Cloud sensing shared information fusion method for vehicle and road cloud cooperative system

Abstract

The invention provides a cloud sensing shared information fusion method for a vehicle and road cloud cooperative system. The method comprises the steps that the vehicle and road cloud cooperative system comprises a cloud, a vehicle end and a road end; the cloud end obtains environment sensing information from other terminals through network communication; a space-time two-dimensional calibration system is constructed, and the space-time two-dimensional calibration system is adopted to calibrate the environmental perception information; dividing a unified global ID for each target by adopting a global target feature matching method, and performing redundancy removal processing on the calibrated environmental perception information by adopting a confidence evaluation comprehensive redundancy removal mechanism; performing pre-estimation compensation on the environment perception information after redundancy removal by adopting a model based on state estimation and kinematics; transmitting the compensated information to a corresponding terminal by adopting a sensing information routing sharing strategy based on the global ID to complete information sharing; according to the method, a resampling optimization strategy and a collaborative prediction correction algorithm are adopted, compared with a traditional linear prediction compensation method, situation information real-time performance and prediction accuracy are remarkably improved, and timeliness and reliability of follow-up decisions are effectively guaranteed.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data information sharing, and particularly relates to a cloud-end perception shared information fusion method of a vehicle-road cloud cooperative system. BACKGROUND

[0002] Vehicle-road cloud integration is a concept involving the deep integration of vehicles, roads, and cloud computing platforms. This system is accelerating the intelligentization process of urban transportation and demonstrating its important role in improving transportation efficiency, reducing traffic accidents, and improving environmental quality. Vehicle-road cloud integration systems are based on vehicle-road coordination and introduce cloud and other support platforms to achieve safer and more reliable autonomous driving through the use of map and weather platform information. It is a key to the industrialization of intelligent connected vehicles and an important solution to smart transportation and smart cities. Vehicle-road cloud integration systems integrate physical and information spaces of people, vehicles, roads, and clouds through new-generation information and communication technologies, based on system collaborative perception, decision-making, and control, to achieve safe, efficient, energy-saving, and comfortable operation of intelligent connected vehicles and transportation systems.

[0003] Currently, the construction of "vehicle-road cloud integration" has gone through the stages of closed test site technology research and development, open road testing and verification, and pilot application exploration. During the technology research and development stage, 17 national test demonstration zones were successfully established, providing a solid foundation for technology reliability verification and open road testing. At the same time, China has built the world's largest and most advanced 4G and 5G cellular mobile communication network. Meanwhile, the five ministries (Ministry of Industry and Information Technology, Ministry of Public Security, Ministry of Natural Resources, Ministry of Housing and Urban-Rural Development, and Ministry of Transport) jointly issued a notice on the application of intelligent connected vehicles "vehicle-road cloud integration" in pilot cities and identified 20 cities (consortium) as pilot cities. The goal of the pilot work is to build a number of city-level application pilot projects with the same architecture, unified standards, business interconnection, and safe and reliable operation, promote the construction of intelligent roadside infrastructure and cloud control basic platform, increase the vehicle terminal assembly rate, and develop intelligent connected vehicle "vehicle-road cloud integration" system architecture design and various scene applications.

[0004] At present, there are still many problems in the actual deployment and application of the vehicle-road-cloud integrated system, which need to be improved. For example, the traditional vehicle still lacks effective sensing ability and C-V2X communication means to share the sensing information with other end sides, and it is also difficult to obtain the sensing information transmitted by other end sides to supplement its own sensing defects. When the cloud shares the sensing information with other end sides such as vehicle end and road side, it is often affected by the performance difference between devices, resulting in data space-time asynchronous problems. At the same time, after the cloud receives repeated sensing of the same target from multiple end sides, there is a lack of effective technical means to process redundant data, resulting in increased data storage cost and data transmission cost on the communication link. Too much redundant data not only makes it difficult for the management department to analyze and judge, but also brings poor user experience to the driver. In addition, especially for dynamic target state data, due to the influence of transmission delay and processing delay, the real-time performance of the data is poor, which leads to a certain deviation in the subsequent data analysis of the cloud, and the motion state of the dynamic target cannot be accurately estimated, so that the management department cannot have strong foresight for traffic accidents.

