Regional request type vehicle multi-modal perimeter perception and risk warning method, device, equipment, medium, vehicle and system

Abstract

The application provides a regional request type vehicle multi-modal perimeter perception and risk early warning method, device, electronic equipment, computer readable storage medium, vehicle and system. By receiving a perception request instruction containing specified geographic fence parameters and a digital certificate, a target perception mode identifier and a minimum response level certificate identifier are extracted. When the certificate authority meets the response level, a vehicle-mounted physical sensor matrix is started, multi-modal sensor stream data is obtained, a target feature vector is extracted locally, and a desensitized perception data packet is generated. The cloud aggregates multiple data packets in time and space to generate a regional situation time sequence image, and generates and distributes a hierarchical early warning event sequence in combination with a dedicated risk judgment rule library.

Description

Technical Field

[0001] This application relates to the fields of vehicle perception, vehicle-to-everything (V2X) communication, regional security early warning and data anonymization processing, and more specifically, to a regional request-based vehicle multimodal perimeter perception and risk early warning method, device, electronic device, computer-readable storage medium, vehicle and system. Background Technology

[0002] Oil depots, chemical warehouses, substations, research institutes, and major transportation hubs typically have high safety monitoring needs. The safety status of these areas depends not only on the operation of the equipment within the area itself but also on changes in the surrounding roads, perimeter buffer zones, and environment. For example, environmental signals such as thermal anomalies, gas diffusion, abnormal vibrations, or changes in local traffic conditions may first manifest at the outer edges of the area or on adjacent roads. Therefore, these areas require data processing mechanisms that can cover the surrounding space, reflect environmental change processes, and respond according to risk levels.

[0003] In existing technologies, a common approach is to deploy camera equipment, environmental data acquisition devices, or integrated monitoring stations at fixed locations and transmit the data collected by these fixed devices to a monitoring platform for centralized processing. This approach typically involves first installing fixed acquisition terminals at area boundaries or key passageways; then, these fixed acquisition terminals collect data such as images, temperature, gas concentration, or vibration according to a predetermined sampling period; finally, the monitoring platform compares the data returned by the fixed acquisition terminals with preset thresholds and generates corresponding alarm information when a threshold is triggered.

[0004] However, the aforementioned fixed-point monitoring scheme has significant technical shortcomings. Because the location of the data acquisition terminals is relatively fixed after deployment, their sensing range is greatly affected by the installation location, orientation, communication link, and maintenance status, making it difficult to cover the continuous road space at the outer edge of the area. Furthermore, data collected by fixed terminals tends to be singular data sources, failing to accurately reflect the diffusion and changes of risk signals in both time and space. In addition, traditional schemes typically rely on continuous or fixed-period data collection, making it difficult to establish controlled data acquisition links based on the permission level of external requests, the target collection range, and the target sensing type. This results in low adaptability for regional risk identification and an insufficient foundation for the hierarchical distribution of risk warnings. Summary of the Invention

[0005] This application provides a method, device, electronic device, computer-readable storage medium, vehicle, and system for area-request-based multimodal vehicle perimeter perception and risk warning, in order to at least alleviate the technical problems of fixed-point monitoring schemes, such as limited coverage, difficulty in reflecting the risk diffusion process from single-point data sources, and difficulty in establishing controlled perception links according to request permissions and target perception types.

[0006] The technical advantages of the technical solution provided in this application are: This application's regional request-based multimodal perimeter perception and risk warning method for vehicles addresses the technical shortcomings of fixed-point monitoring schemes, such as limited coverage, difficulty in reflecting risk diffusion processes from single-point data sources, and difficulty in establishing controlled perception links according to request permissions and target perception types. By receiving perception request instructions containing specified geofence parameters and digital certificates, and extracting the target perception modality identifier and minimum response level certificate identifier from the perception request instructions, the spatial scope, permission basis, and perception type of perception processing are structurally defined at the request entry point. Compared to traditional fixed-period, fixed-point data collection methods, this approach allows regional risk monitoring to move beyond relying solely on existing fixed installation locations. Instead, it enables the establishment of controlled perception entry points around the requested geographical area, thereby mitigating the problems of insufficient coverage at fixed points and unclear collection targets.

