Electronic map update method and driving warning method based on vehicle-road-cloud cooperative perception

Abstract

This invention relates to an electronic map update method and driving warning method based on vehicle-road-cloud collaborative perception, within the field of intelligent transportation. It effectively alleviates the problem of inaccurate intelligent driving warnings caused by the limitations of existing single-vehicle intelligent perception systems in accurately acquiring road physical conditions due to the limitations of perception dimensions and range. The electronic map update method collects vehicle data, environmental data, and meteorological data of target vehicles within a target area; extracts event feature information from all acceleration data using the E-VESA event feature extraction method; obtains road state information based on the event feature information, environmental data, and meteorological data using the ST-TFM method; and overlays the road state information onto the geographic information layer of the target area in real time, achieving real-time updates of the electronic map within the target area. This invention is applicable to the fields of beyond-line-of-sight, quantitative assessment and active safety warning of road physical conditions.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation, and more specifically to the field of electronic maps. Background Technology

[0002] With socio-economic development and accelerated urbanization, modern transportation systems are increasingly demanding higher levels of safety, efficiency, and intelligence. In particular, the rapid evolution of Advanced Driver Assistance Systems (ADAS) and Autonomous Driving (AD) technologies has made real-time and accurate perception of the road environment a core technological challenge that urgently needs to be addressed. However, current technological solutions still have many limitations when dealing with complex and ever-changing road conditions.

[0003] First, the mainstream "single-vehicle intelligence" technology path has inherent perception bottlenecks. While sensors such as cameras, millimeter-wave radar, and lidar on vehicles can effectively identify obstacles and lane markings around the vehicle, their perception capabilities have three major limitations: First, the perception range is limited, typically confined to the line-of-sight range of a few hundred meters, unable to detect sudden situations beyond visual range or in obscured areas; second, they are susceptible to environmental interference. In adverse weather and lighting conditions such as rain, snow, fog, nighttime, and strong sunlight, sensor performance significantly degrades, leading to unreliable perception results; third, the perception dimension is singular. Onboard sensors primarily acquire geometric and kinematic information, unable to directly measure key physical parameters of the road surface, such as the road surface friction coefficient, water film thickness, road surface temperature, the presence of thin ice (i.e., black ice) that is difficult to distinguish with the naked eye, and the degree of road surface damage and bumps. These physical parameters directly relate to the limits of tire adhesion and dynamic control, and are crucial for the most basic control decisions such as acceleration, braking, and steering. Without this crucial information, intelligent vehicles are like "blind men touching an elephant" when faced with slippery, icy, or damaged road surfaces, unable to adjust their control strategies (such as Automatic Emergency Braking (AEB) and Adaptive Cruise Control (ACC) in advance), thus posing serious safety hazards.

[0004] Secondly, existing road condition monitoring methods based on traffic infrastructure are insufficient to meet the needs of intelligent driving. Currently, road condition monitoring mainly relies on the following methods: First, deploying fixed meteorological monitoring stations or embedded road sensors at key nodes such as highways or bridges. However, such equipment is sparsely deployed, costly, and can only provide discrete, point-like monitoring data, failing to form a continuous, wide-area road network status map. Second, relying on municipal maintenance departments for manual or vehicle inspections. This method is slow to respond and inefficient, with delays of several hours or even days in detecting and reporting road potholes, water accumulation, and other conditions, failing to meet real-time requirements. Third, while existing navigation software can provide macroscopic traffic congestion information, it usually ignores the physical causes of congestion or risk, failing to provide drivers with precise and quantifiable safety warnings such as "the road ahead is slippery due to ice 500 meters ahead."

[0005] In summary, existing electronic map updates suffer from the problem that existing single-vehicle intelligent perception systems are limited in their perception dimensions and range, making it impossible to accurately obtain the physical state of the road, which in turn leads to inaccurate warnings for intelligent driving. Summary of the Invention

[0006] This invention alleviates the problem of inaccurate intelligent driving warnings caused by the limitations of existing single-vehicle intelligent perception systems in accurately acquiring road physical conditions due to the limitations of perception dimensions and range in current electronic map updates. This invention provides the following solution: Option 1: An electronic map update method based on vehicle-road-cloud collaborative perception, including the following steps: Step S1: Collect vehicle data and environmental data of the target vehicle within the target area; the vehicle data includes acceleration data; Step S2: Collect meteorological data for the target area; Step S3: Obtain event feature information from the acceleration data using the E-VESA event feature extraction method; Step S4: Based on the event feature information, the environmental data, and the meteorological data, obtain road condition information using the ST-TFM method; Step S5: The road status information is overlaid onto the geographic information layer of the target area in real time to achieve real-time updates of the electronic map within the target area.

[0007] Furthermore, in one embodiment of the present invention, the E-VESA event feature extraction method in step S3 includes the following steps: Step S31: Remove noise from all the acceleration data to obtain denoised acceleration data; Step S32: Perform continuous wavelet transform (CWT) on the acceleration data to obtain the time-frequency matrix; Step S33: Obtain the center distance of the time-frequency matrix using the two-dimensional moment invariant method; Step S34: Obtain morphological features based on the center distance; Step S35: The morphological features are processed through a neural network classifier to obtain the category probability. If the category probability is greater than the confidence threshold, structured road event feature information is generated. The road event feature information includes the road event type, occurrence probability, severity level, precise timestamp, GPS coordinates, and associated raw CAN data.

[0008] Furthermore, in one embodiment of the present invention, the continuous wavelet transform (CWT) in step S32 is performed through:

[0009] Obtain the time-frequency matrix ,in, For scale parameters; For position parameters; In the vertical direction The first moment The acceleration data mentioned in the paragraph; It is the conjugate of the mother wavelet function.

[0010] Furthermore, in one embodiment of the present invention, the two-dimensional moment invariant method in step S33 is achieved by:

[0011] Obtain center distance ,in, for The order of moments in the direction, for The order of moments in the direction, is the centroid of the time-frequency matrix.

[0012] Furthermore, in one embodiment of the present invention, the ST-TFM method in step S4 includes the following steps: Step S41: Perform local event likelihood inference based on the event feature information to obtain the initial likelihood. Step S42: Based on the initial likelihood and the environmental data, obtain a reputation score through a dynamic source reputation calibration mechanism; Step S43: Based on the reputation score and the meteorological data, road status information is obtained through global confidence fusion; the road status information includes road status and corresponding probability.

[0013] Furthermore, in one embodiment of the present invention, the initial likelihood mentioned in step S41 Through

[0014] Obtain, among which, Indicates the event to be detected. for Observe the data, for Bus observation data, Let be the likelihood probability of the sensor. Let be the conditional probability of the vehicle's state. Let be the prior probability of the event.

[0015] Furthermore, in one embodiment of the present invention, the dynamic source reputation calibration mechanism described in step S42 is achieved through...

[0016] Obtain the target vehicle Reputation rating at time t ;in, The learning rate; For the target vehicle Vehicle data; The environmental data; This is a matching function used to obtain the consistency between the initial likelihood and the environmental feature information.

[0017] Option 2: A driving warning method based on vehicle-road-cloud collaborative perception, including the following steps: Step S01: Based on the real-time location and planned route of the target vehicle to be warned, query the road status information in the electronic map as a structured risk parameter sequence; Step S02: Compare the structured risk parameter sequence with a safety threshold. If the threshold is exceeded, generate an early warning instruction. Step S03: Send the warning command to the target vehicle to be warned, and complete the intelligent driving warning; The electronic map is updated in real time using the electronic map update method based on vehicle-road-cloud collaborative perception in Scheme 1.

[0018] Option 3: An intelligent driving early warning system based on vehicle-road-cloud collaborative perception, including a data collaborative acquisition module, a cloud event feature extraction module, a cloud fusion and state calculation module, a forward-looking risk assessment module, an early warning decision generation module, and an early warning issuance module; The data collaborative acquisition module is used to collect vehicle data and environmental data of target vehicles within the target area; the vehicle data includes acceleration data. The cloud-based event feature extraction module is used to obtain meteorological data of the target area; it is also used to obtain event feature information by extracting all acceleration data through the E-VESA event feature extraction method. The cloud fusion and state calculation module is used to obtain road state information based on the event feature information, the environmental data, and the meteorological data using the ST-TFM method. The forward-looking risk assessment module is used to overlay the road condition information onto the geographic information layer of the target area in real time, so as to realize the real-time updating of the electronic map in the target area; and to query the road condition information in the electronic map as a structured risk parameter sequence based on the real-time location and planned route of the target vehicle. The early warning decision generation module is used to compare the structured risk parameter sequence with a safety threshold, and if the threshold is exceeded, an early warning instruction is generated. The warning delivery module is used to send the warning command to the target vehicle to complete the intelligent driving warning.