[0005] The existing sensing information sharing strategy is usually passive, and most of them adopt the publish / subscribe mode. This mode cannot adapt to the dynamic changes of the scene and user demand over time, and cannot share the sensing data associated with specific targets according to the demand of different targets for different sensing information in the actual scene. This will lead to chaos in the cloud data delivery, reduce user experience, increase data transmission cost, reduce traffic driving safety, and is not conducive to the intelligentization and automation level of the overall system. SUMMARY

[0006] In order to solve the problems existing in the prior art, the present application provides a cloud-end sensing information fusion method for a vehicle-road-cloud collaborative system, which comprises: the vehicle-road-cloud collaborative system comprises a cloud end, a vehicle end and a road end; the cloud end obtains environmental sensing information from other terminals through network communication; a time-space two-dimensional calibration system is constructed, and the environmental sensing information is calibrated by using the time-space two-dimensional calibration system; a global target feature matching method is used to divide a unified global ID for each target, and a confidence evaluation comprehensive deduplication mechanism is used to process the calibrated environmental sensing information; a state estimation and kinematics model based compensation is performed on the deduplicated environmental sensing information; the compensated information is transmitted to the corresponding terminal based on the global ID by using a sensing information routing sharing strategy, and the information sharing is completed.

[0007] The beneficial effects of the present application are as follows:

[0008] In view of the time-space asynchronous problem of multi-source heterogeneous terminal equipment sensing information in the vehicle-road cloud collaborative environment, the application constructs a time-space two-dimensional calibration system, that is, an adaptive time delay compensation algorithm is integrated into the PTP protocol, a dynamic mapping relationship between network jitter characteristics and time delay compensation parameters is established, and accurate correction of the sensing data network transmission time delay is realized. At the same time, a UTM projection optimization mechanism based on traffic situation awareness is designed, the projection center point and coordinate conversion parameters are dynamically adjusted based on real-time traffic flow characteristics, a high-precision local coordinate system for regional information distribution is constructed, and accurate local geographic coordinate basis is provided for subsequent information sharing and distribution.

[0009] In view of the high-frequency information collection of vehicle-end and road-end sensors, repeated reporting between different devices, or the fact that the processing algorithm is not suitable in the data integration process, resulting in a large amount of repeated or redundant data in the information sent by the vehicle-end and road-end to the cloud, the application proposes a Bloom Filters de-redundancy method based on a multi-dimensional confidence evaluation mechanism. According to the distance confidence factor, the freshness scoring mechanism, the device accuracy evaluation system and the environmental interference evaluation factor, a multi-dimensional confidence evaluation mechanism is constructed, the evaluation results of the four dimensions are normalized and then weighted and fused, and finally the comprehensive confidence score of each sensing data is output. On this basis, combined with the Bloom Filters filter, through the dynamic bit array adjustment mechanism, the multi-hash function strategy based on the device ID and the confidence threshold adaptive algorithm, an intelligent data de-redundancy system is constructed, which significantly improves the system processing efficiency and storage space utilization rate while ensuring data integrity compared with the traditional fixed threshold de-redundancy method.

[0010] In view of the fact that the dynamic target state information received by the cloud is often lagging behind the actual situation due to the influence of communication delay and processing delay, and when processing the outdated object situation information, the analysis result is often difficult to accurately reflect the real state change of the object, the application designs a situation error compensation method of particle filtering + CA motion model. Through the construction of the particle filtering prediction model, the lagging information is estimated and compensated, and the dynamic parameter adaptive adjustment mechanism of the CA model is combined to realize accurate error correction in the time-varying communication environment. The resampling optimization strategy and collaborative prediction correction algorithm are adopted, which has a significant improvement in real-time and prediction accuracy compared with the traditional linear prediction compensation method, effectively guaranteeing the timeliness and reliability of subsequent decision-making.

[0011] In view of the problem that the global perception is incomplete due to the different perception capabilities and limited communication resources of each participant in the vehicle-road cloud cooperative scene, and the traditional data distribution method has the problems of weak information correlation and poor timeliness, the application designs a dual-mode perception data cooperative distribution mechanism based on Lightgbm intelligent decision and end-side active request. By fusing the efficient classification algorithm of Lightgbm and the active request / response dual-mode distribution strategy, the precision push of passive distribution is realized by constructing a feature importance evaluation model, combined with the active request mechanism of the end-side demand, the gradient boosting decision tree optimization and dynamic weight adjustment technology are used, compared with the traditional fixed rule distribution method, the information integrity guarantee and communication resource utilization are improved significantly, and the cooperative perception ability of the traffic participants is effectively enhanced.

