A method, system, and equipment for real-time fusion processing of multi-source sensing data for smart roads under edge-cloud collaboration.

By employing a data processing method that integrates edge, cloud, and end devices, the problem of real-time fusion of multi-source sensing data for smart roads in unstable environments has been solved, thereby improving the accuracy and reliability of the data and ensuring the accuracy and reliability of traffic condition assessments.

CN123072201APending Publication Date: 2026-11-10SHANDONG EXPRESSWAY GRP CO LTD INNOVATION RES INST +1
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611314041.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-27
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

In existing technologies, the accuracy and reliability of real-time fusion of multi-source sensing data for smart roads are affected by the unstable output quality of sensing devices. Especially in rainy nights, backlighting, and occlusion environments, multiple sensing devices may make conflicting judgments on the position, speed, or lane of the same vehicle.

Method used

By adopting a data processing approach that integrates edge and cloud, a semantic processing mechanism for heterogeneous sensing data on the edge side, a multi-source real-time fusion mechanism for edge nodes, and a global state fusion mechanism in the cloud, and by combining cross-road segment traffic propagation relationships for anomaly correction, a closed-loop collaborative mechanism is established to achieve real-time data fusion and correction.

Benefits of technology

This improves the real-time fusion accuracy and reliability of multi-source perception data for smart roads, ensuring that the data conforms to the actual movement patterns of roads, and continuously corrects the data through subsequent traffic conditions across road segments, thereby enhancing the accuracy and reliability of traffic condition judgment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123072201A_ABST
    Figure CN123072201A_ABST
Patent Text Reader

Abstract

This application provides a method, system, and device for real-time fusion processing of multi-source sensing data for smart roads under edge-cloud collaboration. It relates to the field of road traffic information processing technology. Through semantic processing of heterogeneous sensing data at the edge, real-time fusion of multi-source data from edge nodes, global state fusion and anomaly correction in the cloud, and closed-loop collaborative adjustment between edge and cloud, it can perform real-time fusion of spatiotemporally heterogeneous and dynamically changing road multi-source sensing data under distributed processing conditions at the edge, cloud, and edge. When conflicts occur in local observations, it obtains traffic states that conform to the actual movement patterns of the road. Furthermore, it continuously corrects the previous fusion results using subsequent traffic states and control execution results across road segments, thereby improving the accuracy and reliability of real-time fusion of multi-source sensing data for smart roads.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of road traffic information processing technology, specifically to a method, system, and device for real-time fusion processing of multi-source sensing data for smart roads under edge-cloud collaboration. Background Technology

[0002] With the development of smart roads, vehicle-road cooperative systems, and intelligent transportation systems, the objects of road traffic status perception have gradually expanded from traditional cross-sectional traffic flow to multi-dimensional information such as vehicle location, vehicle speed, lane occupancy status, vehicle trajectory, traffic events, and road environment conditions. Correspondingly, roadside sensing equipment has evolved from single geomagnetic detectors or cameras to a combination of cameras, millimeter-wave radar, lidar, roadside units (RSUs), meteorological sensors, and vehicle-mounted terminals. Due to differences in sampling frequency, data format, spatial coordinate reference, detection accuracy, and applicable environment among different sensing devices, how to timely and accurately fuse multi-source heterogeneous traffic sensing data to obtain continuous and reliable road traffic operation status has become a crucial technical issue in intelligent road traffic monitoring and control.

[0003] Existing technologies have already researched the fusion of multi-source traffic data. For example, Chinese patent application CN110766936A discloses a traffic operation status perception method and system based on multi-source data fusion. Combining mobile data, fixed sensing data, and road network data can improve the problem of insufficient coverage from a single data source. However, its technical focus is mainly on the structured extraction, fusion, and road network matching of different categories of traffic status data. When multiple heterogeneous sensing devices simultaneously generate conflicting observations of the same traffic object, there is still room for improvement in how to determine the reliability of different data sources as the real-time environment changes, and how to further judge conflicting observations using the actual reachability of vehicles in specific road topologies. Another example is Chinese patent application CN113781767A, which discloses a traffic data fusion method and system based on multi-source sensing. This scheme acquires multi-source traffic data, records the multi-source data synchronously, adds spatiotemporal labels, converts it to a unified format, and then performs spatiotemporal matching and fusion through a data fusion model to obtain a fused traffic dataset. Therefore, existing solutions can already achieve spatiotemporal unification and fusion processing of multi-source traffic data.

[0004] However, in real-world smart road environments, the output quality of sensing devices is not constant. For example, the reliability of cameras may rapidly decrease in rainy nights, backlighting, and obstructed conditions, and different devices may make contradictory judgments about the position, speed, or lane of the same vehicle. Therefore, improving the accuracy and reliability of real-time fusion of multi-source sensing data for smart roads is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0005] In view of this, this application provides a method, system and device for real-time fusion processing of multi-source sensing data for smart roads under edge-cloud collaboration, in order to solve the problem that the output quality of sensing devices is unstable in the prior art, which reduces the accuracy and reliability of real-time fusion of multi-source sensing data for roads.

[0006] In a first aspect, embodiments of this application provide a method for real-time fusion processing of multi-source sensing data for smart roads under edge-cloud collaboration, including: Construct a semantic processing mechanism for heterogeneous sensing data on the edge side for road traffic objects, and convert the raw sensing data generated by different sensing devices into data units with unified time reference, spatial reference and road traffic semantics; Establish a multi-source real-time fusion mechanism based on road spatial association, data source credibility evolution, and traffic movement feasibility at edge nodes to perform object association, dynamic credibility evaluation, and conflict resolution on traffic semantic data from different edge nodes; Establish a global state fusion and anomaly correction mechanism for cross-road segment traffic propagation relationships in the cloud. Generate continuous traffic state at the road network level based on the local fusion results formed by multiple edge nodes, and use subsequent cross-road segment traffic states to reverse verify the previous fusion results. Establish a closed-loop collaborative mechanism that drives the dynamic adjustment of edge and terminal perception strategies based on the global correction results in the cloud, so that the data processing status of the terminal, edge, and cloud evolves in real time with the road traffic status.