[0007] By verifying the permissions of digital certificates and activating the onboard physical sensor matrix corresponding to the target perception modality identifier on the target vehicle when the digital certificate's permissions meet the minimum response level certificate identifier, this application establishes a direct link between the permission verification result and the sensor activation action. Traditional fixed monitoring schemes often struggle to dynamically constrain the acquisition range based on the permission level of external requests and the target perception type. However, this application only activates the corresponding onboard physical sensor matrix when the digital certificate permissions meet the response level, and the activation target corresponds to the target perception modality identifier. This reduces the participation of irrelevant perception modalities, lowers the probability of non-target data being collected and transmitted, and makes the perception link more controllable.

[0008] This application extracts target feature vectors from multimodal sensor stream data at the local processing node of the target vehicle, generates desensitized data packets based on the target feature vectors, and then transmits the desensitized data packets to the cloud aggregation node. This approach moves the feature extraction of multimodal sensor stream data to the vehicle's local processing node. Compared to traditional monitoring schemes that directly transmit collected data back to the platform for centralized processing, this application transmits desensitized data packets based on target feature vectors. This reduces the data burden caused by cross-node transmission of raw multimodal data and makes the data received by the cloud aggregation node more suitable for subsequent spatiotemporal aggregation processing.

[0009] By performing spatiotemporal aggregation processing on multiple desensitized data packets from areas corresponding to specified geofence parameters at cloud aggregation nodes, a regional situation time-series profile is generated. This application incorporates desensitized data packets formed by multiple vehicle local processing nodes into the same spatial range and time series caliber. Traditional fixed-point monitoring schemes are easily affected by the installation location and collection status of individual fixed terminals, making it difficult to reflect continuous changes in environmental signals at the periphery of the area. However, this application, through spatiotemporal aggregation processing of multiple desensitized data packets, enables cloud aggregation nodes to form a regional situation time-series profile, thereby providing a more sufficient data foundation for judging whether risk signals have regional and temporal changes.

[0010] By generating corresponding tiered early warning event sequences based on regional situational time-series profiles and a pre-configured proprietary risk assessment rule base corresponding to specified geofence parameters, and distributing these sequences to preset receiving endpoints, this application establishes a continuous processing link between regional situational data, corresponding regional rules, and early warning distribution targets. Compared to traditional fixed-threshold alarm methods, this application can assess regional situational time-series profiles based on proprietary risk assessment rule bases corresponding to specified geofence parameters, organize the assessment results into tiered early warning event sequences, and distribute them to preset receiving endpoints. This ensures that risk warnings are no longer limited to single alarm triggers, but rather form early warning distribution results that match the regional scope, risk level, and receiving targets, thereby improving the adaptability and response hierarchy of regional security early warning processing. Attached Figure Description

[0011] Figure 1 This application provides a scenario for a region-request-based multimodal vehicle perimeter perception and risk warning method. Figure 2 This application provides an embodiment of a region-request-based multimodal vehicle perimeter perception and risk warning method. Figure 3 This application provides an embodiment of a region-request vehicle multimodal perimeter perception and risk warning device. Figure 4 An electronic device is described in an embodiment of this application; Figure 5 This application provides an embodiment of a computer-readable storage medium. Figure 6 This application describes a vehicle as an example of its embodiments. Figure 7 This application provides a method and system for regional request-based vehicle multimodal perimeter perception and risk warning. Detailed Implementation

[0012] like Figure 1 The image shows a scenario of a region-request-based multimodal vehicle perimeter perception and risk warning method according to an embodiment of this application; as shown... Figure 2The illustration shows a method for regional request-based multimodal perimeter perception and risk warning of vehicles according to an embodiment of this application. The method includes: receiving a perception request instruction containing specified geofence parameters and a digital certificate; extracting a target perception modality identifier and a minimum response level certificate identifier from the perception request instruction; performing permission verification on the digital certificate, and when it is determined that the permissions of the digital certificate meet the minimum response level certificate identifier, activating an onboard physical sensor matrix corresponding to the target perception modality identifier on the target vehicle to acquire multimodal sensor stream data; extracting a target feature vector from the multimodal sensor stream data in the local processing node of the target vehicle, generating a desensitized data packet based on the target feature vector, and transmitting the desensitized data packet to a cloud aggregation node; performing spatiotemporal aggregation processing on multiple desensitized data packets from the area corresponding to the specified geofence parameters at the cloud aggregation node to generate a regional situation time-series profile; generating a corresponding graded warning event sequence based on the regional situation time-series profile and a pre-configured exclusive risk judgment rule base corresponding to the specified geofence parameters, and distributing the graded warning event sequence to a preset receiving endpoint.