[0012] In view of the different needs of different receiving ends and different business scenarios in the vehicle-road cloud cooperative system for data timeliness and processing level, the application proposes an intelligent hierarchical distribution method based on business feature recognition. The business is divided into real-time business, quasi-real-time business and non-real-time business according to the communication object and timeliness requirement, and a distribution decision mechanism based on business feature driving is constructed. Through the establishment of a three-level cooperative distribution system of "edge cloud-regional cloud-center cloud", according to different data receiving objects, combined with the business priority dynamic scheduling algorithm, the hierarchical optimization processing of different timeliness requirement businesses is realized. Compared with the traditional unified distribution mode, the response speed of high real-time business is significantly improved, and the overall resource utilization efficiency of the system is optimized. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 The overall flowchart of the application;

[0014] Figure 2 The cloud perception data flowchart of the application;

[0015] Figure 3 The cloud perception information fusion processing flowchart of the application;

[0016] Figure 4 The purpose area association sharing distribution flowchart of the application based on;

[0017] Figure 5 The intelligent hierarchical distribution flowchart of the application based on business feature recognition. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0019] The drawings are only used for exemplary illustration, and the representation is a schematic diagram, not a physical diagram, and should not be understood as a limitation on the present application; in order to better illustrate the embodiments of the present application, some components in the drawings are omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings can be omitted.

[0020] The same or similar reference numerals in the drawings of the embodiments of the present application correspond to the same or similar components; in the description of the present application, it should be understood that if the terms "upper", "lower", "left", "right", "front", "back" and the like indicate the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore the terms describing the positional relationship in the drawings are only used for exemplary illustration, and should not be understood as a limitation on the present application, for those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0021] A cloud-end perception shared information fusion method of a vehicle-road cloud cooperative system, as shown in Figures 1-5 The method comprises the following steps: the vehicle-road cloud cooperative system comprises a cloud end, a vehicle end and a road end; the cloud end obtains environmental perception information from other terminals through network communication; a time-space two-dimensional calibration system is constructed, and the time-space two-dimensional calibration system is used for calibrating the environmental perception information; a global target feature matching method is used to divide a unified global ID for each target, and a confidence evaluation comprehensive de-redundancy mechanism is used to de-redundancy process the calibrated environmental perception information; a state estimation and kinematics model based method is used to estimate and compensate the de-redundancy environmental perception information; the compensated information is transmitted to the corresponding terminal based on the global ID using a perception information routing sharing strategy, and information sharing is completed.

[0022] In the present embodiment, calibrating the environmental perception information using the time-space two-dimensional calibration system comprises:

[0023] In the time calibration dimension, a dynamic mapping relationship between network jitter characteristics and time delay compensation parameters is established by incorporating an adaptive time delay compensation algorithm into the PTP protocol, so as to realize accurate correction of the network transmission time delay of the perception data;

[0024] In the spatial calibration dimension, the UTM projection optimization mechanism based on traffic situation perception is adopted to dynamically adjust the projection center point and coordinate conversion parameters; a high-precision local geographic reference framework for a specific area is constructed, and the GPS longitude and latitude information uploaded by the vehicle is mapped into the optimized local coordinate system in real time.

[0025] The adaptive delay compensation algorithm is integrated into the PTP protocol to establish a dynamic mapping relationship between network jitter characteristics and delay compensation parameters, including: continuously monitoring and extracting the dynamic characteristics of network jitter, including delay variance and jitter frequency; using intelligent algorithms to establish a dynamic mapping relationship between dynamic characteristics and optimal compensation parameters; by real-time querying of the dynamic mapping relationship and dynamically adjusting the clock offset of PTP, precise and adaptive compensation of random network jitter is realized.

[0026] The intelligent algorithm is used to establish a dynamic mapping relationship between dynamic characteristics and optimal compensation parameters, including:

[0027] Based on real-time traffic flow characteristics, the projection center point and coordinate conversion parameters are dynamically adjusted, and a high-precision local geographic reference framework for information distribution areas is constructed. First, the traffic flow core characteristics, i.e. vehicle density, are calculated through real-time vehicle data:

[0028]

[0029] The vehicle density is converted into weight. In the information distribution area, the area is divided into multiple small grids, and the vehicle density of each grid is counted. The higher the density, the greater the corresponding weight. The dynamic center point is calculated by weighted average, i.e. the denser the traffic, the stronger the influence on the center point position, and the center point will automatically deviate to these areas.

[0030] Then the UTM projection optimization mechanism based on traffic situation perception is adopted to calculate the longitude difference

[0031]

[0032] wherein is the vehicle GPS longitude, is the central meridian longitude dynamically adjusted with the traffic flow. The converted plane coordinates are calculated by the formula:

[0033]

[0034]

[0035] wherein B is the vehicle GPS latitude, For the projection origin coordinates, a and b are fixed coefficients. The GPS longitude and latitude information uploaded by the vehicle is mapped in real time into the optimized local coordinate system through the UTM projection mechanism, providing accurate local geographic coordinate basis for subsequent information sharing and distribution.