[0007] In one possible implementation, the construction of a semantic processing mechanism for heterogeneous sensing data on the edge side, oriented towards road traffic objects, converts raw sensing data generated by different sensing devices into data units with a unified time reference, spatial reference, and road traffic semantics, including: The same vehicle that appears consecutively in the sensing areas of adjacent sensing devices is selected as the cross-device timing reference object; The relative time offset between sensing devices is determined based on the appearance time, departure time, traffic propagation time, and motion state continuity of the same reference object in different sensing devices, and the device timestamps are dynamically corrected. The time-corrected perception results are projected onto a pre-established road topology model to form traffic semantic data units for the detected road traffic objects; The degree of traffic status change is determined by the changes in the position, speed, lane status, and event tags of the same traffic object. When the traffic status is stable, only key status increment information is uploaded. When an abnormal status is detected, the upload frequency is increased and historical traffic status cached within a preset time window before the abnormality occurs is also uploaded.

[0008] In one possible implementation, projecting the time-corrected perception results onto a pre-established road topology model to form traffic semantic data units for the detected road traffic objects includes: The time-corrected perception results are projected onto a pre-established road topology model; The edge nodes do not continuously upload all the original sensing data directly to the edge nodes, but instead form traffic semantic data units for the detected road traffic objects. When an anomaly is detected, a traffic state change label is added to the corresponding traffic semantic data unit.

[0009] In one possible implementation, the establishment of a multi-source real-time fusion mechanism at the edge nodes, based on road spatial association, data source credibility evolution, and traffic movement feasibility, to perform object association, dynamic credibility evaluation, and conflict resolution on traffic semantic data from different edge nodes includes: Construct a local spatiotemporal association network with traffic objects as nodes, and associate multiple traffic semantic data as the same traffic object only when the state transition between multiple traffic semantic data satisfies the actual road traffic topology; A nonlinear dynamic evolution mechanism for data source credibility is established based on perception deviation, environmental degradation, and historical self-consistency. Based on the feasibility of road traffic movement and the competition relationship between candidate states, conflict data is reverse-verified to select the result with the most feasibility of traffic movement from multiple conflict results.

[0010] In one possible implementation, the nonlinear dynamic evolution mechanism for establishing data source credibility based on perceived deviation, environmental degradation, and historical self-consistency includes: Acquire the position, speed, and multi-dimensional evaluation parameters between the current output state of the traffic object from the sensing device and the preceding fused trajectory of the corresponding traffic object; Based on the multi-dimensional evaluation parameters, construct a perceptual loss reliability variable. ; Furthermore, reliable state decay orders are constructed based on the average perceived deviation level and average communication delay level within the historical time window. and trusted recovery order ; The data source credibility of the previous fusion cycle is recursively used to determine the dynamic data source credibility of the current fusion cycle. in: .

[0011] In one possible implementation, the reverse verification of conflict data based on the feasibility of road traffic movement and the competition relationship between candidate states, thereby selecting the result with the highest feasibility of traffic movement from multiple conflict results, includes: Multiple data source candidate states that are identified as the same traffic object are combined with historical fusion results to calculate traffic realizability deformation variables, and multiplicative coupling verification is performed from longitudinal movement, lane migration and road traffic constraints. The conflict competition index is calculated by combining the dynamic credibility of the data source and the state differences between candidate states, and the optimal value is selected as the main fusion state of the object. When the traffic realizability variables corresponding to all candidate states exceed the preset abnormal range, or the difference between the minimum conflict competition index and the second smallest conflict competition index is insufficient to form a stable judgment, the traffic object is marked as a conflict traffic object and uploaded to the cloud for cross-regional verification.

[0012] In one possible implementation, the establishment of a global state fusion and anomaly correction mechanism in the cloud for cross-segment traffic propagation relationships, generating continuous traffic states at the road network level based on the local fusion results formed by multiple edge nodes, and using subsequent cross-segment traffic states to perform reverse verification of the preceding fusion results, includes: The traffic status propagation chain is constructed for adjacent road sections according to road connection relationship, traffic flow direction and intersection connection relationship, and cross-regional trajectory connection is carried out for the same vehicle entering the management area of ​​another edge node. Causal consistency is determined based on the traffic conditions in different edge areas, and the local fusion state is verified and corrected in reverse based on the traffic propagation results of vehicles and traffic events actually observed in subsequent road segments. The corrected global fusion results are combined to form traffic status information at the vehicle object layer, lane status layer, road segment status layer, and road network layer, and further form traffic control demand identifiers.

[0013] In one possible implementation, the step of determining causal consistency based on traffic conditions in different edge areas and performing reverse verification and correction of the local fusion state based on the actual traffic propagation results observed in subsequent road segments of vehicles and traffic events includes: When an edge node determines that a specific road segment is severely congested, but the upstream vehicle status, downstream outbound traffic flow, and adjacent road traffic status of that road segment do not show any changes that match the congestion level, the global consistency of that local congestion status is reduced. When an edge node detects an abnormal parking, and subsequently detects a decrease in vehicle speed behind the vehicle, an increase in the queue behind it, vehicles in adjacent lanes detouring, and a redistribution of lane traffic, the global credibility of the corresponding abnormal parking event is increased. For data marked as conflicting traffic objects, the cloud performs reverse tracking of different candidate states based on their location and time of appearance in subsequent edge areas; If a candidate state can form a road topological continuity relationship with the subsequently observed trajectory, the result is used to reverse the candidate state judgment of the original edge node, and the corresponding correction result is fed back to the original edge node.

[0014] In one possible implementation, the establishment of a closed-loop collaborative mechanism that drives the dynamic adjustment of end-side and edge-side perception strategies based on cloud-based global correction results, enabling the data processing status of the end, edge, and cloud levels to evolve in real time with road traffic conditions, includes: Periodically compare the local traffic status uploaded by each edge node with the global corrected traffic status formed in the cloud to determine the error type and duration of different types of sensing devices in different road areas, and generate edge node feedback parameters. The frequency of data uploads to the edge, the fusion cycle of the edge side, and the cross-regional verification frequency of the cloud are dynamically adjusted based on the feedback results from the cloud and the current road traffic conditions. After implementing traffic control based on cloud-integrated traffic conditions, the actual traffic change trend after the control is executed is collected and compared with the traffic response direction predicted before the control. When there is a discrepancy, the corresponding road area is marked as a suspected deviation area of ​​the fusion result and a re-verification is triggered.