[0019] The electronic map update method and driving warning method based on vehicle-road-cloud collaborative perception described in this invention effectively alleviate the problem of inaccurate intelligent driving warnings caused by the limitations of existing single-vehicle intelligent perception systems in terms of perception dimension and range, which prevents them from accurately acquiring road physical conditions. This effectively improves the safety of intelligent driving. Specific beneficial effects include: 1. The electronic map update method described in this invention, compared with existing technologies, fails to effectively solve the problem of providing forward-looking safety warnings and comprehensive, real-time, and accurate perception of road physical conditions, whether from the perspective of vehicle intelligence or infrastructure.

[0020] This invention, by constructing an innovative architecture for collaborative perception across vehicle, roadside, and cloud platforms, overcomes the limitations of existing single-vehicle intelligent perception in terms of scope and dimensionality. It achieves beyond-line-of-sight, all-weather, and accurate perception of key road condition information such as road friction coefficient and damage index. This invention is a novel technical solution that organically integrates mobile vehicle perception data with fixed roadside perception data. Through collaborative intelligence, it overcomes the shortcomings of existing technologies, providing drivers and autonomous driving systems with reliable and timely electronic maps and decision support, thereby ensuring the safe and efficient operation of road traffic.

[0021] 2. The electronic map update method described in this invention can perform in-depth intelligent analysis of uploaded vehicle data through the E-VESA event feature extraction algorithm, extracting key feature information and laying the foundation for the subsequent accurate fusion of vehicle data, environmental data and meteorological data.

[0022] 3. The intelligent driving early warning method described in this invention, the ST-TFM spatiotemporal confidence fusion model, effectively solves the technical problem of inconsistent quality and difficulty in guaranteeing the credibility of multi-source heterogeneous data in open environments by introducing dynamic reputation calibration and probability inference mechanisms, and significantly improves the accuracy and robustness of the final perception results.

[0023] 4. The driving warning method described in this invention, based on the obtained electronic map, can provide drivers and autonomous driving systems with forward-looking and quantifiable proactive safety warnings, effectively improving driving safety and providing reliable data support for intelligent traffic management and road maintenance.

[0024] The method described in this invention is applicable to the field of beyond-line-of-sight quantitative assessment and proactive safety early warning of road physical conditions. Attached Figure Description

[0025] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of the electronic map updating method described in Implementation Method 1; Figure 2 This is a structural diagram of the intelligent driving early warning system described in Implementation Method Nine; Figure 3 This is an architecture diagram of the intelligent driving early warning system described in Implementation Method 10; Figure 4 This is a schematic diagram of the data acquisition interface of the roadside sensing subsystem described in Implementation Method 10; Figure 5 This is the original bump event location map described in Implementation Method 10; Figure 6 This is a schematic diagram of the dynamic road condition map described in Implementation Method 10; Figure 7 This is an example diagram illustrating the variation in the number of weekly turbulence events as described in Implementation Method 10; Figure 8 This is an example diagram illustrating the change in the number of slippery events per week as described in Implementation Method 10. Detailed Implementation

[0026] Various embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. The embodiments described with reference to the drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0027] Implementation Method 1: The electronic map update method based on vehicle-road-cloud collaborative perception described in this implementation method, such as... Figure 1 As shown, it includes the following steps: Step S1: Collect vehicle data and environmental data of the target vehicle within the target area; the vehicle data includes acceleration data; Step S2: Collect meteorological data for the target area; Step S3: Obtain event feature information from all acceleration data using the E-VESA event feature extraction method; Step S4: Based on the event feature information, the environmental data, and the meteorological data, obtain road condition information using the ST-TFM method; Step S5: The road status information is overlaid onto the geographic information layer of the target area in real time to achieve real-time updates of the electronic map within the target area.

[0028] In this embodiment, the data in step S1 is collected collaboratively by the sensor system deployed on the vehicle and the perception subsystems deployed at key road nodes, combining vehicle data and environmental data as multi-source heterogeneous environmental and road perception data, including: On the vehicle side, vehicle-side perception subsystems are deployed on multiple vehicles (industry vehicles such as taxis, buses, logistics vehicles, sanitation vehicles, or socially crowdsourced vehicles) to build a wide-area mobile perception network. The vehicle-side perception subsystem collects two types of core data in real time: one is vehicle dynamic data from the vehicle controller area network (CAN) bus, and the other is road micro-feature data from the inertial measurement unit (IMU).

[0029] On the roadside, fixed roadside sensing subsystems are deployed at key road nodes, such as bridges, ramps, tunnel entrances, sections prone to water accumulation or icing, and accident-prone areas.

[0030] In this embodiment, the environmental data mentioned in step S1 includes real-time road surface temperature, wetness / dryness, rainfall, wind speed, ambient temperature, humidity, etc. This precise, point-to-point meteorological data provides direct physical evidence for subsequent judgments on road surface conditions such as slipperiness and icing.

[0031] In this embodiment, the meteorological data mentioned in step S2 is obtained by actively and periodically accessing and integrating high-precision meteorological service data from third-party professional service providers through API calls, etc., to obtain macro-level environmental background information, such as regional rainfall forecasts and cold wave warnings.

[0032] In this embodiment, step S5, which involves overlaying the key state parameters calculated in the cloud onto a high-precision geographic information layer in real time to generate and continuously update a dynamic road condition map covering the entire area, and querying and evaluating the quantitative risk level of the target vehicle's driving path as presented on the dynamic map based on the target vehicle's real-time location and planned path, includes: After obtaining precise key state parameters such as the road friction coefficient and road damage index for each road segment through the ST-TFM model on the cloud platform, this implementation method structurally associates these parameters with their corresponding geographic information. This step binds the solution result of each road segment as a data attribute containing a quantified risk value to the corresponding road segment vector data in the high-precision geographic information map, thereby technically constructing and continuously maintaining a dynamic road state data layer covering the entire region. This data layer is dynamically refreshed as the cloud continuously receives new sensing data and the ST-TFM model updates the solution results in real time.

[0033] Optionally, the process further includes acquiring the target vehicle's real-time location information and future driving path information provided by the in-vehicle navigation system or autonomous driving planning module, and performing a spatial matching query operation. This involves using the vehicle's planned path sequence as an index to retrieve the key state parameters corresponding to each future road segment on that path from the dynamic road state data layer. The output of this operation is a structured sequence of risk parameters corresponding to the planned path sequence, containing the risk type and its quantification level for each road segment the vehicle is about to enter. This risk parameter sequence serves as objective and forward-looking data input and is passed to the next processing step for generating early warning decisions.

[0034] By adopting this approach, the solution result for each road segment can be treated as a data attribute containing a quantified risk value and bound to the corresponding road segment vector data in a high-precision geographic information map, constructing and continuously maintaining a dynamic road state data layer covering the entire region. This data layer is dynamically refreshed as new sensing data is continuously received from the cloud and the solution results are updated in real time by the ST-TFM model. Subsequently, after obtaining the real-time location and future travel path information of the target vehicle, the system will perform a spatial matching query operation, that is, using the vehicle's planned path sequence as an index, retrieving the key state parameters corresponding to each future road segment on the path from the dynamic road state data layer. The output of this operation is a structured risk parameter sequence corresponding to the planned path sequence, which is passed to the next processing step as an objective and forward-looking data input.

[0035] In this embodiment, the dynamic road condition data with high confidence, calculated by the ST-TFM model in step S4, is preferably provided as a standardized data product to third-party application systems. Furthermore, real-time road friction coefficient maps and road damage index heat maps from electronic maps are authorized for provision to traffic management departments, road maintenance departments, and autonomous driving systems.

[0036] The electronic map updating method described in this embodiment organically combines mobile sensing data, which has wide coverage but may fluctuate in accuracy, with fixed sensing data, which has limited coverage but extremely high accuracy, in step S1. Vehicle data provides a dynamic and real-time "crowdsourced" information foundation for the entire road network, while roadside environmental data provides continuous and reliable "ground truth" and calibration basis for the road conditions of key nodes. The two complement each other and together constitute a multi-source, heterogeneous, and high-value sensing data foundation, providing comprehensive and reliable data input for subsequent accurate road condition calculations. This is a technical advantage that existing single sensing methods do not possess.

[0037] Step S1, combined with the meteorological data obtained in step S2, completes the centralized aggregation of all basic sensing data. At this point, we have raw high-frequency dynamic response data from a large number of vehicles, precise environmental meteorological data from key fixed stations, and macro-meteorological background data from third-party services, thus constructing a complete data loop for a comprehensive, three-dimensional, and high-confidence road condition assessment.