[0013] Optionally, before receiving the sensing request instruction, the method further includes: obtaining the center point hash coordinates and preset coverage radius of the initial monitoring area, and obtaining a pre-configured normal sensing interval parameter, wherein the normal sensing interval parameter is used to set the normal sensing interval of the sensing request instruction; establishing a mobile tracking fence based on the center point hash coordinates and the preset coverage radius, and generating the specified geofence parameter; and generating the sensing request instruction based on the specified geofence parameter, the normal sensing interval parameter, and the digital certificate.

[0014] Preferably, the specific implementation process for obtaining the center point hash coordinates and preset coverage radius of the initial monitoring area before receiving the perception request instruction is as follows: First, read the spatial boundary configuration record to be performed for perimeter perception. The spatial boundary configuration record is not simply a map display range, but a spatial index basis used to limit subsequent vehicle matching, vehicle physical sensor matrix preheating, perception request instruction issuance, and the attribution relationship of desensitized perception data packets. The spatial boundary configuration record can be composed of a fence identifier, a fence type field, an original center coordinate field, a boundary radius field, a buffer radius field, an authorization level field, and a moving target location source field. The fence identifier is used to distinguish different perception ranges in the subsequently specified geofence parameters. The fence type field is used to distinguish long-term fixed areas, time-active areas, event-extended areas, or mobile tracking areas. The original center coordinate field is used to provide the geometric positioning basis of the initial monitoring area. The boundary radius field is used to limit the main coverage range of the initial monitoring area. The buffer radius field is used to form a preliminary matching range before the target vehicle enters the core range. The authorization level field is used to form an authorization association with the digital certificate when generating the perception request instruction. The moving target location source field is used to provide the update source of the real-time location of the moving target when the fence type field represents the mobile tracking area. After reading the spatial boundary configuration record, the original center coordinate field is first normalized to convert it to a road segment coordinate system consistent with the vehicle location reporting standard. Then, the road segment coordinate system is hashed to obtain the center point hash coordinates. In this application, the center point hash coordinates are a spatial index field that converts the geometric center point of the initial monitoring area into a location that cannot be directly reconstructed as a precise physical location. This spatial index field is not used as display coordinates but participates in the establishment of mobile tracking fences, anonymous vehicle location comparison, and encapsulation of specified geofence parameters, enabling the target vehicle to determine whether it is within the corresponding sensing range based on the anonymized spatial relationship. After determining the center point hash coordinates, the boundary radius field is read, and the boundary radius field is validated according to the fence type field to form the preset coverage radius. In this application, the preset coverage radius is a distance boundary parameter that, together with the hash coordinates of the center point, defines the spatial range of the target. The preset coverage radius can be configured according to the road distribution at the outer edge of the initial monitoring area, the available sensing distance of the vehicle-mounted physical sensor matrix, the spatial location of the risk signal outside the perimeter, and the update speed of the moving target's real-time position. For a mobile tracking area, the preset coverage radius does not represent the static boundary outside a fixed building, but rather the sensing radius surrounding the real-time position of the moving target; for a long-term fixed area, the preset coverage radius represents the normal sensing range surrounding the hash coordinates of the center point of the initial monitoring area.Therefore, the center point hash coordinates and the preset coverage radius jointly participate in the establishment of the mobile tracking fence, and enable the perception request command to use an anonymous spatial range as the entry point, rather than the publicly available precise location coordinates, thereby reducing the exposure range of the original location coordinates in the request link.