[0015] Secondly, embodiments of this application provide a real-time fusion processing system for multi-source sensing data of smart roads under edge-cloud collaboration, including: The edge semantic processing module is used to build an edge heterogeneous sensing data semantic processing mechanism for road traffic objects, and to convert the raw sensing data generated by different sensing devices into data units with unified time reference, spatial reference and road traffic semantics. The edge fusion module is used to establish a multi-source real-time fusion mechanism at edge nodes based on road spatial association, data source credibility evolution, and traffic movement feasibility. It performs object association, dynamic credibility evaluation, and conflict resolution on traffic semantic data from different edge nodes. The cloud-based global fusion module is used to establish a global state fusion and anomaly correction mechanism for cross-road segment traffic propagation relationships in the cloud. It generates continuous traffic states at the road network level based on the local fusion results formed by multiple edge nodes, and uses subsequent cross-road segment traffic states to reverse verify the previous fusion results. The closed-loop collaborative adjustment module is used to establish a closed-loop collaborative mechanism that drives the dynamic adjustment of the perception strategies of the end and edge sides based on the global correction results in the cloud, so that the data processing status of the end, edge and cloud evolves in real time with the road traffic status.

[0016] Thirdly, embodiments of this application provide an electronic device, including: processor; Memory; And a computer program, wherein the computer program is stored in the memory, the computer program including instructions that, when executed by the processor, cause the electronic device to perform the method described in any one of the first aspects.

[0017] Fourthly, embodiments of this application provide a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform the method described in any one of the first aspects.

[0018] In this embodiment, by semantic processing of heterogeneous sensing data at the edge, real-time fusion of multi-source data from edge nodes, global state fusion and anomaly correction in the cloud, and closed-loop collaborative adjustment between the edge and cloud, real-time fusion of spatiotemporally heterogeneous and dynamically changing road multi-source sensing data can be achieved under distributed processing conditions at the edge, cloud, and edge. When conflicts occur in local observations, traffic states that conform to the actual movement patterns of the road can be obtained. Furthermore, subsequent traffic states and control execution results across road segments can be used to continuously correct the previous fusion results, thereby improving the accuracy and reliability of real-time fusion of multi-source sensing data for smart roads. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating a real-time fusion processing method for multi-source sensing data of smart roads under edge-cloud collaboration, provided as an embodiment of this application; Figure 2 This application provides a schematic diagram of the dynamic reliability evolution curve of a data source, as shown in the embodiments of the present application. Figure 3 A scatter plot for resolving candidate observation conflicts is provided in an embodiment of this application. Figure 4 This is a schematic diagram of a real-time fusion processing system for multi-source sensing data of smart roads under edge-cloud collaboration, provided as an embodiment of this application.

[0021] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0022] To better understand the technical solution of this application, the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0023] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0024] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0025] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0026] Example 1 See Figure 1 The real-time fusion processing method for multi-source sensing data of smart roads under edge-cloud collaboration provided in this embodiment includes: S101, Construct a semantic processing mechanism for heterogeneous sensing data on the edge side for road traffic objects, and convert the raw sensing data generated by different sensing devices into data units with unified time reference, spatial reference and road traffic semantics.

[0027] This embodiment is applied to a smart highway environment equipped with various road sensing devices, including cameras, millimeter-wave radar, lidar, geomagnetic detectors, roadside units (RSUs), meteorological sensors, and vehicle-mounted terminals. Each road sensing device is communicatively connected to its corresponding edge processing unit, and multiple edge processing units are further communicatively connected to edge computing nodes set up in the road area. Each edge computing node is connected to the cloud platform through a communication network.

[0028] Among them, cameras are used to acquire images and video information of vehicles and road maintenance personnel; millimeter-wave radar is used to acquire information on the distance, radial velocity and direction of movement of traffic objects; lidar is used to acquire spatial point clouds and target positions of traffic objects; geomagnetic detectors are used to detect the passing status of vehicles and lane occupancy status; roadside units (RSUs) are used to acquire vehicle position, speed, heading and vehicle identification information sent by networked vehicles; and meteorological sensors are used to acquire environmental status information such as rainfall, visibility, road ambient brightness and temperature.

[0029] Based on the aforementioned data acquisition equipment, this embodiment first continuously collects information on road traffic objects and road traffic environment through multiple types of sensing devices installed on the road, and then forms traffic semantic data with a unified time reference and road spatial semantics at the end-side nodes.

[0030] Taking a fusion period of 100ms as an example, each camera, millimeter-wave radar, lidar, and RSU acquires road traffic data according to its respective acquisition frequency. For each sensing device, its local acquisition time is recorded as... The edge or end-side system maintains a unified reference clock. For the same traffic object appearing consecutively within the sensing areas of adjacent sensing devices, the time detected by different devices and the road propagation time are recorded, and the device time offset is obtained from multiple consecutive reference traffic objects. For sensing devices with different sensing locations, the corresponding traffic propagation time is determined based on the road path distance and movement state of the reference traffic object between the two sensing locations, and the traffic propagation time is subtracted from the observation time difference recorded by the two sensing devices. The remaining time difference is used as the relative time offset estimate between the devices. For example: Camera A is located at 0m on the road, and Radar B is located at 20m on the road. The same vehicle appears in A at 10:00:00.000 and in B at 10:00:05.000. These 5 seconds cannot be directly used as the time offset because the vehicle itself needs time to travel 20 meters, i.e., traffic propagation time. The time offset should be 5 seconds minus the traffic propagation time. In this embodiment, this traffic propagation time can be obtained by dividing 20 meters by the average speed of the vehicle.

[0031] Let the first The estimated time offset obtained by each device in the current calibration period is: The unified time corresponding to the device data is then represented as: In this embodiment, the time offset is not determined based on a single target passage event, but rather updated based on a time offset sequence formed by at least five most recent valid reference objects. When the time offset of one reference object is significantly greater than that of the others, that reference object can be excluded as an anomaly. For data reception delays caused by network jitter, the data generation time and the actual reception time of the edge node are also recorded, and the difference between the two is used to determine the data transmission delay level (described later). Through the above processing, vehicle positions, speeds, and event states from different devices can be mapped to a unified fusion period.

[0032] Establish a road topology model, which at least stores the road number, driving direction, lane number, lane center line and lane boundary, road speed limit, intersection connection relationship, entrance lane and exit lane relationship, ramp merging or diverging relationship, stop line position, traffic signal phase, and the effective sensing range of the sensing device.

[0033] The target locations output from different sensing devices are transformed to a unified road coordinate system, and the targets are mapped to corresponding road sections and lanes. Traffic semantic data units are generated for each detected traffic object, for example: in: For identifying traffic objects; The longitudinal position of the traffic object along the centerline of the road; The lateral offset of a traffic object relative to the centerline of its lane; For speed; In terms of direction of motion; Number the lane it belongs to; For traffic object type; The data source identifier for generating this semantic data; The raw perceived quality information output by the device itself.