[0038] In step S4, event feature information extracted through intelligent analysis using the cloud-based E-VESA algorithm is received; environmental data from the roadside sensing subsystem (mainly weather stations) in step S1 is received and processed, such as real-time road surface temperature, dryness / wetness, rainfall, wind speed, ambient temperature, and humidity. This precise, point-to-point meteorological data provides direct physical evidence for subsequent judgments regarding road surface slipperiness, icing, and other conditions. Furthermore, high-precision meteorological service data from third-party professional service providers is integrated and fused to obtain macro-level environmental background information, such as regional rainfall forecasts and cold wave warnings, which will serve as important prior knowledge. The aforementioned structured, spatiotemporally aligned road event feature information, the environmental data, and the meteorological data together constitute the high-quality input dataset required for executing the ST-TFM method.

[0039] In step S5, the calculated road condition information is overlaid onto the high-precision geographic information layer in real time to generate and continuously update a dynamic high-precision electronic road condition map covering the entire area. At the same time, the calculated high-quality dynamic road condition data can also be used as a standardized service to provide data support for third-party applications such as traffic management, road maintenance and autonomous driving systems.

[0040] Implementation Method Two: This implementation method further defines the electronic map update method described in Implementation Method One. In this implementation method, the E-VESA event feature extraction method in step S3 includes the following steps: Step S31: Remove noise from all acceleration data to obtain denoised acceleration data; Step S32: Perform continuous wavelet transform (CWT) on the acceleration data to obtain the time-frequency matrix; Step S33: Obtain the center distance of the time-frequency matrix using the two-dimensional moment invariant method; Step S34: Obtain morphological features based on the center distance; Step S35: The morphological features are processed through a neural network classifier to obtain the category probability. If the category probability is greater than the confidence threshold, structured road event feature information is generated. The road event feature information includes the road event type, occurrence probability, severity level, precise timestamp, GPS coordinates, and associated raw CAN data.

[0041] In this embodiment, step S31 preferably uses a digital bandpass filter to filter out noise in irrelevant frequency bands.

[0042] In this embodiment, in step S34, a set of Hu moment invariants that are invariant to image translation, rotation, and scaling can be derived using the central moment. These invariants are combined into a low-dimensional, robust eigenvector. This feature vector can stably characterize the core morphological features of an impact event.

[0043] In this embodiment, the neural network classifier mentioned in step S35 is a pre-trained lightweight neural network classifier. (For example, MobileNetV3-Small). This classifier is trained offline using a large amount of labeled data. Its inference process in the cloud can be represented as: The output is a tuple containing two parts: It is a probability vector representing the probability of the event belonging to N predefined categories (such as potholes, speed bumps, manhole covers, normal driving, etc.). It is a continuous value between [0, 1], representing the severity of the event, which is derived by the network from the impact characteristics through regression.

[0044] In this embodiment, when the category probability is in step S35 When the confidence level exceeds a preset high-confidence threshold (e.g., 0.95), an event is determined to have occurred, and a structured road event feature information is generated. This information is a data object (e.g., in JSON format) containing key fields such as road event type, probability of occurrence, severity level, precise timestamp, GPS coordinates, and associated raw CAN data (e.g., vehicle speed). This step completes the transformation from raw vibration data to high-value event information with clear semantics in the cloud.

[0045] In this embodiment, based on vehicle data, which includes multi-dimensional information such as vehicle dynamic data (CAN), high-frequency inertial measurement data (IMU), and vehicle radar data, the E-VESA event feature extraction algorithm can perform in-depth intelligent analysis on the raw sensor data, thereby accurately extracting key event information such as road damage, bumps, road surface wetness, and water accumulation.

[0046] This implementation further defines step S3, illustrating the E-VESA event feature extraction method with an example. This method identifies key road events such as potholes and bumps from the acceleration data stream. It completes the preprocessing and aggregation of all input data, including vehicle-side feature information extracted through intelligent analysis by the cloud-based E-VESA algorithm, environmental feature information from roadside weather stations, and macroscopic background feature information from third-party meteorological services. This structured, spatiotemporally aligned data collectively constitutes the high-quality input dataset required for the next stage of executing the ST-TFM spatiotemporal confidence fusion model.

[0047] Implementation Method 3: This implementation method further defines the electronic map updating method described in Implementation Method 2. In this implementation method, the continuous wavelet transform (CWT) mentioned in step S32 is achieved through:

[0048] Obtain the time-frequency matrix ,in, For scale parameters; For position parameters; In the vertical direction The first moment The acceleration data mentioned in the paragraph; It is the conjugate of the mother wavelet function.

[0049] In this embodiment, the Mexican Hat Wavelet is preferably used as the mother wavelet function because its shape is similar to the road impact signal.

[0050] This embodiment further defines step S32, illustrating the Continuous Wavelet Transform (CWT) scheme. CWT analyzes each acceleration data segment. CWT decomposes the data through scaling and shifting a mother wavelet function. Through CWT, the one-dimensional time series is transformed into an information-rich two-dimensional time-frequency graph, forming a unique "time-frequency fingerprint" of the event, accurately capturing "when" and "at what frequency" the impact event occurred.

[0051] Implementation Method Four: This implementation method further defines the electronic map updating method described in Implementation Method Two. In this implementation method, the two-dimensional moment invariant method in step S33 is achieved through:

[0052] Obtain center distance ,in, for The order of moments in the direction, for The order of moments in the direction, is the centroid of the time-frequency matrix.

[0053] This embodiment further defines step S33 and provides an example of the two-dimensional moment invariant method, which can reduce data dimensionality and extract robust features.

[0054] Implementation Method 5: This implementation method further defines the electronic map update method described in Implementation Method 1. In this implementation method, the ST-TFM method in step S4 includes the following steps: Step S41: Perform local event likelihood inference based on the event feature information to obtain the initial likelihood. Step S42: Based on the initial likelihood and the environmental data, obtain a reputation score through a dynamic source reputation calibration mechanism; Step S43: Based on the reputation score and the meteorological data, road status information is obtained through global confidence fusion; the road status information includes road status and corresponding probability.

[0055] In this embodiment, the global confidence fusion in step S43 to obtain road state information includes: The model fuses all available evidence within a unified Bayesian framework to update the state of a given road segment (e.g., friction coefficient). posterior probability distribution Its expression is:

[0056] In the formula, For environmental and historical prior probabilities, it represents the initial estimate of the friction coefficient of the road segment before any real-time vehicle data is seen; This is the weighted joint likelihood of all relevant data evidence. Among them, the data points... final dynamic weights It is a comprehensive reflection of multiple factors, and its expression is:

[0057] in It is a dynamic reputation score calculated by the submodule. It is a time decay factor used to reduce the impact of outdated data. It is a spatial consensus enhancement factor, used to increase the influence of events that have obtained verification from surrounding data.

[0058] Among them, weight It innovatively combines dynamic reputation scores calculated by submodules. Time decay factor used to reduce the impact of outdated data And spatial consensus enhancement factors used to improve the impact of events by obtaining verification from surrounding data. .

[0059] This implementation further defines step S4, illustrating the ST-TFM method with an example. It processes the aggregated multi-source data, dynamically calculating a confidence score for each data point based on data source, timeliness, spatial consensus, and event relevance. This score is then used as weights for weighted fusion, ultimately accurately calculating key state parameters such as the dynamic road friction coefficient and road damage index. This method scientifically quantifies the dynamic credibility of each uploaded data point to address the deep uncertainty issues arising from diverse data sources, varying quality, and uneven spatiotemporal distribution in an open crowdsourced perception environment.

[0060] This method employs a phased processing architecture, decomposing the complex fusion process into two core stages: local event likelihood inference and global confidence fusion. The model not only provides the most probable road state estimate but also quantifies the uncertainty of that estimate. Furthermore, it can learn and dynamically calibrate the credibility of different data sources, sending a signal to the decision-making system (especially autonomous driving systems) that "I am not entirely confident in my current judgment," allowing the vehicle to adopt a more conservative driving strategy.

[0061] Specifically, in this embodiment, to address the challenges of deep uncertainty and dynamic credibility in large-scale open crowdsourced perception systems, an original ST-TFM model is employed. The core objective of this model is to upgrade the solution of road states from a deterministic weighted calculation process to a dynamic, adaptive probabilistic inference process. This method decomposes the complex fusion process into two stages through an innovative, phased architecture: a local event likelihood inference stage and a global confidence fusion stage. The local event likelihood inference stage is step S41, in which the received data packets reported by the vehicle are initially processed in the cloud. By analyzing the logical correlation between data from various sensors inside the vehicle, the initial likelihood of the event reported by the vehicle is inferred.