[0015] Preferably, in the specific technical implementation of obtaining a pre-configured normal perception interval parameter and setting the normal perception interval of the perception request instruction before receiving the perception request instruction, the perception frequency configuration record corresponding to the fence identifier is first read. The perception frequency configuration record is used to describe the request sending rhythm of different fence types before the danger level identifier is triggered. The perception frequency configuration record may include a fence identifier field, a fence type field, a normal perception interval field, an alarm state perception interval field, a perception modality constraint field, and a vehicle matching refresh field. The normal perception interval parameter is formed after the normal perception interval field is read. The alarm state perception interval field is used to adjust the sending rhythm of the perception request instruction after the generation of the graded warning event sequence. The perception modality constraint field is used to limit the target perception modality identifier that the perception request instruction is allowed to carry in the future. The vehicle matching refresh field is used to limit the refresh cycle of vehicle position comparison after the center point hash coordinates are updated. In this application, the normal perception interval parameter is a time field that can be carried by the perception request command, which converts the time sampling density required for the initial monitoring area before the hazard level indicator is triggered. The normal perception interval parameter is not simply the alert time or the alarm response time, but rather a request rhythm parameter used to control the time interval at which the perception request command enters the target vehicle matching process before the dynamic event fence extension processing mechanism and independent direct reporting transmission link are triggered. When configuring the normal perception interval parameter, the basic interval caliber can be determined first based on the fence type field; for example, a minute-level interval can be used for long-term fixed areas, and a second-level interval can be used for mobile tracking areas. Then, based on the perception mode energy consumption parameters corresponding to the vehicle physical sensor matrix, the transmission capacity parameters of the desensitized perception data packets, the aggregation waiting length of the cloud aggregation node within the time slice, and the requirements of the regional situation time series profile for continuous time windows, the basic interval caliber is adjusted to obtain the normal perception interval parameter that matches the specified geofence parameter. The sensing mode energy consumption parameter represents the energy consumption burden of the on-board physical sensor matrix corresponding to the target sensing mode identifier within a unit sampling period. This parameter participates in the adjustment of the basic interval caliber, ensuring that the normal sensing interval parameter matches the continuous operating capability of the on-board physical sensor matrix. The desensitized sensing data packet transmission capacity parameter represents the amount of desensitized sensing data packets that the target vehicle can transmit within one normal sensing interval. This parameter also participates in the adjustment of the basic interval caliber, ensuring that the normal sensing interval parameter matches the time-slice aggregation rhythm of the cloud aggregation node.Before the normal perception interval parameter is written into the perception request command, it needs to be checked for consistency with the preset coverage radius: when the preset coverage radius covers a large number of road segments, the normal perception interval parameter can use a longer interval to reduce irrelevant duplicate collections; when the mobile tracking fence updates its real-time position quickly following the moving target, the normal perception interval parameter can use a shorter interval to ensure a good correspondence between the center point hash coordinates, the vehicle matching range, and the issuance time of the perception request command. After the above processing, the normal perception interval parameter participates in setting the normal perception interval of the perception request command, so that after the target vehicle enters the mobile tracking fence, it can perform the perception task corresponding to the target perception modality identifier according to the normal perception interval, instead of the target vehicle choosing an inconsistent sampling time caliber on its own.

[0016] Preferably, the specific implementation process for establishing a mobile tracking fence and generating the specified geofence parameters based on the center point hash coordinates and the preset coverage radius is as follows: First, a mobile target spatial anchor point carrying record is established. The center point hash coordinates are written into the center position field of the mobile target spatial anchor point carrying record, and the preset coverage radius is written into the range field of the mobile target spatial anchor point carrying record. Then, the mobile target spatial anchor point carrying record after writing the center position field and range field is indexed and encapsulated to form a mobile target spatial anchor point record that can be continuously updated. In this application, the mobile target spatial anchor point record binds the anonymous spatial position and spatial coverage range corresponding to the mobile target at the current time to the same reading caliber, so that subsequent vehicle position comparison does not need to directly read the precise physical coordinates of the mobile target. Subsequently, the anonymous position field and single session-level random communication token of the candidate vehicle are read according to the vehicle position reporting caliber, and the anonymous position field and the center point hash coordinates are placed under the same road segment index caliber for adjacency calculation to form a candidate vehicle fence relationship record. The candidate vehicle fence relationship record includes a single-session-level random communication token for the candidate vehicle, a hash identifier for the road segment where the candidate vehicle is located, the segment distance relationship between the candidate vehicle and the hash coordinates of the center point, a matching flag indicating whether the candidate vehicle is within a preset coverage radius, and a preheating flag indicating whether the candidate vehicle is within a buffer radius. This candidate vehicle fence relationship record is further used to establish the mobile tracking fence: when the matching flag in the candidate vehicle fence relationship record indicates that the candidate vehicle is within the preset coverage radius, the candidate vehicle fence relationship record is included in the core response record of the mobile tracking fence; when the preheating flag in the candidate vehicle fence relationship record indicates that the candidate vehicle is within the buffer radius but has not yet entered the preset coverage radius, the candidate vehicle fence relationship record is included in the preheating response record of the mobile tracking fence. The core response record is used to determine target vehicles that have entered the preset coverage radius when a subsequent perception request command is issued, and the preheating response record is used to determine candidate vehicles approaching the preset coverage radius when the onboard physical sensor matrix is ​​preheated. Both the core response record and the preheating response record originate from the candidate vehicle fence relationship record and jointly participate in the updating of the mobile tracking fence. In this application, the mobile tracking fence is an updatable spatial constraint object formed by the center point hash coordinates, the preset coverage radius, the buffer radius, and the candidate vehicle fence relationship record. The mobile tracking fence is not a static map graphic, but rather, after the real-time position of the moving target changes, the center point hash coordinates are regenerated and the candidate vehicle fence relationship record is recalculated, so that the matching status of the target vehicle can follow the real-time position update of the moving target.After establishing the mobile tracking fence, the fence identifier, fence type field, center point hash coordinates, preset coverage radius, buffer radius, authorization level field, and vehicle matching refresh field are read from the mobile target spatial anchor point record, core response record, and preheating response record. These fields are then aggregated and encapsulated to generate the specified geofence parameters. These specified geofence parameters are subsequently read as the spatial range field of the perception request command. Their technical function is to enable the perception request command to limit whether the target vehicle belongs to the current request range, and to enable the cloud aggregation node to identify whether multiple desensitized perception data packets belong to the same fence area based on the same specified geofence parameters.