[0034] Set horizontal offset One of the objectives is to determine in subsequent steps whether a vehicle has approached the lane boundary, thereby avoiding relying solely on discrete lane numbers to determine whether a vehicle can complete lane migration within a fusion cycle. For cases where cameras, millimeter-wave radar, or lidar cannot directly output all the aforementioned variables, only the fields that can be determined are output; missing fields are recorded as null values, and artificially created data is not used for filling in.

[0035] Edge nodes establish state caches for the same traffic object with a duration of 2–5 seconds. When the speed change rate, position change rate, and lane status of a vehicle are all within a stable range for multiple consecutive fusion cycles, only the object identifier, position increment, speed, and timestamp of the current traffic object are uploaded. When at least one of the following states occurs: sudden deceleration, continuous lane changes, abnormal parking, driving in the wrong direction, road maintenance personnel encroaching on the motor vehicle lane, road obstacles, or rapid increase in vehicle queue length, the edge node increases the data upload frequency of the corresponding object and simultaneously uploads the historical state cached before the anomaly occurred.

[0036] S102 establishes a multi-source real-time fusion mechanism at edge nodes based on road spatial association, data source credibility evolution, and traffic movement feasibility, and performs object association, dynamic credibility evaluation, and conflict resolution on traffic semantic data from different end-side nodes.

[0037] Edge nodes receive traffic semantic data uploaded by multiple end-side nodes in their management area. First, they associate data that may represent the same traffic object, then dynamically determine whether different sensing devices are currently reliable, and finally select the state that can be truly realized in the sense of traffic movement when multi-source observation conflicts occur.

[0038] For connected vehicles, if the RSU data has a legally usable anonymous vehicle identifier, the anonymous vehicle identifier is used first in conjunction with the road location for association. For traffic objects for which a unique vehicle identifier cannot be obtained, an object association threshold is constructed based on road segment consistency, road connectivity, direction of movement, longitudinal position, speed, lane position, and preceding trajectory.

[0039] For example, the If the longitudinal predicted position difference between an object detected by a device and an existing trajectory is no greater than a preset association distance, and the vehicle's movement direction conforms to the permitted traffic direction of the road, and there is an actual road connection relationship from the preceding lane to the candidate lane, then the object is considered as a candidate observation of the same object.

[0040] After completing the traffic object association, for the same traffic object, candidate states from cameras, millimeter-wave radar, and lidar may be obtained simultaneously. Then, based on the deviation of the current traffic object state output by the device from the corresponding traffic object's previous fused trajectory in terms of position, speed, and lane state, the degree of perception deviation is obtained. Environmental degradation factors are determined based on rain, snow, fog, nighttime conditions, backlighting, and road flooding. The degree of communication delay is determined based on the time difference between the data generation time and the edge node reception time. The degree of perceptual loss is determined based on the target occlusion ratio, the number of consecutive frame drops, or the degree of object information loss. The traffic complexity factor is determined based on the current traffic density of the road area, the proportion of intersecting traffic flows, and the degree of vehicle queuing. The degree of historical continuity consistency is obtained based on the extent to which the output results of the sensing device remain consistent with the edge fusion state over the most recent fusion cycles. The specific process is as follows: 1. Perceived deviation degree Let the first There are longitudinal positional deviations between the candidate state currently output by each sensing device and the predicted state of the preceding fused trajectory for the same traffic object. Speed ​​deviation and lane deviation Determine the overall deviation: in: This is the reference distance for positional deviation; The speed deviation is the reference speed; This is the absolute value of the difference in lane number between the current candidate lane and the preceding merged lane.

[0041] Again After truncation and normalization, we get: in: This is to pre-set a deviation from the baseline. For example, in urban road applications, the deviation can be adjusted based on the on-site road calibration results. Set to 1-3m, The speed is set to 2-5 m / s; however, the specific values ​​mentioned above are not necessary limitations of this invention.

[0042] 2. Environmental degradation factors Rainfall intensity, low visibility, low light intensity, or backlight intensity are each normalized to: Considering that multiple adverse environmental conditions may occur simultaneously, this embodiment does not simply calculate an average, but instead uses the product of the remaining effective environmental states to determine the environmental degradation factor: Therefore, as long as a clearly unfavorable environmental condition occurs, This means that degradation will increase; when multiple adverse environmental conditions occur simultaneously, the degree of degradation will be further enhanced. Different types of equipment can select different environmental variables. For example, cameras mainly use changes in rainfall, visibility, and brightness; millimeter-wave radar can mainly use heavy rainfall and target density; lidar can mainly use rain, fog, and the effective echo ratio of point clouds.

[0043] 3. Communication delay level Let the first The data generation time for each device is The edge node receiving time is The communication delay is ,pass Obtain the degree of communication delay.

[0044] 4. Degree of perceptual loss Perception of the degree of loss The determination is based on target occlusion, frame drops, and missing fields. For example, let the current target occlusion ratio be... ,recent The effective output ratio of data in each fusion cycle is: Then it can be expressed as: in: and All are limited to the range of 0 to 1.

[0045] 5. Traffic complexity factor Based on the current vehicle density in the sensing area Intersecting traffic flow degree and queuing occupancy Normalize to 0-1 respectively to obtain the traffic complexity factor: This avoids using a single traffic flow indicator to describe complex traffic conditions.

[0046] 6. Degree of historical continuity and consistency Regarding the first The device recently For each effective fusion period, calculate the normalized deviation of the observed state relative to the final fusion state. ,get: therefore: The closer the continuous output results of the device are to the final fusion state, The closer it is to 1. In this embodiment, it can be taken as... One integration cycle.

[0047] Furthermore, the first variable is calculated from the above variables. The sensing failure variables of a single sensing device: in: For the first The number of cycles in which each device has continuously and stably output valid data up to the current fusion cycle; Enhancement coefficient for continuous and stable operation.

[0048] The By using a multiplicative coupling method, the current perception deviation, environmental degradation, transmission delay and perception information loss are nonlinearly amplified. At the same time, the continuous and stable output of the device forms reverse suppression, so that the device will produce a significant unreliable state change when the environment suddenly deteriorates or the output suddenly becomes abnormal.

[0049] To avoid the sensing devices from operating continuously and stably for a long time To continuously increase the number of consecutive stable output cycles, this embodiment sets a preset upper limit. In this embodiment ,in In this embodiment, the maximum number of stable periods is preset. The value can be between 10 and 50. This sets a preset upper limit for the number of consecutive stable output cycles. When the number of consecutive stable output cycles reaches After that, it will no longer increase with continuous stable working time. When it is detected that the current output of the sensing device does not meet the preset stability condition, it will... Reset to 0. The preset stability condition means that the current sensing deviation, communication latency, and sensing loss of the sensing device are all within the corresponding normal range. This embodiment provides a directly implementable method for determination: in: This represents the maximum enhancement coefficient. To stabilize the periodic scale parameters. For example, it can be set. , As the number of consecutive and stable working cycles of the equipment increases, Gradually approaching Instead of growing indefinitely.