[0062] The global confidence fusion stage consists of steps S42 to S44, in which the initial likelihood, environmental data, and meteorological data from different vehicles, which have been processed in the first stage, are finally probabilistically fused.

[0063] Implementation Method Six: This implementation method further defines the electronic map updating method described in Implementation Method Five. In this implementation method, the initial likelihood mentioned in step S41... Through

[0064] Obtain, among which, Indicates the event to be detected. For IMU observation data, for Bus observation data, Let be the likelihood probability of the sensor. Let be the conditional probability of the vehicle's state. Let be the prior probability of the event.

[0065] In this embodiment, the initial likelihood To determine the posterior probability of a pothole event actually occurring by combining IMU and CAN data; sensor likelihood probability. To determine the probability of observing the current IMU vibration characteristics (extracted by the vehicle-side E-VESA algorithm) under the condition of a real pothole event; the prior probability of the event. This represents the initial probability of a pothole existing at a given location, without considering any real-time sensor data from the current vehicle. This probability can be set based on macroscopic information such as historical data statistics, road grade, and seasonal factors.

[0066] This implementation further defines step S41, providing an example of local event likelihood inference. The local event likelihood inference stage performs preliminary processing of the received road event feature information in the cloud. Its purpose is not to directly accept the data, but to infer the initial likelihood of the event reported by the vehicle by analyzing the logical correlation between data from various sensors inside the vehicle. This process is equivalent to conducting a "pre-screening" of the internal evidence chain before trusting a report. When a "bump event" is reported by a vehicle at a certain location, this stage comprehensively analyzes the IMU vibration characteristics and CAN bus data within the event packet. The inference process is based on Bayesian network modeling. Through this stage, each isolated event report is assigned an initial probability value based on its internal data consistency, providing high-quality input for the next stage of global fusion.

[0067] Implementation Method Seven: This implementation method further defines the electronic map updating method described in Implementation Method Five. In this implementation method, the dynamic source reputation calibration mechanism described in step S42 is achieved through...

[0068] Obtain the target vehicle Reputation rating at time t ;in, The learning rate; For the target vehicle Vehicle data; The environmental data; This is a matching function used to obtain the consistency between the initial likelihood and the environmental feature information.

[0069] In this embodiment, the learning rate The matching function is used to control the update speed. Used to calculate the consistency between data.

[0070] This embodiment further defines step S42, illustrating the dynamic source reputation calibration mechanism. This mechanism abandons static data source weights and uses data from the roadside perception subsystem as the "ground truth" or "examiner" for a local area to continuously and dynamically calibrate the reputation score Ri of each crowdsourced vehicle. The initial value of the reputation score can be set according to the weight of the basic data source, for example, 0.95 for roadside devices. Subsequently, when a vehicle passes through the coverage area of ​​the roadside subsystem, the system automatically compares the consistency between the data reported by the vehicle and the actual roadside data, and updates its reputation score through an online learning method, namely exponential moving average. Through this mechanism, vehicles that provide accurate data over a long period will gradually increase their reputation score, thus allowing the system to "automatically" trust more reliable data sources in fusion decision-making.

[0071] Implementation method eight, the driving early warning method based on vehicle-road-cloud cooperative perception described in this implementation method includes the following steps: Step S01: Based on the real-time location and planned route of the target vehicle to be warned, query the road status information in the electronic map as a structured risk parameter sequence; Step S02: Compare the structured risk parameter sequence with a safety threshold. If the threshold is exceeded, generate an early warning instruction. Step S03: Send the warning command to the target vehicle to be warned, and complete the intelligent driving warning; The electronic map is updated in real time using any one of the electronic map update methods based on vehicle-road-cloud collaborative perception, from implementation method one to seven.

[0072] In this embodiment, step S02 involves real-time determination of whether the risk level of the road segment ahead exceeds a pre-set safety threshold. If the threshold is exceeded, an early warning decision-making mechanism is immediately triggered to generate a structured early warning instruction containing the specific risk type, precise location, and corresponding countermeasures, including: After obtaining a structured sequence of risk parameters corresponding to the target vehicle's driving path, it is compared with an adaptive safety threshold that is dynamically adjusted based on the vehicle's current dynamic state and risk type.

[0073] When the risk level is determined to exceed the threshold, a structured warning instruction is generated. The instruction includes the risk type, precise location, quantified risk level, a confidence field derived from the ST-TFM model output, and suggested countermeasures.

[0074] By adopting the above technical solution, when the module determines that the quantitative risk level of the road section ahead exceeds the currently calculated adaptive safety threshold, the system will immediately trigger the early warning decision mechanism.

[0075] Optionally, the step of sending warning commands to target vehicles via vehicle-to-everything (V2X) communication and providing clear risk alerts to drivers through the in-vehicle human-machine interface in the form of voice or vision, while simultaneously providing standardized services using the calculated high-quality dynamic road condition data to support third-party applications such as traffic management, road maintenance, and autonomous driving systems, including: After generating a structured warning instruction, the instruction will be sent via vehicle-to-everything (V2X) communication to target vehicles that meet the warning criteria and are located ahead of the risky road section. Upon receiving the warning instruction, the vehicle's onboard communication unit will parse it and extract key fields such as risk type, precise location, quantification level, confidence level, and suggested countermeasures.

[0076] Subsequently, the in-vehicle human-machine interface system will provide clear, non-intrusive risk warnings to the driver in one or more forms based on the parsed information. In this way, the driver can have sufficient reaction time to learn about and understand potential hazards ahead, and thus take appropriate driving actions.

[0077] The driving warning method described in this embodiment queries and evaluates the quantitative risk level of the road segment ahead of the target vehicle on the dynamic map based on the real-time location and planned route of the target vehicle. In step S02, it is determined in real time whether the risk level of the road segment ahead output by the forward-looking risk assessment module exceeds the preset safety threshold. If it exceeds the threshold, the early warning decision mechanism is immediately triggered to generate a structured early warning instruction containing the specific risk type, precise location and suggested countermeasures. In step S03, the generated warning command is sent to the target vehicle through vehicle-to-everything (V2X) communication. The driver can also be given a clear risk warning through the vehicle-mounted human-machine interface in the form of voice or vision.

[0078] Implementation Method Nine: This implementation method describes an intelligent driving early warning system based on vehicle-road-cloud collaborative perception, such as... Figure 2 As shown, it includes a data collaborative acquisition module, a cloud event feature extraction module, a cloud fusion and status calculation module, a forward-looking risk assessment module, an early warning decision generation module, and an early warning issuance module; The data collaborative acquisition module is used to collect vehicle data and environmental data of target vehicles within the target area; the vehicle data includes acceleration data. The cloud-based event feature extraction module is used to obtain meteorological data of the target area; it is also used to obtain event feature information by extracting all acceleration data through the E-VESA event feature extraction method. The cloud fusion and state calculation module is used to obtain road state information based on the event feature information, the environmental data, and the meteorological data using the ST-TFM method. The forward-looking risk assessment module is used to overlay the road condition information onto the geographic information layer of the target area in real time, so as to realize the real-time updating of the electronic map in the target area; and to query the road condition information in the electronic map as a structured risk parameter sequence based on the real-time location and planned route of the target vehicle. The early warning decision generation module is used to compare the structured risk parameter sequence with a safety threshold, and if the threshold is exceeded, an early warning instruction is generated. The warning delivery module is used to send the warning command to the target vehicle to complete the intelligent driving warning.

[0079] In this embodiment, the data collaborative acquisition module includes a vehicle-side perception submodule deployed on the vehicle and a road perception submodule deployed at key road nodes, collaboratively acquiring vehicle data and environmental data of the target area as multi-source heterogeneous environmental and road perception data.

[0080] In this embodiment, the vehicle-side perception submodule encapsulates the raw high-frequency acceleration data stream segments collected by the vehicle and the synchronously collected CAN data points, along with metadata such as precise timestamps, geographic coordinate information, and vehicle unique identifiers, into a standardized data packet format, which serves as vehicle data.

[0081] In this embodiment, the cloud event feature extraction module, the cloud fusion and state calculation module, the forward-looking risk assessment module, the early warning decision generation module, and the early warning issuance module are integrated into the cloud data platform.

[0082] The vehicle-side perception submodule uploads vehicle data to the cloud data platform in real time via a wireless communication network. The roadside sensing subsystem (mainly a weather station) uploads the environmental data it collects to the cloud data platform in real time; The cloud data platform actively accesses and integrates high-precision meteorological service data from third-party service providers to obtain macro-level environmental background information.