[0036] A global target feature matching method is used to divide a unified global ID for each target, including: calculating the similarity of two targets according to the sensing parameters of the targets in different sensors; setting a threshold, if the similarity is within the set threshold, a unified global ID in the cloud is divided for it, otherwise, no global ID is divided; wherein the sensing parameters include longitude and latitude information, speed and direction angle information.

[0037] In this embodiment, a confidence evaluation comprehensive de-redundancy mechanism is used to process the calibrated environmental perception information, including: establishing a comprehensive confidence evaluation mechanism according to the distance relationship between the sensing end side and the sensing target, the time freshness of the sensing data, the device accuracy evaluation system and the environmental interference evaluation factor, normalizing the results and then performing weighted fusion to output the comprehensive confidence score of each sensing data.

[0038] In this embodiment, a cloud-end perception sharing information fusion method of a vehicle-road-cloud collaborative system specifically includes:

[0039] S1, a multi-modal communication interface is deployed in the cloud, including 4G / 5G wireless network, wired network and special communication link, supporting flexible adaptation of multiple communication protocols such as UDP and MQTT. Through the heterogeneous network communication capability, seamless connection with the vehicle end, roadside sensing device (RSU), traffic management department and third-party enterprise system is realized, ensuring full-dimensional perception information collection and aggregation of actual traffic scenes.

[0040] S2, the original timestamp of the sensing data and the network transmission jitter characteristics are collected in real time, the adaptive time delay compensation algorithm is integrated into the PTP protocol, the dynamic mapping relationship between the network jitter characteristics and the time delay compensation parameters is established, and the accurate correction of the sensing data transmission time delay is realized. At the same time, a UTM projection optimization mechanism based on traffic situation perception is designed, the projection center point and the coordinate conversion parameters are dynamically adjusted based on real-time traffic flow characteristics, a high-precision local coordinate system for regional information distribution is constructed, and the GPS longitude and latitude information uploaded by the vehicle is mapped in real time into the optimized local coordinate system, providing accurate local geographic coordinate basis for subsequent information sharing and distribution.

[0041] S3, a confidence evaluation model based on rules is constructed according to the distance confidence factor, the time scoring mechanism, the device accuracy evaluation system and the environmental interference evaluation factor, the evaluation results of the four dimensions are normalized and then weighted fused, and a reliability score in the range of 0-1 is attached to the data.

[0042] For data at different confidence levels, an adaptive Bloom filter is used for differentiated processing: 1) A dynamic bit array adjustment strategy is designed to optimize storage space allocation in real time according to data flow, and automatically expand the bit array size when the data density exceeds a preset threshold; 2) A multi-hash function optimization mechanism is adopted to achieve rapid initial screening of redundant data through the main hash function group, supplemented by a secondary verification hash function to reduce the false judgment rate, and finally obtain the unique data with the highest confidence and the strongest real-time performance of the same type at the same time.

[0043] S4. Construct an error compensation model that integrates improved particle filtering and adaptive CA motion model. A prediction model containing 200-500 particles is constructed, each carrying third-order state variables (position, velocity, and acceleration). Lag information is probabilistically estimated through importance sampling. A dynamic adjustment mechanism for CA model parameters is designed, including: 1) an adaptive acceleration correction algorithm based on vehicle motion characteristics; 2) an online update strategy for the time-varying noise covariance matrix; and 3) motion state abrupt change detection and model switching logic. Simultaneously, a system resampling optimization strategy is employed to prevent particle degradation, and Kalman weighted fusion is performed between the latest received real observation data and the prediction results to form a closed-loop correction.

[0044] S5. Design a dual-mode perception data collaborative distribution mechanism based on LightGBM intelligent decision-making and edge-side active request / response. First, construct a feature importance evaluation model based on LightGBM. Then, use a gradient boosting decision tree algorithm to perform multi-dimensional feature analysis on the multi-source perception data aggregated in the cloud (including data timeliness, spatial coverage, target type, motion status, etc.) to achieve intelligent classification based on the destination region of interest during passive distribution. Simultaneously, design an edge-side dynamic request / response mechanism to support vehicles, roadside devices, pedestrians, traffic management departments, and third-party enterprises to dynamically request personalized data from the cloud based on their own perception needs and intentions. Employ dynamic weight adjustment technology to optimize the resource allocation ratio between passive push and active response in real time: 1) an adaptive bandwidth allocation algorithm based on network load status; 2) an intelligent caching strategy considering data correlation; and 3) a priority interruption mechanism for emergency events.