[0050] Further obtain the Average sensing deviation level of each sensing device within a historical time window and average communication delay level : Construct the order of reliable state decay respectively and trusted recovery order : When the current perceived deviation and communication latency are lower than the corresponding historical average levels, the reliable recovery order is... The increase accelerates the decay of untrusted states; when the current perception deviation or communication delay is higher than the corresponding historical average level, the corresponding recovery item will no longer be increased, thereby avoiding excessively strong trusted recovery effects when the sensing device is in an abnormal state.

[0051] The data source reliability of the previous fusion cycle is recursively derived into the dynamic data source reliability of the current fusion cycle using the following formula: in: For systems starting up for the first time and without historical data, the following settings can be configured uniformly: Alternatively, after the sensing device has undergone on-site calibration, the initial confidence level of the device can be set to the normalized value of the calibration accuracy. This is especially important when historical samples are insufficient. When calculating the average value for each period, the system uses all currently available valid periods and does not require waiting for a complete historical window before performing the fusion, thus solving the problem of not being able to calculate the historical average value during the system startup phase.

[0052] When the current perceived deviation or communication latency suddenly exceeds the device's historical normal level, the order of trust state decay occurs. The increase causes the preceding trusted state to decay rapidly through power-law deformation; when the current perceived deviation of the device decreases below the historical level and remains consistent for multiple consecutive fusion cycles, the trusted recovery order increases. The increase in credibility causes the untrusted state to decay rapidly, thereby gradually restoring the credibility of the data source.

[0053] like Figure 2 As shown, when the camera enters rainy night or occlusion conditions, both the environmental degradation factor and the degree of perception deviation increase simultaneously, causing its dynamic reliability to drop rapidly. When the rainy night and occlusion conditions are lifted and stable observations are continuously output, the historical consistency improves, allowing the dynamic reliability to gradually recover. Millimeter-wave radar and lidar exhibit different reliability evolution curves due to varying degrees of influence from their respective environments. Figure 2 This is used to illustrate the above trend of credibility evolution, and does not limit the specific curve values.

[0054] like Figure 3 As shown, the same vehicle may generate multiple candidate states from the camera, millimeter-wave radar and lidar respectively within the same fusion cycle.

[0055] In this embodiment, traffic objects are defined. The The candidate states are: in: Indicates the longitudinal position of the road; For speed; Lane number.

[0056] The edge node first reads the state of the previous fusion cycle: And based on the horizontal position of the previous cycle Determine whether the vehicle has entered the lane boundary buffer zone.

[0057] For the j-th candidate state, construct traffic realizability deformation variables: in: This is the longitudinal distance reference quantity, used to suppress excessively small normalization denominators at low speeds or when stationary. It can be set to 0.5–2m. This refers to the number of lanes a traffic object needs to cross to move from the lane it was in in the previous fusion cycle to the j-th candidate lane. Specifically, .

[0058] The time is determined based on the current vehicle speed, the length of the fusion cycle, and the road structure. The maximum reasonable number of lane moves allowed within the specified timeframe. For example, when the vehicle was far from the lane boundary in the previous cycle: When a vehicle has already entered the lane boundary buffer zone in the previous cycle, or shows a clear lane-changing trend for two or more consecutive cycles: For longer fusion cycles, the shortest reasonable lane change time can also be preset. The process can be expanded, but vehicles are generally not allowed to cross multiple full lanes instantaneously within a fusion period of 100-200ms.

[0059] This is a road connection legality factor, with a value ranging from 0 to 1. One possible implementation is: when the candidate state fully conforms to the lane connection relationship, road traffic direction, and the current traffic signal permitted direction, When the road topology allows but is currently in a critical migration or uncertain state, When the candidate state corresponds to a road that cannot be directly reached from the current lane or completely violates the road's direction of travel, Determined directly by querying the road topology database. This eliminates the need for subjective scoring by humans.

[0060] The equivalent velocity required for the vehicle to move from the previous merged position to the j-th candidate position is defined as: in: This represents the time interval between the current fusion cycle and the previous fusion cycle.

[0061] The current permissible speed is determined by combining the speed limit of the road, the current lane traffic conditions, and the traffic signal conditions; Determined based on road speed limits and current traffic conditions. In one implementation: in: Speed ​​limits on roads; The reasonable upper limit speed for the current traffic flow in the lane can be determined based on the recent vehicle speed distribution. and This allows for short-term fluctuations. For areas before red light stop lines or construction speed limit zones, the corresponding traffic control restrictions will be added. .

[0062] This is the velocity reference constant, which can be set to 0.5–2 m / s. This is a road conflict amplification factor, which can be preset according to the complexity of the road area. A lower value is used for ordinary straight road sections, and a higher value is used for intersections, merging zones, diverging zones, or ramp areas. For example, an ordinary straight road section can be set to... Merging, diverging, and ramp areas can be equipped with... The core area of ​​the intersection can be set up The specific parameters mentioned above can be adjusted through road calibration.

[0063] During implementation, to prevent numerical overflow caused by consecutive multiplications, direct calculation may be avoided. And calculate its logarithmic form: in: These represent the three multiplicative factors mentioned above.

[0064] Therefore, by multiplicatively coupling the three constraints of longitudinal motion continuity, lane migration rationality, and road traffic feasibility, candidate states that violate road traffic motion laws are nonlinearly amplified. After obtaining the traffic feasibility deformation variables of multiple candidate states, the edge nodes further combine the dynamic credibility of the data source and the degree of conflict between multiple candidate states to construct the conflict competition index of the j-th candidate traffic state: Where: M is the number of data sources that generate valid candidate observations for the same traffic object within the current fusion period. ; The state difference between the j-th candidate state and the k-th candidate state is determined by the longitudinal position difference, speed difference, and lane difference. The candidate state difference normalization benchmark value can be set to 1 in this embodiment; It is a positive stability constant, which can be set to 0.05 to 0.2.

[0065] In this embodiment, The method of determination is as follows: in, The normalized baseline distance is the longitudinal position difference of the candidate states. Normalized baseline velocity for candidate state velocity differences: This converts longitudinal position difference, speed difference, and lane difference into dimensionless quantities, avoiding inconsistencies in dimensions caused by directly adding different physical quantities.