[0083] In this embodiment, the data collaborative acquisition module is responsible for performing collaborative acquisition of multi-source heterogeneous data. Specifically, this module is used to control and acquire vehicle dynamic data and road micro-feature data from vehicle-side perception subsystems deployed on multiple vehicles, and simultaneously control and acquire local environmental and road physical state data from roadside perception subsystems deployed at key road nodes; The cloud-based event feature extraction module is responsible for executing the E-VESA (Edge Event Feature Analysis) algorithm on the cloud platform. This module processes the collected high-frequency raw data in real time, intelligently extracting the continuous data stream into structured, lightweight event feature information that includes event type, severity level, and location information. The cloud-based fusion and state calculation module is responsible for receiving and aggregating data uploaded from vehicle-side, roadside, and third-party platforms, and executing the core ST-TFM (Spatiotemporal Confidence Fusion Model) algorithm. This module processes the aggregated multi-source data, dynamically calculates a confidence score for each data point, and uses this score as a weight for weighted fusion to ultimately accurately calculate key state parameters such as the road friction coefficient and road damage index. The forward-looking risk assessment module is responsible for overlaying key state parameters calculated in the cloud onto a high-precision geographic information layer to generate a dynamic road condition map. This module is also used to query and assess the quantitative risk level of a target vehicle's driving path as presented on the dynamic map, based on the vehicle's real-time location and planned route. The early warning decision generation module is responsible for determining in real time whether the risk level of the road segment ahead, output by the forward-looking risk assessment module, exceeds a preset adaptive safety threshold. If it exceeds the threshold, the early warning decision mechanism is immediately triggered, generating a structured early warning instruction that includes the specific risk type, precise location, and suggested countermeasures. The early warning issuance and service module is responsible for sending generated early warning commands to target vehicles via vehicle-to-everything (V2X) communication and providing risk alerts through the in-vehicle human-machine interface. Simultaneously, this module also provides standardized services using the calculated high-quality dynamic road condition data, offering data support to third-party applications such as traffic management, road maintenance, and autonomous driving systems via application programming interfaces (APIs).

[0084] Implementation Method 10: This implementation method employs an electronic map update method and a driving warning method based on vehicle-road-cloud collaborative perception. The intelligent driving warning method is based on the electronic map update method based on vehicle-road-cloud collaborative perception described in Implementation Method 1, combined with the E-VESA event feature extraction method optimized in Implementation Method 2, the continuous wavelet transform (CWT) optimized in Implementation Method 3, the two-dimensional moment invariant method optimized in Implementation Method 4, the ST-TFM method optimized in Implementation Method 5, the local event likelihood inference optimized in Implementation Method 6, and the dynamic source reputation calibration mechanism optimized in Implementation Method 7. The driving warning method based on vehicle-road-cloud collaborative perception is based on the driving warning method described in Implementation Method 8. The intelligent driving warning system is based on the intelligent driving warning system described in Implementation Method 9, such as... Figure 3 As shown.

[0085] Step A1: Collect multi-source heterogeneous environmental and road perception data in collaboration with sensor systems deployed on vehicles and perception subsystems deployed at key road nodes.

[0086] Specifically, to construct a perception network with broad coverage and rich information dimensions, data collection is achieved through the coordinated deployment of mobile and fixed perception units. The mobile perception units are vehicle-mounted perception subsystems deployed on multiple vehicles, while the fixed perception units are roadside perception subsystems deployed at key road nodes. This step may include the following sub-steps: Step A11: Access the vehicle controller local area network (CAN) bus through the vehicle-side perception subsystem to build a mobile perception network and collect vehicle dynamic data; Specifically, at the vehicle end, vehicle-side perception subsystems are deployed on multiple vehicles to construct a wide-area mobile perception network. These subsystems connect to the vehicle's CAN bus via a standard interface, reading and parsing parameters reflecting the vehicle's real-time operating status in a real-time, non-intrusive manner, forming a vehicle dynamic dataset. This dynamic data may include wheel speeds, steering wheel angle, and braking and accelerator pedal states. Furthermore, this step also includes acquiring the activation status of the Anti-lock Braking System (ABS) or Electronic Stability Program (ESP), which directly reflects the tire-road adhesion state. The purpose of obtaining the ABS / ESP activation status is that the intervention of these systems directly indicates that the adhesion between the vehicle's tires and the road surface has reached its limit, providing strong evidence of whether the road surface is wet or icy, and offering crucial underlying physical information for subsequent state calculations.

[0087] Step A12: Collect road micro-feature data through the inertial measurement unit (IMU) built into the vehicle-side perception subsystem; Specifically, in addition to collecting data on the vehicle's own state, the vehicle-side perception subsystem also uses its built-in inertial measurement unit (IMU), particularly a high-precision triaxial accelerometer, to sense changes in vehicle attitude caused by road morphology. This step involves continuously collecting vertical acceleration data streams during vehicle movement at a high sampling frequency (e.g., 100Hz or higher) to accurately quantify and characterize road surface irregularities such as bumps, potholes, cracks, and manhole cover settlement. These high-frequency vibration signals directly reflect the microscopic physical morphology of the road and are the core source of raw data for subsequent calculations of the road damage index.

[0088] Step A13: Deploy fixed roadside sensing subsystems at key road nodes, construct a fixed sensing network, and continuously collect high-precision road surface physical state data; Specifically, along the roadside, fixed roadside sensing subsystems are deployed at key nodes to continuously and accurately monitor the condition of specific high-risk or critical locations. These key nodes can be selected based on historical traffic and road engineering data, such as bridge sections (prone to icing) and low-lying urban sections (prone to water accumulation). This step involves deploying remote road condition sensors and pavement temperature sensors at these nodes to continuously collect high-precision pavement physical condition data, such as real-time pavement temperature, dry / wet condition, water film thickness, and the presence of icing. Because this data has a fixed source and has undergone professional calibration, it can serve as high-confidence benchmark data in subsequent fusion algorithms.

[0089] Step A14: Collect local micro-meteorological data as environmental data through the roadside sensing subsystem.

[0090] Specifically, to comprehensively understand the environmental factors affecting road conditions, the roadside sensing subsystem also includes equipment for collecting local micro-meteorological data, such as weather transmitters or integrated micro-weather stations. This step aims to continuously collect and record real-time meteorological parameters around the node, such as ambient temperature, humidity, rainfall, wind speed, and visibility. This local micro-meteorological data, combined with the road surface physical state data collected in step A13, can provide direct environmental evidence for determining the causes and trends of risks such as road icing and water accumulation. (Refer to...) Figure 4 , Figure 4 This is a schematic diagram of the data acquisition interface of the roadside sensing subsystem (weather station) in this embodiment. As shown in the figure, the roadside sensing subsystem (mainly a weather station) can collect and report precise physical and environmental parameters of key nodes in real time, such as... Figure 4The pop-up window displays air temperature (14℃), road temperature (14℃), relative humidity (93%), water film thickness (1.05mm), and slipperiness level (0.6). This high-precision ground condition data will be uploaded to the cloud platform as one of the data sources in step A2, and will serve as the benchmark data for reputation calibration (DSRC) and fusion solution of the ST-TFM model in subsequent step A4.

[0091] Step A2: Collect and upload the original dynamic data from the vehicle and the roadside weather station to the cloud platform, and access third-party data.

[0092] Specifically, in the embodiments of the present invention, to achieve centralized advanced analytics in the cloud, it is necessary to uniformly aggregate data distributed across vehicles and roadsides. The purpose of this step is to securely and efficiently transmit the various types of raw data collected in the preceding step (A1) to the cloud data platform and integrate necessary external data sources, providing a complete and consistent data foundation for subsequent cloud-based intelligent analysis (such as the event feature extraction to be performed in step A3 and the fusion calculation in step A4). This step may specifically include the following sub-steps: Step A21: Prepare and upload the original dynamic data package from the vehicle side; Specifically, within the vehicle-side perception subsystem, the raw high-frequency acceleration data stream segments collected and cached in steps A11 and A12 are... The system, along with synchronously collected CAN data points and metadata such as precise GPS timestamps, geographic coordinates, and vehicle unique identifiers, is encapsulated into a standardized data packet format. To ensure security and privacy during data transmission, lightweight encryption algorithms (such as AES) can be used to encrypt the data packets. Subsequently, the vehicle-side perception subsystem utilizes its integrated wireless communication module to reliably upload the packaged and encrypted data packets to a pre-defined cloud data platform receiving endpoint in real time via high-bandwidth, low-latency cellular networks such as 4G / 5G. Transmission optimization strategies such as breakpoint resumption and data compression can be employed to address potential network fluctuations and ensure data integrity and transmission efficiency. Through this process, the cloud platform can continuously receive raw, high-frequency dynamic response data from a large number of vehicles, either uncompressed or only after basic filtering and synchronization processing, along with their synchronized vehicle status information.