[0045] S6. Classify services into real-time, near-real-time, and non-real-time services based on communication objects and timeliness requirements, and construct a distribution decision mechanism driven by service characteristics. By establishing a three-level collaborative distribution system of "edge cloud - regional cloud - central cloud", and combining different data receiving objects with a dynamic scheduling algorithm based on service priority, hierarchical optimization processing of services with different timeliness requirements can be achieved.

[0046] In this embodiment, the cloud-based perception and sharing information fusion method, distribution strategy, and system in step S1 consist of a vehicle-side terminal, an intelligent roadside system, and a cloud service platform. Figure 2The cloud-based sensing data flow diagram shown includes:

[0047] The vehicle-side components include intelligent connected vehicles with sensing capabilities and non-intelligent connected vehicles combined with mobile intelligent terminals. They are primarily responsible for sensing their own and other vehicles' status information, uploading and receiving sensing information, real-time route planning, and assisting with safe driving.

[0048] The intelligent roadside system mainly refers to cameras and radars with sensing information detection capabilities, as well as roadside MECs with corresponding computing and communication capabilities. It is primarily responsible for environmental and object perception and recognition, scene-specific sensing information collection, traffic incident monitoring, and the uploading and receiving of sensing information.

[0049] The cloud service platform mainly refers to a cloud service platform that provides road environment perception information and traffic object perception information for vehicles, roadside devices, and management departments, including but not limited to central cloud, regional cloud, edge cloud, and enterprise cloud. It is primarily used for receiving multi-source perception information, spatiotemporal synchronization of multi-source perception information, redundancy processing of multi-source perception information, error compensation of multi-source perception information, storage of multi-source perception information, priority scheduling of perception information, and formulation of perception information sharing strategies.

[0050] The cloud service platform flexibly adapts to the diverse communication needs of vehicle-side, roadside, management, and enterprise-side terminals by deploying various communication interfaces such as wired networks, wireless networks, and dedicated lines. It efficiently integrates multi-source perception data to complete the acquisition of comprehensive perception information in actual traffic scenarios in the cloud.

[0051] Intelligent connected vehicles can directly upload perceived information to the cloud in real time via 4G / 5G networks, while non-intelligent connected vehicles upload data via 4G / 5G networks using mobile intelligent terminals (such as smartphones or in-vehicle OBD devices), ensuring that all types of vehicles can access the cloud system. Secondly, roadside cameras and radar, among other sensing devices, transmit the raw data to local roadside MEC nodes for preliminary processing. The processed data is then uploaded to the cloud via highly reliable wired networks (such as fiber optics or Ethernet) or low-latency 4G / 5G wireless networks to meet the transmission needs of different scenarios. Finally, traffic management departments and enterprises upload traffic management data and operational data to the cloud via dedicated line connections (such as VPNs or fiber optic leased lines), ensuring the security and stability of data transmission while supporting the efficient transmission of large-scale data.

[0052] In this embodiment, spatiotemporal asynchronous processing of multi-source sensing information is performed in step S2. By incorporating an adaptive delay compensation algorithm into the PTP protocol, a dynamic mapping relationship between network jitter characteristics and delay compensation parameters is established, thereby achieving accurate correction of the transmission delay of sensing data network.

[0053] For the calculation of clock offset, firstly, let the master clock timestamps be t1 (Sync transmission) and t2 (Delay_Req reception), and the slave clock timestamps be t1' (Sync reception) and t2' (Delay_Req transmission). Then the clock offset is calculated as follows. The calculation formula is:

[0054]

[0055] The formula for calculating transmission delay (Delay) is:

[0056] Delay = [(t1' - t1) + (t2 - t2')] / 2

[0057] Periodic synchronization correction is performed every Δt cycle to dynamically adjust the clock deviation. The correction amount Δ is given by where (α and β are the PID control coefficients):

[0058] Δ = α·Offset + β·d(Offset) / dt

[0059] The delay compensation amount is obtained by combining the dynamic mapping relationship between network jitter characteristics and delay compensation parameters:

[0060]

[0061] Where σ is the standard deviation of network delay, J_max is the maximum jitter value within the observation window, J_avg is the average network jitter, and α, β, γ are dynamic weight coefficients.