[0066] Similarly, to avoid large products from multiple candidate states, the implementation can follow these steps: The calculation is performed, and then the results are compared based on their logarithmic values. Since the logarithmic function is monotonically increasing, it does not change the order of the candidate states.

[0067] Finally, based on the conflict competition index of each candidate state, the following is determined: and will This represents the primary fusion state in the current fusion cycle.

[0068] like Figure 3 As shown, assuming the vehicle was in lane 2 in the previous cycle, the three devices are currently generating candidate results near lanes 1, 2, and 3, respectively. Even if two anomalous candidates are spatially close to each other, as long as the vehicle cannot cross from the previous state to the corresponding lane within one fusion cycle, its... This will significantly increase the chances of being incorrectly selected simply because multiple devices are close in space.

[0069] when At that time, due to the original conflict competition formula Since the value is 0, the conflict competition index is not used. In this case, only the conflict competition index for this candidate state is calculated. .like If the candidate state is positive, then the candidate state is temporarily accepted; otherwise, it is marked as an abnormal candidate state.

[0070] when Let the minimum and second-minimum conflict competition indices be respectively and Calculate the relative separation degree between the two: in: To prevent small positive numbers with a denominator of 0.

[0071] If all candidate states or If the current local observations are insufficient to stably determine the main fusion state, then it is considered that the current local observations are insufficient. For example, historical road data can be used to verify the fusion. , The initial discrimination range is set, and then calibration is performed using actual road data. At this time, edge nodes are not forced to select results, but instead upload the candidate state set, the dynamic credibility of each candidate, the traffic feasibility deformation variable, and the corresponding timestamp to the cloud.

[0072] It should be noted that, in order to ensure that the above formula can be directly implemented, this embodiment first performs dimensionless processing on the state variables entering the formula. For any positive outlier that needs to be normalized... The following truncation and normalization method is adopted: in: and These are the lower limit of normal and the upper limit of abnormality for the corresponding states, which can be predetermined using road equipment calibration data or historical operating data; after normalization, the values ​​of each abnormal state quantity are all between 0 and 1.

[0073] S103 establishes a global state fusion and anomaly correction mechanism in the cloud for cross-road segment traffic propagation relationships. It generates continuous traffic states at the road network level based on the local fusion results formed by multiple edge nodes, and uses subsequent cross-road segment traffic states to reverse verify the previous fusion results.

[0074] After the edge nodes complete local fusion, they upload vehicle continuous trajectories, road traffic flow, average speed, lane occupancy, queue length, and conflict objects that cannot be stably determined at the edge to the cloud.

[0075] Based on the road topology, the cloud connects multiple adjacent edge nodes according to the actual travel direction of vehicles. For example, when a vehicle enters road segment B from road segment A via an intersection, the vehicle exit status of the corresponding edge node in A is associated with the vehicle entry status of the corresponding edge node in B. For the same traffic object, a cross-regional trajectory chain is formed based on the time of leaving the previous edge area, the time of entering the next edge area, the road connection direction, the longitudinal distance, and the speed continuity. For congestion or accident events, traffic state propagation relationships are established based on upstream vehicle speed decreases, queue length increases, vehicles in adjacent lanes detouring, and downstream outbound flow decreases.

[0076] The cloud platform uses actual traffic conditions observed in subsequent road areas to verify data that conflict with the previous edge area.

[0077] For example, edge nodes at time For a vehicle, candidate states A and B are maintained. If the deduction continues according to candidate state A, the vehicle should enter road segment B; if the deduction continues according to candidate state B, the vehicle should enter road segment C. If a traffic object with a continuous trajectory to the vehicle is subsequently detected in road segment B, but no corresponding object is found in road segment C, the global consistency of candidate state A is increased, and A is used as the correction result of the previous state.

[0078] Similarly, when an edge node determines that road congestion has occurred, but its upstream, downstream, and adjacent roads do not show any corresponding changes in traffic flow propagation, the cloud reduces the credibility of this local congestion event. Conversely, when abnormal parking occurs, and subsequent cycles show deceleration of vehicles behind, queue lengthening, and detours by adjacent lanes, the global credibility of the abnormal parking event is increased. The cloud feeds back these correction results to the original edge node, which can be used to update historical self-consistency.

[0079] The cloud ultimately generates at least four levels of data. First, the traffic object layer, including vehicle trajectories, vehicle speeds, lane changes, and abnormal driving behavior. Second, the lane status layer, including lane flow, average speed, lane occupancy, and queue length. Third, the road segment status layer, including congestion levels, effective capacity, abnormal parking conditions, and road accident conditions. Fourth, the road network layer, including traffic flow direction, congestion spread trends, accident impact areas, and key traffic control areas.

[0080] The aforementioned traffic conditions are further provided to traffic signal control, ramp control, variable message signs, traffic guidance, and vehicle-road cooperative early warning systems.

[0081] S104 establishes a closed-loop collaborative mechanism that drives the dynamic adjustment of end-side and edge-side perception strategies based on the global correction results in the cloud, enabling the data processing status of the end, edge, and cloud levels to evolve in real time with the road traffic status.

[0082] After obtaining the global traffic status in the cloud, the current processing does not end. Instead, the global correction results are reused in the next fusion cycle at both the edge and endpoint sides. The cloud compares the local edge fusion results with the final global correction results.

[0083] If a camera is found to continuously deviate from its target position in a rainy night environment, the environmental degradation status under the corresponding environmental conditions will be increased, so that the device will be able to perform better in the next cycle. Increase. If a millimeter-wave radar is found to continuously experience target mismatches, reduce its historical continuity consistency. If cross-device time misalignments occur consecutively, the device time offset is updated. Therefore, the cloud-based correction results can directly affect the reliability of the data source in the next cycle, rather than just being used for retrospective recording.

[0084] For areas experiencing accidents, abnormal parking, or rapid congestion, the upload frequency of corresponding sensing devices is increased, while the edge fusion cycle is shortened. For example, the fusion cycle can be set to 100–200 ms under normal conditions; in abnormal areas, it can be shortened to 50–100 ms. For traffic events whose impact extends across multiple edge nodes, the cross-regional processing frequency in the cloud is increased. This allows the frequencies of edge data upload, edge fusion, and cross-regional processing in the cloud to be dynamically adjusted according to changes in traffic conditions.

[0085] After the cloud-based system outputs signal timing adjustments, ramp control, traffic guidance, or vehicle-road cooperative warnings based on the fusion results, it continues to collect data on the traffic conditions after control. For example, the cloud-based system determines that queues at intersections are increasing and implements adjustments to the green light timing. If the queue length at the corresponding approach lane decreases as expected after control, it indicates that the preceding traffic condition identification and control requirements are basically consistent. If the queue condition after control is inconsistent with the predicted direction for a long period, the corresponding area is marked as a suspected fusion deviation area, and the frequency of multi-source data cross-validation in that area is increased.