[0093] Step A22: Collect and upload data from the roadside weather station; Specifically, after continuously collecting environmental data in steps A13 and A14, the roadside sensing subsystems (mainly weather stations) deployed at key road nodes also periodically or when significant changes occur, uploading data containing precise environmental parameters such as road surface temperature, wetness / dryness, rainfall, and wind speed, along with their corresponding station identifiers and timestamps, to the same cloud data platform via their communication interfaces. This high-precision ground environmental data from fixed nodes is crucial for subsequent road condition assessments (such as icing or slippery conditions).

[0094] Step A23: The cloud data platform actively accesses and integrates high-precision meteorological service data from third-party service providers.

[0095] Specifically, to further enrich the data dimensions and improve the accuracy of subsequent state calculations, the cloud data platform of this invention will proactively and periodically access and integrate high-precision meteorological service data from third-party professional service providers through application programming interfaces (APIs). The purpose of this step is to acquire macro-level environmental background information, such as regional rainfall forecasts, cold wave warnings, and large-scale gridded data on road surface temperature and humidity covering the entire urban road network. This macro-meteorological data provides the system with an important prior knowledge base. For example, when a cold wave warning is received, the system can increase its sensitivity to "road icing" events in advance and dynamically adjust the prior probability settings of the subsequent ST-TFM model, thereby improving the model's predictive ability and accuracy.

[0096] By adopting the above technical solution, the cloud data platform has completed the centralized aggregation of all basic sensing data. At this point, the platform has acquired raw high-frequency dynamic response data from a large number of vehicles, precise environmental meteorological data from key fixed stations, and macro-meteorological background data from third-party services. This has built a complete data loop for a comprehensive, three-dimensional, and high-confidence road condition assessment, and prepared the platform for the next step (step A3) to perform event feature extraction in the cloud.

[0097] Step A3: Execute the E-VESA event feature extraction algorithm in the cloud and aggregate multi-source data.

[0098] Specifically, in this embodiment of the invention, after receiving the raw high-frequency dynamic data packets from the vehicle-side system uploaded in step A2, and the data from the roadside perception subsystem, the cloud data platform will perform key event feature extraction and multi-source data aggregation operations. The purpose of this step is to transform the raw, semantically undefined data stream into structured information usable by the subsequent fusion model, and to aggregate all necessary data sources to prepare high-quality input for the next stage of the ST-TFM fusion model. This step may specifically include the following sub-steps: Step A32: Collect data from roadside weather stations and high-precision meteorological data from third parties.

[0099] Reference Figure 5 , Figure 5 This is the original bump event location map generated after the E-VESA algorithm is executed in the cloud in this embodiment. As shown in the figure, after receiving and processing the original IMU data streams from a large number of vehicles (as described in step A31), the cloud platform uses CWT, Hu moment invariants and neural network classifiers to solve them into a massive number of dense "bump points" (as shown by the yellow triangles in the figure) and overlays them on the map. Figure 5 The pop-up information (license plate number: F609, event: bumpy spot, reporting time: 2022-09-23 10:33:26) is a structured event feature information generated in the cloud by the E-VESA algorithm. The original event location set shown in this figure will be used as input to the next step (A4) of the ST-TFM model for confidence evaluation and fusion.

[0100] This step may specifically include the following sub-steps: Step A42: Dynamically evaluate and update the reputation of the vehicle-side data source through a dynamic source reputation calibration mechanism; Specifically, to address the issue of dynamically changing data quality across different vehicles, this invention abandons static data source weights and innovatively employs a dynamic source reputation calibration mechanism. This mechanism uses data from the roadside perception subsystem as the "ground truth" or "examiner" for a local area, continuously and dynamically calibrating the reputation score of each crowdsourced vehicle. The initial value for the reputation score can be set based on the weights of the underlying data sources, for example, 0.95 for roadside devices and 0.75 for industry vehicles. Subsequently, when a crowdsourced vehicle... When a vehicle passes through an area covered by the roadside subsystem, the system automatically compares the data reported by the vehicle with the measured data from the roadside subsystem, and updates its reputation score using an online learning method, namely exponential moving average. The expression for this is:

[0101] In the formula, For data source Updated reputation score; Assign a credit score to the current credit rating. The learning rate is used to control the update speed. This is a matching function used to calculate the consistency between vehicle data and roadside equipment data. The higher the consistency, the closer the score is to 1. Through a dynamic source reputation calibration mechanism, the reputation score of "honest" vehicles that provide accurate data over a long period will gradually increase, thus allowing the system to "automatically" trust more reliable data sources in fusion decision-making.

[0102] Step A43: Perform global confidence fusion and finally calculate the road state parameters including uncertainty metrics; Specifically, the model fuses all available data within a unified Bayesian framework to update the state of a road segment (e.g., friction coefficient). posterior probability distribution Its expression is:

[0103] In the formula, For environmental and historical prior probabilities, this represents our initial estimate of the friction coefficient for that road segment before seeing any real-time vehicle data. For example, if a high-precision weather service reports "road icing," this prior distribution would be concentrated in the low-value range. This is the weighted joint likelihood of all relevant data evidence. Among them, the data points... final dynamic weights It is a comprehensive reflection of multiple factors, and its expression is:

[0104] This weighting incorporates the dynamic reputation score calculated by the dynamic source reputation calibration mechanism module. Time decay factor used to reduce the impact of outdated data And spatial consensus enhancement factors used to improve the impact of events by obtaining verification from surrounding data. .

[0105] By adopting the above technical solution, the final output of the ST-TFM model is not a single numerical value, but a probability distribution, such as a normal distribution. Among them, the expected value This represents the most likely estimate of the road friction coefficient, while the variance... This represents the system's uncertainty about the estimated value. When the data is sparse or contradictory, the variance will be large, which can send a signal to the decision-making system (especially autonomous driving systems) that "I am not entirely confident in my current judgment," allowing the vehicle to adopt a more conservative driving strategy.

[0106] Step A5: Overlay the key status parameters calculated in the cloud onto a high-precision geographic information layer in real time to generate and continuously update a dynamic road condition map covering the entire area. Based on the real-time location and planned route of the target vehicle, query and assess the quantitative risk level of its driving path as presented on the dynamic map. (Refer to...) Figure 6 , Figure 6This is a schematic diagram of the dynamic traffic map (comprehensive risk) generated after the cloud-based ST-TFM model fusion in this embodiment. As shown in the figure, this dynamic map is the final output of step A4 and constitutes the dynamic traffic map described in step A5. It no longer displays all the original event points generated in step A3 (such as...). Figure 5 Instead of displaying the map as shown in the image, it only shows high-confidence risk points confirmed after confidence fusion using the ST-TFM model (as indicated by the yellow triangles). Furthermore, the map also integrates roadside weather station data, high-precision meteorological data, and vehicle CAN bus data (such as ABS activation status) to calculate the road surface slippage level and... Figure 6 The legend ("Dry-Wet-Waterlogged") in the upper right corner and the road markings of different colors (as shown by the green markings in the figure) provide a visual representation. This comprehensive risk map forms the objective data basis for the present invention to perform forward-looking risk assessments and generate early warnings (step A6).

[0107] Specifically, in embodiments of the present invention, to transform the abstract data calculated in the preceding steps into spatial intelligent information that can directly guide driving decisions, this method deeply binds these quantified road state parameters with geospatial information and provides personalized, future-oriented risk assessments for vehicles in motion. This step may specifically include the following sub-steps: Step A51: Based on the key state parameters, generate and dynamically maintain a road state data layer covering the entire area; Specifically, after obtaining precise key state parameters such as the road friction coefficient and road damage index for each road segment through the ST-TFM model on the cloud platform, this invention structurally correlates these parameters with their corresponding geographic information. This step treats the solution result for each road segment as a data set containing a quantified risk value and its uncertainty (i.e., the variance of the probability distribution). The data attributes are bound to the corresponding road segment vector data in the high-precision geographic information map. In this way, a dynamic, full-area road condition data layer is technically constructed and continuously maintained. This data layer is dynamically refreshed as new sensing data is continuously received from the cloud and the ST-TFM model updates its solution results in real time, ensuring that it reflects the latest actual road conditions. In one feasible implementation, this layer can be visualized as a risk heat map; for example, the friction coefficient values ​​of different intervals can be rendered in different colors from green to red, thus presenting the physical state of the entire urban road network in an intuitive way.