[0062] Then, spatial synchronization of the perceived information is performed. A coordinate transformation model is designed to map the GPS latitude and longitude information uploaded by the vehicle to the optimized local coordinate system in real time. A spatial coordinate system alignment model based on UTM spatial projection is designed to synchronize and align the spatial coordinates of the data. Given WGS-84 coordinates (λ, φ), the calculation process of the zone number, the transformed x-coordinate, and the transformed y-coordinate in the UTM projection is as follows (where k0 = 0.9996, T = tan²φ, C = e'²cos²φ, A = (λ - λ0)cosφ):

[0063]

[0064]

[0065] Northing = k0·[S + tanφ·(A² / 2 + (5-T+9C)A 4 / twenty four)]

[0066] In this embodiment, a feature matching method for reference frame normalization of multiple targets in different reference frames is performed in step S3, such as...Figure 3 The flowchart shown is for cloud-based sensing information fusion processing. It includes the following processes:

[0067] First, information on their category, position, velocity, and orientation angle is used as the matching criterion. Then, sensing objects are extracted from collaborative sensing information from other sources. First, the category is used to determine if they belong to the same class. Then, their feature vectors are extracted. To avoid interference from different units of measurement in the calculation results, Min_Max normalization is chosen to map their feature vectors to the range [0,1]. Then, their Euclidean distance is calculated using the following formula:

[0068]

[0069] Where A and B are the feature vectors of the target. , where represents position, velocity, and direction angle, respectively.

[0070] The similarity between two targets is calculated using the following formula:

[0071]

[0072] If the similarity exceeds a set threshold, they are determined to be the same target; otherwise, they are determined to be different objects. Once identified as the same object, it is added to a linked list for use in the next step.

[0073] A multi-dimensional confidence evaluation mechanism is constructed based on four dimensions: distance confidence factor, freshness scoring mechanism, equipment accuracy assessment system, and environmental interference assessment factor. The evaluation results of these four dimensions are normalized and then weighted and fused to output a comprehensive confidence score for each sensed data point. Building upon this, an intelligent data redundancy removal system is constructed by combining it with Bloom filters, employing a dynamic bit array adjustment mechanism, a device ID-based multi-hash function strategy, and a confidence threshold adaptive algorithm.

[0074] First, the reliability of data uploaded from other endpoints is quantitatively evaluated: Let the i-th sensed data be D_i, and its confidence score be:

[0075]

[0076] Where S_i ∈ [0,1] is the sensor quality score, T_i ∈ [0,1] is the spatiotemporal freshness score, and H_i ∈ [0,1] is the detection distance score. α+β+γ=1 represents the weighting coefficients.

[0077] To accelerate the filtering of redundant data, a Bloom filter mechanism is introduced. Specifically, the system maintains a dynamic Bloom filter with parameters (m, k) for each target ID, where m is the size of the bit array and k is the number of hash functions. The message's confidence level is used as a key feature and mapped to the bit array through multiple hash functions. When a new message arrives, M... i = (ID i C i ,T i The system first checks the Bloom filter corresponding to the target ID and calculates k hash positions:

[0078] If all H positions are set and the following conditions are met:

[0079]

[0080] If the message is deemed redundant, it is considered redundant; otherwise, the filter is updated and the fusion process begins. This method achieves collaborative filtering of confidence and timeliness through quantization parameters (ϵ,δ) while maintaining O(k) time complexity.

[0081] The initial size of the bit array is m_0, and the dynamic adjustment strategy is as follows:

[0082]

[0083] in Let t be the estimated data volume at time t, and p be the target misclassification rate.

[0084] Multiple hashing optimization is performed using a primary-secondary dual hash group, where the expressions for the primary and secondary hash groups are as follows:

[0085]

[0086]

[0087] Calculate data similarity using a similarity function:

[0088]

[0089] in τ is the attenuation coefficient.

[0090] In this embodiment, the method of predicting and compensating for the deredundant perception information based on state estimation and kinematic model includes: constructing a particle filter prediction model; using the particle filter prediction model to estimate and compensate for the state of the lag information; using the dynamic parameter adaptive adjustment mechanism of the particle filter prediction model to correct the error in the time-varying communication environment; and using a resampling optimization strategy and a collaborative prediction correction algorithm to reasonably estimate the state of the dynamic target after a time delay of t, thereby obtaining the compensated situational error.

[0091] In this embodiment, step S4 involves constructing an error compensation model that integrates the improved particle filter and the adaptive CA motion model, including:

[0092] By combining the dynamic parameter adaptive adjustment mechanism of the CA model, accurate error correction in time-varying communication environments is achieved. A resampling optimization strategy and a collaborative prediction correction algorithm are employed to reasonably estimate the state of a dynamic target after a time delay of t, thereby compensating for situational errors and improving data timeliness and accuracy.

[0093] First, the target state vector is defined:

[0094]

[0095] in

[0096] The adaptive state equation of the improved CA motion model is:

[0097] Where the state transition matrix is:

[0098]

[0099] For the implementation of particle filtering, the particle set can be represented as:

[0100]

[0101] Considering the noise and delay in sensor measurements, Kalman filtering or extended Kalman filtering can be introduced to optimize the prediction results. By dynamically adjusting the error between the predicted value and the actual measured value, the situational error can be compensated.