[0086] As can be seen from the above embodiments, the embodiments of this application, through semantic processing of heterogeneous sensing data on the edge side, real-time fusion of multi-source road sensing data with dynamic changes in reliability at the edge and cloud, can perform real-time fusion of multi-source road sensing data with spatiotemporal heterogeneity under distributed processing conditions of the edge, cloud, and cloud. When conflicts occur in local observations, traffic states that conform to the actual movement laws of the road can be obtained, and the subsequent traffic states and control execution results of cross road segments can be used to continuously correct the previous fusion results, thereby improving the accuracy and reliability of real-time fusion of multi-source sensing data of smart roads.

[0087] Example 2 Corresponding to the above embodiments, this application also provides a real-time fusion processing system for multi-source sensing data of smart roads under edge-cloud collaboration.

[0088] See Figure 4 The edge-cloud collaborative smart road multi-source perception data real-time fusion processing system 20 provided in this embodiment includes: The edge-side semantic processing module 201 is used to construct an edge-side heterogeneous sensing data semantic processing mechanism for road traffic objects, which converts the raw sensing data generated by different sensing devices into data units with unified time reference, spatial reference and road traffic semantics.

[0089] The edge fusion module 202 is used to establish a multi-source real-time fusion mechanism at the edge node based on road spatial association, data source credibility evolution, and traffic movement feasibility, and to perform object association, dynamic credibility evaluation, and conflict resolution on traffic semantic data from different end nodes.

[0090] The cloud-based global fusion module 203 is used to establish a global state fusion and anomaly correction mechanism for cross-road segment traffic propagation relationships in the cloud. It generates continuous traffic states at the road network level based on the local fusion results formed by multiple edge nodes, and uses subsequent cross-road segment traffic states to reverse verify the previous fusion results.

[0091] The closed-loop collaborative adjustment module 204 is used to establish a closed-loop collaborative mechanism that drives the dynamic adjustment of the perception strategies of the end side and edge side based on the global correction results in the cloud, so that the data processing status of the end, edge and cloud evolves in real time with the road traffic status.

[0092] It should be noted that the specific content involved in the embodiments of this application can be found in the description of the above method embodiments, and will not be repeated here for the sake of brevity.

[0093] Example 3 Corresponding to the above embodiments, this application also provides an electronic device.

[0094] See Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 5 As shown, the electronic device 30 may include a processor 301, a memory 302, and a communication unit 303. These components communicate via one or more buses. Those skilled in the art will understand that the electronic device structure shown in the figures does not constitute a limitation on the embodiments of this application. It may be a bus topology or a star topology, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0095] The communication unit 303 is used to establish a communication channel, thereby enabling the electronic device to communicate with other devices.

[0096] The processor 301 serves as the control center of the electronic device, connecting various parts of the device via various interfaces and lines. It executes software programs and / or modules stored in the memory 302, and calls data stored in the memory to perform various functions and / or process data. The processor can be composed of integrated circuits (ICs), such as a single packaged IC or multiple packaged ICs with the same or different functions connected together. For example, the processor 301 may consist only of a central processing unit (CPU). In this embodiment, the CPU may have a single processing core or include multiple processing cores.

[0097] Memory 302 is used to store the execution instructions of processor 301. Memory 302 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0098] When the execution instructions in memory 302 are executed by processor 301, the electronic device 30 is able to perform some or all of the steps in the above method embodiments.

[0099] Example 4 Corresponding to the above embodiments, this application also provides a computer-readable storage medium, wherein the computer-readable storage medium may store a program, wherein when the program runs, it can control the device where the computer-readable storage medium is located to execute some or all of the steps in the above method embodiments. In specific implementation, the computer-readable storage medium may be a magnetic disk, an optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0100] Example 5 Corresponding to the above embodiments, this application also provides a computer program product containing executable instructions that, when executed on a computer, cause the computer to perform some or all of the steps in the above method embodiments.

[0101] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0102] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0103] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0104] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0105] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A method for real-time fusion processing of multi-source sensing data for smart roads under edge-cloud collaboration, characterized in that, include: Construct a semantic processing mechanism for heterogeneous sensing data on the edge side for road traffic objects, and convert the raw sensing data generated by different sensing devices into data units with unified time reference, spatial reference and road traffic semantics; Establish a multi-source real-time fusion mechanism based on road spatial association, data source credibility evolution, and traffic movement feasibility at edge nodes to perform object association, dynamic credibility evaluation, and conflict resolution on traffic semantic data from different edge nodes; Establish a global state fusion and anomaly correction mechanism for cross-road segment traffic propagation relationships in the cloud. Generate continuous traffic state at the road network level based on the local fusion results formed by multiple edge nodes, and use subsequent cross-road segment traffic states to reverse verify the previous fusion results. Establish a closed-loop collaborative mechanism that drives the dynamic adjustment of edge and terminal perception strategies based on the global correction results in the cloud, so that the data processing status of the terminal, edge, and cloud evolves in real time with the road traffic status.

2. The method according to claim 1, characterized in that, The aforementioned mechanism for constructing a semantic processing mechanism for heterogeneous sensing data on the edge side, oriented towards road traffic objects, converts raw sensing data generated by different sensing devices into data units with unified time reference, spatial reference, and road traffic semantics, including: Select the same vehicle that appears consecutively in the sensing areas of adjacent sensing devices as the cross-device timing reference object; The relative time offset between sensing devices is determined based on the appearance time, departure time, traffic propagation time, and motion state continuity of the same reference object in different sensing devices, and the device timestamps are dynamically corrected. The time-corrected perception results are projected onto a pre-established road topology model to form traffic semantic data units for the detected road traffic objects; The degree of traffic status change is determined by the changes in the position, speed, lane status, and event tags of the same traffic object. When the traffic status is stable, only key status increment information is uploaded. When an abnormal status is detected, the upload frequency is increased and historical traffic status cached within a preset time window before the abnormality occurs is also uploaded.

3. The method according to claim 2, characterized in that, The process of projecting the time-corrected perception results onto a pre-established road topology model to form traffic semantic data units for the detected road traffic objects includes: The time-corrected perception results are projected onto a pre-established road topology model; The edge nodes do not continuously upload all the original sensing data directly to the edge nodes, but instead form traffic semantic data units for the detected road traffic objects. When an anomaly is detected, a traffic state change label is added to the corresponding traffic semantic data unit.