[0108] Step A52: Obtain the real-time location and future driving path information of the target vehicle; Specifically, to achieve proactive early warning, this invention acquires relevant driving information of the target vehicle in real time through vehicle-to-everything (V2X) communication. This information includes the vehicle's current precise location, speed, and heading angle provided by the vehicle's onboard GPS. Simultaneously, the system also obtains the vehicle's future driving path information from its onboard navigation system or autonomous driving planning module. This path information is presented as a sequence of future road segments arranged in temporal or spatial order, such as an ordered list containing a series of unique link identifiers (Link IDs) or geographic coordinates, covering a range of several kilometers to tens of kilometers in the future.

[0109] Step A53: Perform a spatial matching query to obtain a sequence of structured risk parameters corresponding to the vehicle's travel path.

[0110] Specifically, this invention performs a spatial matching query operation, that is, using the vehicle planned path sequence obtained in step A52 as an index, it searches within the dynamic road state data layer generated in step A51. This operation matches each future road segment in the path sequence with the road segment vector data in the data layer and extracts the key state parameters corresponding to that road segment. The output of this operation is a structured sequence of risk parameters corresponding to the planned path sequence. For example, a JSON array where each element represents a future road segment and contains a unique identifier for that road segment, its distance from the current location, and the expected road friction coefficient. and its uncertainty and road damage index The risk parameter sequence, as an objective and forward-looking data input, is passed to the next processing step of this invention for generating early warning decisions. Through this step, this invention achieves "beyond-line-of-sight" perception, enabling vehicles to know in advance potential physical risks on the road beyond their line of sight.

[0111] Step A6: In real time, determine whether the risk level of the road section ahead exceeds the preset safety threshold. If it exceeds the threshold, immediately trigger the early warning decision mechanism to generate a structured early warning instruction containing the specific risk type, precise location, and countermeasures.

[0112] Specifically, in this invention, to transform the objective risk data calculated in the preceding steps into active safety commands that can directly guide driving behavior, this method performs real-time risk assessment and decision generation on the acquired risk parameter sequence. This step may specifically include the following sub-steps: Step A61: Calculate the adaptive safety threshold based on the real-time dynamic status of the target vehicle and the type of risk ahead; Specifically, this invention employs an adaptive safety threshold for risk assessment, rather than a fixed constant. This is based on the fact that the same road risk poses different levels of danger to vehicles in different conditions. For example, a moderate reduction in the coefficient of friction (such as...) What might not pose a serious threat to a small car traveling at low speed could mean a risk of loss of control to a heavy truck traveling at high speed. Therefore, the adaptive safety threshold is a dynamic variable that comprehensively considers the vehicle's current dynamic state and the type of risk. In one feasible implementation, the threshold is dynamically adjusted based on real-time speed obtained from the vehicle, vehicle type (e.g., heavy truck or small car), and specific risk type (e.g., "low friction coefficient" or "road damage") obtained from a risk parameter sequence. By employing an adaptive threshold, more personalized and targeted assessments of risks for different vehicles and driving conditions can be achieved, thereby improving the accuracy and effectiveness of early warnings.

[0113] Step A62: Compare the risk values ​​in the structured risk parameter sequence with the adaptive safety threshold in real time to determine whether an early warning is triggered; Specifically, this invention compares each risk value in the risk parameter sequence generated in step A5 with its corresponding adaptive safety threshold calculated in step A61 in real time. For example, for friction coefficient risk, the system will compare the expected value output by the ST-TFM model... Compared with the safety threshold of the friction coefficient, when If the water level is below this threshold, a risk of slipperiness is assessed. For road damage risk, the system will use a road damage index. Compared with the safety threshold of the damage index, when If the value exceeds this threshold, a risk of bumpiness is identified. This comparison process is continuous, ensuring real-time monitoring of all potential risk points along the vehicle's path.

[0114] Step A63: When the risk level is determined to exceed the threshold, generate a structured early warning instruction containing key fields such as specific risk type, precise location, response measures, and confidence level.

[0115] Specifically, when the judgment result of step A62 is true, i.e., the risk level exceeds the threshold, the system will immediately trigger the early warning decision mechanism and generate a structured, machine-readable early warning instruction. The early warning instruction is designed as a data object containing multiple key fields, such as a JSON object. This instruction includes a risk type field to specify whether it is a "low friction" or "severe turbulence" risk; a precise location field to indicate the geographical coordinates of the risk's starting point; a quantified risk level field, i.e., the actual calculated risk value; and a crucial confidence level field. The confidence level is derived from the variance of the probability distribution output by the ST-TFM model. This clearly indicates the reliability of the warning to the receiving end, a key feature distinguishing this invention from traditional deterministic warnings. Furthermore, the instruction includes a response field, the content of which is matched against a pre-defined driving strategy knowledge base based on the risk type and severity level, such as "recommend reducing speed to below 50 km / h" or "recommend avoiding emergency lane changes." By generating such a structured and comprehensive warning instruction, this invention ensures that downstream execution units can clearly and unambiguously understand the specific circumstances of the risk and take the most appropriate response.

[0116] Step A7: Send the warning command to the target vehicle through vehicle-to-everything (V2X) communication, and provide the driver with clear risk warnings in the form of voice or vision through the in-vehicle human-machine interface. At the same time, provide data support for third-party applications such as traffic management, road maintenance and autonomous driving systems as a standardized service using the calculated high-quality dynamic road condition data.

[0117] Specifically, in this embodiment of the invention, to effectively transmit the early warning decisions generated in the preceding steps to end users and realize their application value, this method uses vehicle-to-everything (V2X) communication technology to issue instructions and achieves information sharing and empowerment through standardized data interfaces. This step may specifically include the following sub-steps: Step A71: Using vehicle-to-everything (V2X) communication technology, the structured warning command is sent to the target vehicle and parsed. Specifically, after generating a structured warning instruction, this invention will send the instruction to target vehicles that are ahead of the risky road section and meet the warning conditions via vehicle-to-everything (V2X) communication technology. The V2X communication technology can utilize 4G / 5G cellular networks to achieve communication between the vehicle and the cloud platform, or, in scenarios with suitable hardware, employ C-V2X direct communication technology. Upon receiving the warning instruction, the vehicle's onboard communication unit will parse it and extract key fields such as the risk type, precise location, quantified risk level, confidence level derived from the ST-TFM model, and suggested countermeasures, for further processing by the onboard system.

[0118] Step A72: Present the warning command in a visual or voice format through the vehicle's in-vehicle human-machine interface (HMI) system; Specifically, after the in-vehicle system completes the parsing of the warning command, the in-vehicle human-machine interface (HMI) system will provide the driver with clear, non-intrusive risk warnings in one or more forms based on the parsed information. In one feasible implementation, the system can visualize the risk area on the corresponding location on the in-vehicle navigation map using highlighted icons or color rendering (e.g., red represents high risk); simultaneously, through the in-vehicle audio system, the system will clearly announce the specific risk content to the driver in a voice broadcast format, such as "500 meters ahead, road surface icy risk, high confidence level, it is recommended to keep the speed below 40 km / h." Through this multimodal warning method, the driver can be aware of and understand the potential danger ahead with sufficient reaction time, thereby taking appropriate driving actions to effectively avoid risks. Furthermore, only vehicles about to pass through dangerous or abnormal points will receive the relevant warning information, avoiding interference with unrelated vehicles.

[0119] Step A73: Provide standardized services to third-party application systems through the application programming interface (API) to the dynamic road status data calculated by the cloud platform.

[0120] Specifically, in addition to providing real-time warnings to drivers, the dynamic road condition data with high confidence, calculated by the ST-TFM model on the cloud platform, will also be provided as a standardized data product to third-party application systems. The cloud platform will authorize the provision of real-time road friction coefficient maps, road damage index heat maps, and other data to traffic management departments, road maintenance departments, and autonomous driving systems through secure and standard application programming interfaces (APIs). For example, traffic management departments can use this data for dynamic traffic flow management and intelligent speed limits; road maintenance departments can use road damage data for precise and proactive maintenance planning, thereby improving management efficiency and reducing costs; and autonomous driving systems can directly use these key physical environmental parameters as important inputs to their decision-making and planning modules to improve their safety and reliability under complex and extreme conditions. Through this process, the present invention achieves multi-faceted empowerment of perception data, maximizing its application value.