[0017] Preferably, in the specific technical implementation of generating the perception request instruction based on the specified geofence parameters, the normal perception interval parameters, and the digital certificate, the following steps are taken: First, the fence identifier, fence type field, center point hash coordinates, preset coverage radius, buffer radius, and authorization level fields in the specified geofence parameters are read, and the normal perception interval field in the normal perception interval parameters is read to establish a perception request carrying record. In this application, the perception request carrying record incorporates spatial constraints, temporal constraints, and permission constraints into the same data carrying object before the perception request instruction is generated, so that when the perception request instruction is sent to the target vehicle, it can simultaneously carry vehicle matching basis, sampling time basis, and certificate verification basis. Subsequently, the certificate issuance level label, certificate validity period field, and request signature field in the digital certificate are read, and the certificate issuance level label is checked against the authorization level field in the specified geofence parameters to form a request permission verification record. The request permission verification record is used to determine whether the digital certificate can initiate a perception request instruction corresponding to the specified geofence parameter. When the request permission verification record indicates that the digital certificate matches the authorization level field, the certificate issuance level label, certificate validity period field, and request signature field are written into the perception request bearer record. When the request permission verification record indicates that the digital certificate does not match the authorization level field, the generation of the perception request instruction is stopped to prevent perception request bearer records with mismatched permissions from entering the vehicle matching process. After completing the permission verification, a perception frequency field is configured for the perception request bearer record based on the fence type field in the perception request bearer record and the normal perception interval field corresponding to the normal perception interval parameter. The target perception modality identifier is read from the perception modality constraint field based on the fence type field in the perception request bearer record and written into the perception request bearer record. The perception frequency field is used to carry the request sending rhythm corresponding to the normal perception interval field. The perception frequency field is subsequently sent to the target vehicle along with the perception request instruction, enabling the target vehicle to acquire multimodal sensor stream data according to the request sending rhythm corresponding to the normal perception interval field. The target perception modality identifier is subsequently used directly to activate the onboard physical sensor matrix on the target vehicle. The normal perception interval parameter is subsequently used directly to set the sampling time caliber on the target vehicle before triggering the hazard level indicator. The digital certificate is subsequently used directly on the target vehicle to perform permission verification. Finally, the perception request bearer record, after writing the specified geofence parameter field, normal perception interval field, target perception modality identifier, certificate issuance level label, certificate validity period field, and request signature field, undergoes instruction encapsulation processing to generate the perception request instruction.In this application, the perception request instruction is structured request data that can be parsed by the target vehicle. The perception request instruction solidifies the technical relationship between the specified geofence parameters, the normal perception interval parameters, and the digital certificate into the same transmission unit, so that after receiving the perception request instruction, the target vehicle can first perform permission verification based on the digital certificate, then confirm whether it is within the corresponding range based on the specified geofence parameters, and then perform corresponding multimodal sensor stream data acquisition and processing based on the target perception modality identifier and the normal perception interval parameters.