4. The method according to claim 1, characterized in that, The establishment of a multi-source real-time fusion mechanism at edge nodes, based on road spatial association, data source credibility evolution, and traffic movement feasibility, performs object association, dynamic credibility evaluation, and conflict resolution on traffic semantic data from different edge nodes, including: Construct a local spatiotemporal association network with traffic objects as nodes, and associate multiple traffic semantic data as the same traffic object only when the state transition between multiple traffic semantic data satisfies the actual road traffic topology; A nonlinear dynamic evolution mechanism for data source credibility is established based on perception deviation, environmental degradation, and historical self-consistency. Based on the feasibility of road traffic movement and the competition relationship between candidate states, conflict data is reverse-verified to select the result with the most feasibility of traffic movement from multiple conflict results.

5. The method according to claim 4, characterized in that, The nonlinear dynamic evolution mechanism for establishing data source credibility based on perception deviation, environmental degradation, and historical self-consistency includes: Acquire the position, speed, and multi-dimensional evaluation parameters between the current output state of the traffic object from the sensing device and the preceding fused trajectory of the corresponding traffic object; Based on the multi-dimensional evaluation parameters, construct a perceptual loss reliability variable. ; Furthermore, reliable state decay orders are constructed based on the average perceived deviation level and average communication delay level within the historical time window. and trusted recovery order ; The data source credibility of the previous fusion cycle is recursively used to determine the dynamic data source credibility of the current fusion cycle. in: .

6. The method according to claim 4 or 5, characterized in that, The method involves reverse verification of conflict data based on the feasibility of road traffic movement and the competition relationship between candidate states, thereby selecting the result with the highest feasibility of traffic movement from multiple conflict results, including: Multiple data source candidate states that are identified as the same traffic object are combined with historical fusion results to calculate traffic realizability deformation variables, and multiplicative coupling verification is performed from longitudinal movement, lane migration and road traffic constraints. The conflict competition index is calculated by combining the dynamic credibility of the data source and the state differences between candidate states, and the optimal value is selected as the main fusion state of the object. When the traffic realizability variables corresponding to all candidate states exceed the preset abnormal range, or the difference between the minimum conflict competition index and the second smallest conflict competition index is insufficient to form a stable judgment, the traffic object is marked as a conflict traffic object and uploaded to the cloud for cross-regional verification.

7. The method according to claim 1, characterized in that, The aforementioned global state fusion and anomaly correction mechanism for cross-road segment traffic propagation relationships established in the cloud generates continuous traffic states at the road network level based on the local fusion results formed by multiple edge nodes, and uses subsequent cross-road segment traffic states to reverse-verify the preceding fusion results, including: The traffic status propagation chain is constructed for adjacent road sections according to road connection relationship, traffic flow direction and intersection connection relationship, and cross-regional trajectory connection is carried out for the same vehicle entering the management area of ​​another edge node. Causal consistency is determined based on the traffic conditions in different edge areas, and the local fusion state is verified and corrected in reverse based on the traffic propagation results of vehicles and traffic events actually observed in subsequent road segments. The corrected global fusion results are combined to form traffic status information at the vehicle object layer, lane status layer, road segment status layer, and road network layer, and further form traffic control demand identifiers.

8. The method according to claim 7, characterized in that, The process of determining causal consistency based on traffic conditions in different edge areas, and performing reverse verification and correction of the local fusion state based on the actual traffic propagation results observed in subsequent road segments of vehicles and traffic events, includes: When an edge node determines that a specific road segment is severely congested, but the upstream vehicle status, downstream outbound traffic flow, and adjacent road traffic status of that road segment do not show any changes that match the congestion level, the global consistency of that local congestion status is reduced. When an edge node detects an abnormal parking, and subsequently detects a decrease in vehicle speed behind the vehicle, an increase in the queue behind it, vehicles in adjacent lanes detouring, and a redistribution of lane traffic, the global credibility of the corresponding abnormal parking event is increased. For data marked as conflicting traffic objects, the cloud performs reverse tracking of different candidate states based on their location and time of appearance in subsequent edge areas; If a candidate state can form a road topological continuity relationship with the subsequently observed trajectory, the result is used to reverse the candidate state judgment of the original edge node, and the corresponding correction result is fed back to the original edge node.

9. The method according to claim 1, characterized in that, The aforementioned closed-loop collaborative mechanism, which establishes a cloud-based global correction result to drive dynamic adjustment of the edge and terminal perception strategies, enables the data processing status at the terminal, edge, and cloud levels to evolve in real time with road traffic conditions, including: Periodically compare the local traffic status uploaded by each edge node with the global corrected traffic status formed in the cloud to determine the error type and duration of different types of sensing devices in different road areas, and generate edge node feedback parameters. The frequency of data uploads to the edge, the fusion cycle of the edge side, and the cross-regional verification frequency of the cloud are dynamically adjusted based on the feedback results from the cloud and the current road traffic conditions. After implementing traffic control based on the cloud-integrated traffic status, the actual traffic change trend after the control is executed is collected and compared with the traffic response direction predicted before the control. When there is a discrepancy, the corresponding road area is marked as a suspected deviation area of ​​the fusion result and a re-verification is triggered.

10. A real-time fusion processing system for multi-source sensing data of smart roads under edge-cloud collaboration, characterized in that, include: The edge semantic processing module is used to build an edge heterogeneous sensing data semantic processing mechanism for road traffic objects, and to convert the raw sensing data generated by different sensing devices into data units with unified time reference, spatial reference and road traffic semantics. The edge fusion module is used to establish a multi-source real-time fusion mechanism at edge nodes based on road spatial association, data source credibility evolution, and traffic movement feasibility. It performs object association, dynamic credibility evaluation, and conflict resolution on traffic semantic data from different edge nodes. The cloud-based global fusion module is used to establish a global state fusion and anomaly correction mechanism for cross-road segment traffic propagation relationships in the cloud. It generates continuous traffic states at the road network level based on the local fusion results formed by multiple edge nodes, and uses subsequent cross-road segment traffic states to reverse verify the previous fusion results. The closed-loop collaborative adjustment module is used to establish a closed-loop collaborative mechanism that drives the dynamic adjustment of the perception strategies of the end and edge sides based on the global correction results in the cloud, so that the data processing status of the end, edge and cloud evolves in real time with the road traffic status.

Citation Information

Patent Citations

  • Traffic operation state sensing method and system based on multi-source data fusion

    CN110766936A

  • Traffic data fusion method and system based on multi-source perception

    CN113781767A