[0102] The Kalman weighted fusion formula is:

[0103]

[0104] The Kalman gain is:

[0105]

[0106] By combining the target's historical trajectory with environmental context information (such as road constraints and target behavior patterns), the prediction results are further revised, thereby improving the timeliness and accuracy of the data and ensuring that the system can more reliably reflect the target's true state.

[0107] In step S5 of this embodiment, a dual-mode sensing data collaborative distribution mechanism based on LightGBM intelligent decision-making and active requests from the edge is designed, such as... Figure 4 As shown, the process includes:

[0108] LightGBM's intelligent decision-making passive distribution first divides destination areas based on static map data and the data requirements of management departments and enterprises. Traffic participants within the same area are associated with the same data-sharing object. Simultaneously, geographic features are extracted from the collected multi-device data, projected using a dynamic UTM coordinate system, and the area is divided according to different precision levels to facilitate subsequent area matching and classification. Furthermore, multi-dimensional geographic feature vectors are constructed by combining static features such as regional POI (Point of Interest) density and road type distribution to enhance the accuracy of area identification. These features are input into the data classification model for training, constructing an optimal classification hyperplane to achieve data classification, dynamically adapting to data changes and improving model accuracy. To improve model performance, an imbalanced dataset processing strategy is adopted: weight adjustments or focal loss are applied to areas with small sample sizes to ensure accurate identification of small-area data. Bayesian hyperparameter optimization automatically adjusts key parameters of LightGBM such as tree depth and learning rate to construct the optimal classification decision boundary. Finally, SHAP value analysis visualizes the contribution of different geographic features to the classification results, providing interpretable evidence for management decisions.

[0109] The edge-side proactive request / response mechanism achieves intelligent data distribution through the following process: When the vehicle's OBU or roadside RSU detects missing or insufficiently confident local sensing data, it first generates a data demand level based on a lightweight demand assessment model (inputs include current vehicle speed, road type, surrounding vehicle density, etc.). Then, it initiates a tiered request (three levels: regular / urgent / safety critical) to the edge cloud via a low-latency communication interface. After receiving the request, the cloud dynamically allocates transmission resources based on network load status and service priority, compresses the target data packet using differential coding technology, and sends it to the requesting terminal in real time through the optimal link.

[0110] In step S6 of this embodiment, a multi-level cloud collaborative distribution mechanism based on service objects is designed, such as... Figure 5 As shown, the process includes:

[0111] In each cloud tier, a perception information classification algorithm and intelligent routing mechanism are deployed to dynamically determine forwarding paths based on the content of the perception information and the relevance of the communication objects. Building upon this, services are categorized into real-time, near-real-time, and non-real-time services according to the communication objects and timeliness requirements, constructing a service characteristic-driven distribution decision mechanism. By establishing a three-tiered collaborative distribution system of "edge cloud - regional cloud - central cloud," and combining different data receiving objects with a dynamic scheduling algorithm based on service priorities, hierarchical optimization processing of services with different timeliness requirements is achieved. Compared to the traditional unified distribution model, this method significantly improves the response speed of high real-time services while optimizing the overall system resource utilization efficiency.

[0112] It is worth noting that each implementation detail of the cloud-based perception information fusion and perception data sharing method and system based on the vehicle-road-cloud system described in the embodiments of the present invention can be referred to the working process of the cloud-based perception information fusion and perception data sharing method and system based on the vehicle-road-cloud system described in the above embodiments, and will not be repeated here.

[0113] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium. When the program is executed, it can implement the steps of the method. The storage medium may be a magnetic disk, an optical disk, etc.