[0018] Optionally, the step of verifying the permissions of the digital certificate and, upon determining that the permissions of the digital certificate meet the minimum response level certificate identifier, activating the on-board physical sensor matrix corresponding to the target perception modality identifier on the target vehicle to acquire multimodal sensor stream data includes: parsing the certificate issuance level tag of the digital certificate; comparing the certificate issuance level tag with the minimum response level certificate identifier pre-stored locally on the target vehicle to determine whether the permission priority of the certificate issuance level tag reaches the permission priority of the minimum response level certificate identifier; if the permission priority of the certificate issuance level tag reaches the permission priority of the minimum response level certificate identifier, then removing the perception restriction interception of the on-board operating system of the target vehicle; after removing the perception restriction interception, activating the corresponding on-board physical sensor matrix according to the target perception modality identifier to acquire the multimodal sensor stream data.

[0019] Preferably, the specific implementation process of parsing the certificate issuance level label of the digital certificate is as follows: After the target vehicle receives the perception request instruction, the digital certificate bearer field is first read from the perception request instruction, and the certificate structure parsing process is performed on the digital certificate bearer field to obtain the digital certificate parsing record. The digital certificate parsing record includes a certificate body field, a certificate issuance field, a certificate validity period field, a certificate purpose field, a request signature field, and a certificate issuance level label field. The certificate body field is used to characterize the source of the perception request instruction, the certificate issuance field is used to characterize the issuance chain of the digital certificate, the certificate validity period field is used to limit the time range for the digital certificate to participate in the permission verification, the certificate purpose field is used to limit whether the digital certificate can be used for area-request-type vehicle multimodal perimeter perception requests, the request signature field is used to perform integrity verification on the specified geofence parameters, target perception modality identifier, and minimum response level certificate identifier in the perception request instruction, and the certificate issuance level label field is used to carry the certificate issuance level label. In specific implementation, the request payload field is first read from the perception request instruction. The request payload field includes specified geofence parameters, target perception modality identifier, and minimum response level certificate identifier. A digest consistency check is performed on the request signature field and the request payload field to form a request payload check record. This request payload check record indicates whether the specified geofence parameters, target perception modality identifier, and minimum response level certificate identifier in the request payload field are consistent with the signature content corresponding to the request signature field. When the request payload check record indicates that there are no field misalignments or substitutions in the request payload field, the certificate issuance field and the certificate purpose field are then read to form a certificate purpose check record. This certificate purpose check record determines whether the digital certificate belongs to a certificate type capable of initiating area-request vehicle multimodal perimeter perception requests. When the certificate purpose check record indicates that the digital certificate belongs to a certificate type that can be used for area-request vehicle multimodal perimeter perception requests, the certificate issuance level label is extracted from the certificate issuance level label field. In this application, the certificate issuance level label converts the authorization level of the digital certificate in the issuance chain corresponding to the certificate issuance field into an authorization index field that can be compared locally by the target vehicle. This authorization index field is then entered into the subsequent certificate level comparison record in the form of the certificate issuance level label. The certificate issuance level label does not directly activate the vehicle-mounted physical sensor matrix; instead, it serves as an input field for subsequent comparison with the lowest response level certificate identifier. Thus, the digital certificate parsing record, request payload verification record, and certificate purpose verification record sequentially participate in the extraction of the certificate issuance level label, ensuring that subsequent authorization verification does not rely on the textual description of the perception request instruction but rather on the certificate issuance level label after field parsing and integrity verification.

