continues to decrease and the driving trajectory is horizontal, the current video segment is analyzed to determine the corresponding stage type. If it is in the stage of increased congestion or increased smoothness, it is determined that there is a risk of inappropriate fine-grained adjustment or loss of target frames. Long-tail effect dynamic recognition is then performed to complete target differentiation.

6. The cloud-edge-device collaborative highway video hierarchical processing and transmission optimization method according to claim 1, characterized in that, In step two, based on the preliminary outline of the vehicle targets involved in the long tail effect, multiple vehicle targets are dynamically identified when the long tail effect is formed. Based on the dynamic differences of vehicle target driving recorded on the real-time road surface, the dynamic differences are divided into speed adjustment delay and trajectory offset delay. When dynamic differences occur, the outline points of the corresponding vehicle targets involved in the long tail effect are distinguished, and a real-time target outline is established based on the dynamic consistency of the outline points.

7. The cloud-edge-device collaborative highway video hierarchical processing and transmission optimization method according to claim 1, characterized in that, In step two, after establishing the real-time target contour and comparing it with the preliminary contour, if they match, the differentiation is completed, and the video acquisition parameters are adjusted. When the adjustment time conflicts with the adjustment time of the congestion aggravation stage or the smoothness increase stage, the video acquisition parameters of the current adjustment time are adjusted first, and the adjustment process of the current congestion aggravation stage or smoothness increase stage continues after the long tail effect disappears. If they do not match, the differentiation is not completed, and the contour point identification and search are continued while the video acquisition parameters are adjusted synchronously.

8. The cloud-edge-device collaborative highway video hierarchical processing and transmission optimization method according to claim 1, characterized in that, In step three, the image parameter is grayscale value. If the grayscale value of the auxiliary line area image is the same as that of the surrounding road surface area image, the current auxiliary line area image is marked as a missing line area. If the grayscale value of the auxiliary line area image and the surrounding road surface area image deviates from the image parameter deviation threshold, the current auxiliary line area is marked as a faded area. Areas other than those mentioned above are marked as lined areas.

9. The cloud-edge-device collaborative highway video hierarchical processing and transmission optimization method according to claim 1, characterized in that, In step three, after entering the missing line area or faded area, the reference standard for the vehicle target trajectory offset in the video is missing. The known movement trajectory of the vehicle target is determined based on the adjacent line area, and the reserved workshop distance value when the vehicle target meets another vehicle is obtained by combining the known movement trajectory passage process. The distance value range is constructed based on the identified reserved workshop distance value.

10. The cloud-edge-device collaborative highway video hierarchical processing and transmission optimization method according to claim 1, characterized in that, In step three, at the beginning of entering the missing line area or faded area, the reserved workshop distance value of the historical adjacent passing time is used as the starting value to record the current vehicle target's offset trajectory. The reserved workshop distance value is collected at subsequent adjacent passing times. If the current vehicle target's offset trajectory continues to change and the reserved workshop distance value fluctuates in the direction of leaving the distance value range, the abnormal vehicle target offset in the highway video is identified and uploaded to the cloud edge for early warning.

Images