[0121] Reference Figure 3 This application also provides an intelligent driving early warning system based on vehicle-road-cloud collaborative perception. This system belongs to the same concept as the aforementioned method embodiment, and its specific implementation process is detailed in the method embodiment, and will not be repeated here. The system includes: The data collaborative acquisition module is used to collaboratively collect multi-source heterogeneous environmental and road perception data through vehicle-side perception subsystems deployed on multiple vehicles and perception subsystems deployed at key road nodes. The cloud-based event feature extraction module, deployed in the cloud, uses the E-VESA (Cloud Event Feature Analysis) algorithm to process the collected high-frequency raw data locally in real time, extracting it into structured, lightweight event feature information. The cloud fusion and state calculation module is used to receive and aggregate data uploaded from vehicle terminals, roadside, and third-party platforms. It uses ST-TFM (Spatiotemporal Confidence Fusion Model) to process the aggregated multi-source data. By dynamically calculating the confidence score for each data point and using it as a weight for weighted fusion, it can accurately calculate key state parameters such as road friction coefficient and road damage index. The forward-looking risk assessment module is used to overlay key status parameters calculated in the cloud onto a high-precision geographic information layer in real time, generate and continuously update a dynamic road condition map covering the entire area, and query and evaluate the quantitative risk level of the target vehicle's driving path presented on the dynamic map based on the real-time location and planned route of the target vehicle. The early warning decision generation module is used to determine in real time whether the risk level of the road segment ahead, output by the forward-looking risk assessment module, exceeds the preset adaptive safety threshold. If it exceeds the threshold, the early warning decision mechanism is immediately triggered to generate a structured early warning instruction containing the specific risk type, precise location, and countermeasures. The early warning issuance and service module is used to issue generated early warning commands to target vehicles through vehicle-to-everything (V2X) communication and provide risk alerts through the in-vehicle human-machine interface. At the same time, this module is also used to provide data support to third-party applications such as traffic management, road maintenance and autonomous driving systems through application programming interfaces (APIs) as standardized services for the calculated high-quality dynamic road condition data.

[0122] Reference Figure 7 and Figure 8 , Figure 7 and Figure 8 This is an example diagram of long-term data statistical analysis of bumpy and slippery events performed by the cloud platform in this embodiment.

[0123] Figure 7 The section on "Weekly Changes in the Number of Bump Events" demonstrates how the cloud platform analyzes and statistically analyzes massive amounts of raw vehicle-side IMU data using the E-VESA algorithm logic in step A3 to determine the total number of bump events each week. As shown in the figure, the system can quantitatively reflect the changing trend of road damage over time, reaching a peak (21,897 times) around September 7, 2025. This clearly demonstrates the system's ability to macroscopically monitor and quantitatively statistically analyze road physical damage.

[0124] Figure 8 The section on "Weekly Changes in the Number of Slippery Events" demonstrates how the cloud platform, through the ST-TFM model in step A4, integrates vehicle-side data (such as CAN bus data), roadside weather station data, and high-precision meteorological data to calculate the total number of slippery events each week. As shown in the figure, the system can sensitively capture fluctuations in road surface slipperiness risk caused by factors such as weather changes, for example, reaching a peak (26 times) around August 31, 2025.

[0125] By adopting the above technical solution, this invention can not only identify single risks, but also perform long-term data statistics and trend analysis. These statistical results (such as...) Figure 7 As shown) can be used as the prior probability of the ST-TFM model in step A4 (such as... It provides historical data support and can serve as a standardized data service in step A7, providing road maintenance departments with quantitative decision-making basis on "where" and "when" to carry out maintenance, further highlighting the beneficial effects and practical application value of the present invention.

[0126] This application also provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the intelligent driving early warning method based on vehicle-road-cloud collaborative perception described in any of the foregoing embodiments.

[0127] This application also provides an electronic device, including a processor and a memory; the memory is used to store computer program instructions; the processor is used to execute the computer program instructions stored in the memory to implement the intelligent driving early warning method based on vehicle-road-cloud collaborative perception as described in any of the foregoing embodiments.

[0128] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor or other programmable logic device, whose function is to execute instructions to implement or perform the methods and steps disclosed in this invention.

[0129] In embodiments of the present invention, the memory may include volatile memory (such as random access memory, RAM) and non-volatile memory (such as read-only memory, ROM), or a combination of both. The memory is used to store program code for implementing the above methods and data required for method execution. The storage medium may be any general-purpose or special-purpose computer capable of accessing the data, such as a USB flash drive, portable hard drive, magnetic disk, or optical disk.

[0130] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments, including components, without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. An electronic map update method based on vehicle-road-cloud collaborative perception, characterized in that, Includes the following steps: Step S1: Collect vehicle data and environmental data of the target vehicle within the target area; the vehicle data includes acceleration data; Step S2: Collect meteorological data for the target area; Step S3: Obtain event feature information from the acceleration data using the E-VESA event feature extraction method; Step S4: Based on the event feature information, the environmental data, and the meteorological data, obtain road condition information using the ST-TFM method; Step S5: The road status information is overlaid onto the geographic information layer of the target area in real time to achieve real-time updates of the electronic map within the target area.

2. The electronic map updating method according to claim 1, characterized in that, The E-VESA event feature extraction method described in step S3 includes the following steps: Step S31: Remove noise from all the acceleration data to obtain denoised acceleration data; Step S32: Perform continuous wavelet transform (CWT) on the acceleration data to obtain the time-frequency matrix; Step S33: Obtain the center distance of the time-frequency matrix using the two-dimensional moment invariant method; Step S34: Obtain morphological features based on the center distance; Step S35: The morphological features are processed through a neural network classifier to obtain the category probability. If the category probability is greater than the confidence threshold, structured road event feature information is generated. The road event feature information includes the road event type, occurrence probability, severity level, precise timestamp, GPS coordinates, and associated raw CAN data.

3. The electronic map updating method according to claim 2, characterized in that, The continuous wavelet transform (CWT) mentioned in step S32 is achieved through: Obtain the time-frequency matrix ,in, For scale parameters; For position parameters; In the vertical direction The first moment The acceleration data mentioned in the paragraph; It is the conjugate of the mother wavelet function.

4. The electronic map updating method according to claim 2, characterized in that, The two-dimensional moment invariant method described in step S33 is achieved through: Obtain center distance ,in, for The order of moments in the direction, for The order of moments in the direction, is the centroid of the time-frequency matrix.

5. The electronic map updating method according to claim 1, characterized in that, The ST-TFM method described in step S4 includes the following steps: Step S41: Perform local event likelihood inference based on the event feature information to obtain the initial likelihood. Step S42: Based on the initial likelihood and the environmental data, obtain a reputation score through a dynamic source reputation calibration mechanism; Step S43: Based on the reputation score and the meteorological data, road status information is obtained through global confidence fusion; the road status information includes road status and corresponding probability.

6. The electronic map updating method according to claim 5, characterized in that, The initial likelihood mentioned in step S41 Through Obtain, among which, Indicates the event to be detected. for Observe the data, for Bus observation data, Let be the likelihood probability of the sensor. Let be the conditional probability of the vehicle's state. Let be the prior probability of the event.

7. The electronic map updating method according to claim 5, characterized in that, The dynamic source reputation calibration mechanism described in step S42 is achieved through... Obtain the target vehicle Reputation rating at time t ;in, The learning rate; For the target vehicle Vehicle data; The environmental data; This is a matching function used to obtain the consistency between the initial likelihood and the environmental feature information.

8. A driving early warning method based on vehicle-road-cloud collaborative perception, characterized in that, Includes the following steps: Step S01: Based on the real-time location and planned route of the target vehicle to be warned, query the road status information in the electronic map as a structured risk parameter sequence; Step S02: Compare the structured risk parameter sequence with a safety threshold. If the threshold is exceeded, generate an early warning instruction. Step S03: Send the warning command to the target vehicle to be warned, and complete the intelligent driving warning; The electronic map is updated in real time using any one of the electronic map update methods based on vehicle-road-cloud collaborative perception as described in claims 1 to 7.

9. An intelligent driving early warning system based on vehicle-road-cloud collaborative perception, characterized in that, It includes a data collaborative acquisition module, a cloud event feature extraction module, a cloud fusion and status calculation module, a forward-looking risk assessment module, an early warning decision generation module, and an early warning issuance module; The data collaborative acquisition module is used to collect vehicle data and environmental data of target vehicles within the target area; the vehicle data includes acceleration data. The cloud-based event feature extraction module is used to obtain meteorological data for the target area; It is also used to extract event feature information from all acceleration data using the E-VESA event feature extraction method; The cloud fusion and state calculation module is used to obtain road state information based on the event feature information, the environmental data, and the meteorological data using the ST-TFM method. The forward-looking risk assessment module is used to overlay the road condition information onto the geographic information layer of the target area in real time, so as to realize the real-time updating of the electronic map in the target area; and to query the road condition information in the electronic map as a structured risk parameter sequence based on the real-time location and planned route of the target vehicle. The early warning decision generation module is used to compare the structured risk parameter sequence with a safety threshold, and if the threshold is exceeded, an early warning instruction is generated. The warning delivery module is used to send the warning command to the target vehicle to complete the intelligent driving warning.

Images