[0020] Preferably, in the specific technical implementation of comparing the certificate issuance level label with the lowest response level certificate identifier pre-stored locally in the target vehicle to determine whether the permission priority of the certificate issuance level label reaches the permission priority of the lowest response level certificate identifier, the permission priority sequence table in the local permission configuration area of ​​the target vehicle is first read, and the lowest response level mapping record corresponding to the lowest response level certificate identifier is read from the permission priority sequence table. In this application, the permission priority sequence table is certificate level sorting data pre-written in the local permission configuration area of ​​the target vehicle. The permission priority sequence table maps different certificate issuance level labels to comparable permission priorities, and the permission priorities are used to express the response boundaries of the target vehicle to different perception request commands. The target vehicle's local permission configuration area can have a permission priority sequence table and a minimum response level certificate identifier written into it during the vehicle initialization configuration phase, or it can have these elements rewritten after the target vehicle receives the permission configuration update data packet. After the permission configuration update data packet is written into the target vehicle's local permission configuration area, the certificate level sorting data in the packet is used to update the permission priority sequence table, and the minimum responsive certificate level field is used to update the minimum response level certificate identifier. In this application, the minimum response level certificate identifier is a boundary identifier in the target vehicle's local permission configuration area used to define the minimum responsive certificate level. The written minimum response level certificate identifier establishes a fixed mapping relationship with the permission priority sequence table, enabling the target vehicle to use the same comparison standard when receiving different digital certificates. After reading the minimum response level mapping record, the certificate issuance level label is written into the certificate level comparison record, and the permission priority corresponding to the certificate issuance level label in the comparison record is queried according to the permission priority sequence table to form a certificate issuance level permission record. The certificate level comparison record is used to store the certificate issuance level label after it has been processed by the request payload verification record and the certificate purpose verification record. This record continues to participate in the generation of the certificate issuance level permission record, ensuring that the permission priority of the certificate issuance level label has a traceable field source. Subsequently, the permission priority in the certificate issuance level permission record is compared sequentially with the permission priority in the lowest response level mapping record to form a certificate permission comparison record.The certificate permission comparison record is used to indicate whether the permission priority of the certificate issuance level label reaches the permission priority of the lowest response level certificate identifier. When the permission priority of the certificate issuance level label is lower than the permission priority of the lowest response level certificate identifier, a rejection response flag is written to the certificate permission comparison record. The rejection response flag is used to block the perception request command from entering the subsequent perception restriction interception removal process. When the permission priority of the certificate issuance level label reaches the permission priority of the lowest response level certificate identifier, a responsive flag is written to the certificate permission comparison record. The responsive flag serves as the triggering basis for subsequently removing the perception restriction interception of the target vehicle's in-vehicle operating system. Through this processing, the lowest response level certificate identifier is no longer an abstract text threshold, but a local comparison boundary continuously associated with the permission priority sequence table, the lowest response level mapping record, and the certificate permission comparison record.

[0021] Preferably, if the permission priority of the certificate issuance level label reaches the permission priority of the lowest response level certificate identifier, the specific implementation process for lifting the perception restriction interception of the target vehicle's onboard operating system is as follows: After the certificate permission comparison record is written with a responsive flag, the specified geofence parameter, target perception modality identifier, and normal perception interval parameter in the perception request instruction are read first, and the specified geofence parameter, the target perception modality identifier, the normal perception interval parameter, and the responsive flag are written into the perception authorization bearer record. In this application, the perception authorization bearer record incorporates the responsive flag, spatial range field, perception modality field, and sampling time field into the same reading caliber, so that the subsequent lifting of the perception restriction interception is not solely based on the digital certificate, but is simultaneously constrained by the specified geofence parameter, the target perception modality identifier, and the normal perception interval parameter. The spatial range field is provided by the specified geofence parameter, and is used to limit whether the target vehicle belongs to the geographical range of the current perception request instruction; the perception modality field is provided by the target perception modality identifier, and is used to limit the perception modalities that are allowed to be called subsequently; the sampling time field is provided by the normal perception interval parameter, and is used to limit the time caliber for subsequent acquisition of multimodal sensor stream data. Subsequently, the perception restriction and interception status record of the vehicle's onboard operating system is read. The perception restriction and interception status record includes a perception interface interception flag, a sensor call interception flag, a perception modality whitelist field, and a data upload interception flag. In this application, the perception restriction and interception of the onboard operating system is a local access control state that blocks the call entry of the onboard physical sensor matrix, the sampling entry of the multimodal sensor stream data, and the transmission entry of the desensitized perception data packet before the target vehicle completes the permission verification; the perception restriction and interception of the onboard operating system does not physically disconnect the onboard physical sensor matrix, but retains the interception flag in the local call link, so that unauthorized perception request instructions cannot trigger the perception modality corresponding to the target perception modality identifier. After reading the perception restriction interception status record, based on the responsive flag in the perception authorization bearer record, the perception interface interception flag is rewritten from the intercepted state to the callable state, and the target perception modality identifier is written into the perception modality whitelist field to form a perception restriction release record. The perception restriction release record is then used to rewrite the sensor call interception flag and the data upload interception flag, ensuring that the sensor call interception flag only enters the callable state for target perception modality identifiers within the perception modality whitelist field, and that the data upload interception flag only enters the transmittable state for desensitized data packets subsequently generated by the perception modality corresponding to the target perception modality identifier.The transmittable state continues to limit the transmission process of subsequent desensitized data packets into the vehicle communication node, ensuring that the transmission range of the desensitized data packets remains consistent with the perception modality whitelist field. Therefore, removing the perception restriction interception of the target vehicle's onboard operating system does not mean opening all sampling channels of the onboard physical sensor matrix. Instead, it binds specified geofence parameters, target perception modality identifiers, and normal perception interval parameters based on the perception authorization bearer record, ensuring that the activation range of the subsequent onboard physical sensor matrix corresponds to the perception request command that has passed the permission verification.