[0114] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A cloud-based perception sharing information fusion method for a vehicle-road cloud cooperation system, characterized in that, The application relates to a vehicle-road-cloud cooperative system and a calibration method thereof. The vehicle-road-cloud cooperative system comprises a cloud terminal, a vehicle terminal and a road terminal; the cloud terminal obtains environment sensing information from other terminals through network communication; A time-space dual-dimension calibration system is constructed to calibrate the environment sensing information by using the time-space dual-dimension calibration system; A global target feature matching method is used to divide a unified global ID for each target, and a confidence evaluation comprehensive deduplication mechanism is used to deduplicate the calibrated environment sensing information. A state estimation and kinematics model-based method is used to estimate and compensate the deduplicated environment sensing information; a sensing information routing sharing strategy is used to transmit the compensated information to the corresponding terminal based on the global ID, so that information sharing is completed. 2.The cloud perception information fusion method of a vehicle-road-cloud cooperative system according to claim 1, characterized in that, The environment sensing information comprises real-time situation information of traffic participants, road environment information and traffic event information in a scene. 3.The cloud perception sharing information fusion method of a vehicle-road-cloud cooperative system according to claim 1, characterized in that, The calibration of the environment sensing information by using the time-space dual-dimension calibration system comprises the following steps: In the time calibration dimension, a dynamic mapping relationship between network jitter characteristics and time delay compensation parameters is established by integrating an adaptive time delay compensation algorithm into the PTP protocol, so that the network transmission time delay of the sensing data is accurately corrected. In the space calibration dimension, a UTM projection optimization mechanism based on traffic situation awareness is used to dynamically adjust the projection center point and the coordinate conversion parameters; a high-precision local geographic reference framework for a specific area is constructed, and the GPS longitude and latitude information uploaded by the vehicle is mapped into the optimized local coordinate system in real time. 4.The cloud perception sharing information fusion method of a vehicle-road-cloud cooperative system according to claim 3, characterized in that, The dynamic mapping relationship between the network jitter characteristics and the time delay compensation parameters is established by integrating the adaptive time delay compensation algorithm into the PTP protocol, which comprises the following steps: the dynamic characteristics of the network jitter are continuously monitored and extracted, the dynamic characteristics comprise the time delay variance and the jitter frequency; an intelligent algorithm is used to establish a dynamic mapping relationship between the dynamic characteristics and the optimal compensation parameters; the clock offset of the PTP is dynamically adjusted by real-time querying of the dynamic mapping relationship, so that the random network jitter is accurately and adaptively compensated. 5.The cloud perception sharing information fusion method of a vehicle-road-cloud cooperative system according to claim 3, characterized in that, The UTM projection optimization mechanism based on traffic situation awareness is used to dynamically adjust the projection center point and the coordinate conversion parameters, which comprises the following steps: the projection center point and the coordinate conversion parameters are dynamically adjusted based on the real-time traffic flow characteristics, a high-precision local geographic reference framework for a specific area is constructed, and the GPS longitude and latitude information uploaded by the vehicle is mapped into the optimized local coordinate system in real time through the UTM projection mechanism. 6.The cloud perception sharing information fusion method of a vehicle-road-cloud cooperative system according to claim 1, characterized in that, The global target feature matching method is used to divide a unified global ID for each target, which comprises the following steps: The similarity of two targets is calculated according to the sensing parameters of the targets in different sensors; a threshold is set, if the similarity is within the set threshold, a global ID of the cloud terminal is divided for the target, otherwise, no global ID is divided; wherein the sensing parameters comprise longitude and latitude information, speed and direction angle information. 7.The cloud perception sharing information fusion method of a vehicle-road-cloud cooperative system according to claim 1, characterized in that, The confidence evaluation comprehensive deduplication mechanism is used to deduplicate the calibrated environment sensing information, which comprises the following steps: a comprehensive confidence evaluation mechanism is established according to the distance relationship between the sensing terminal side and the sensing target, the time freshness of the sensing data, the device accuracy evaluation system and the environmental interference evaluation factor; the results are normalized and then weighted and fused to output the comprehensive confidence score of each sensing data. 8.The cloud perception sharing information fusion method of a vehicle-road-cloud cooperative system according to claim 1, wherein, The pre-estimation and compensation of the de-redundant perception information based on state estimation and kinematic model includes: constructing a particle filter prediction model, using the particle filter prediction model to estimate and compensate the lag information; using the dynamic parameter adaptive adjustment mechanism of the particle filter prediction model to correct the error under the time-varying communication environment; using the resampling optimization strategy and the cooperative prediction correction algorithm to reasonably estimate the state of the dynamic target after a delay of t time, and obtain the compensation situation error. 9.The cloud perception sharing information fusion method of a vehicle-road-cloud cooperative system according to claim 1, wherein, The perception information routing sharing strategy includes a machine learning passive sharing strategy and a dual-mode data distribution mechanism of end-side active request / response; The machine learning passive sharing strategy includes: dividing the static map data in the scene into destination areas, associating the traffic participants in the same area in a same data sharing object; training a machine learning model in combination with the regional POI density and road type distribution; and realizing the classification and distribution of data by constructing an optimal classification hyperplane; The dual-mode data distribution mechanism of end-side active request / response includes: real-time service distribution, quasi-real-time service distribution and non-real-time service distribution; by establishing a "edge cloud-regional cloud-central cloud" three-level cooperative distribution system, according to different data receiving objects, in combination with a business priority dynamic scheduling algorithm, hierarchical optimization processing of services with different time efficiencies is realized.

Images