[0022] Preferably, in the specific technical implementation of activating the corresponding vehicle-mounted physical sensor matrix according to the target perception modality identifier after the perception restriction interception is lifted, in order to obtain the multimodal sensor stream data, the perception modality whitelist field in the perception restriction lifting record is first read, and the target perception modality identifier in the perception modality whitelist field is written into the sensor matrix activation record. The sensor matrix activation record is used to transmit the target perception modality identifier that has been released through the perception restriction lifting record to the reading process of the vehicle-mounted physical sensor matrix, so that the vehicle-mounted physical sensor matrix only performs sampling channel positioning based on the target perception modality identifier in the perception modality whitelist field. In this application, the vehicle-mounted physical sensor matrix is ​​a structured mapping record between the perception modes stored locally on the target vehicle and the vehicle-mounted sensor sampling channels, rather than a simple mathematical matrix. The rows of the vehicle-mounted physical sensor matrix can correspond to perception modes such as thermal gradient distribution attributes, gas concentration abrupt change attributes, environmental vibration attributes, electromagnetic intensity change attributes, or vehicle motion direction attributes. The columns of the vehicle-mounted physical sensor matrix can correspond to vehicle image sampling channels, vehicle body temperature sampling channels, in-vehicle gas sampling channels, three-axis acceleration sampling channels, vehicle communication strength sampling channels, or positioning speed sampling channels. The matrix elements at the intersection of rows and columns in the vehicle-mounted physical sensor matrix are used to record the activation relationship between the corresponding perception mode and the corresponding vehicle-mounted sensor sampling channel, the sampling frequency field, the sampling data format field, and the end-side cache field. After reading the vehicle-mounted physical sensor matrix, based on the target perception mode identifier in the sensor matrix activation record, the matrix element corresponding to the target perception mode identifier is located in the vehicle-mounted physical sensor matrix, and the sampling channel identifier, sampling frequency field, and sampling data format field are read from the matrix element to form a sampling channel configuration record. The sampling channel configuration record continues to participate in the multimodal sampling entry configuration. The multimodal sampling entry configuration is used to write the sampling channel identifier, sampling frequency field, and sampling data format field into the corresponding vehicle sensor sampling channel, so that the corresponding vehicle sensor sampling channel performs data sampling according to the same perception request instruction: when the target perception mode identifier corresponds to the thermal gradient distribution attribute, the vehicle image sampling channel and the vehicle body temperature sampling channel are started according to the sampling channel configuration record to form a thermal gradient sampling data stream; when the target perception mode identifier corresponds to the gas concentration change attribute, the vehicle interior gas sampling channel is started according to the sampling channel configuration record to form a gas concentration sampling data stream; when the target perception mode identifier corresponds to the environmental vibration attribute, the three-axis acceleration sampling channel is started according to the sampling channel configuration record to form an environmental vibration sampling data stream.The thermal gradient sampling data stream, the gas concentration sampling data stream, and the environmental vibration sampling data stream are timestamped according to the sampling time caliber corresponding to the normal sensing interval parameter to form a timestamped sampling data stream. The timestamped sampling data stream is then encapsulated into a frame sequence according to the sampling data format field to obtain the multimodal sensing stream data. In this application, the multimodal sensing stream data is a local sensing data sequence formed by sampling data streams corresponding to at least two sensing modes according to the same sensing request command and the same sampling time caliber. The multimodal sensing stream data will not be directly used as cross-node transmission content in the subsequent process, but will instead enter the local processing node of the target vehicle for target feature vector extraction, so that the subsequent desensitized sensing data package can carry the structured results of different sensing modes such as thermal gradient distribution attributes, gas concentration abrupt change attributes, and environmental vibration attributes.

[0023] Optionally, the step of extracting a target feature vector from the multimodal sensor stream data and generating a desensitized data packet based on the target feature vector in the local processing node of the target vehicle includes: performing feature separation and extraction on the multimodal sensor stream data to obtain a numerical sequence reflecting regional thermal gradient distribution attributes, gas concentration abrupt change attributes, and environmental vibration attributes; constructing the target feature vector based on the numerical sequence; performing privacy isolation processing on the hardware identification code of the target vehicle and replacing the hardware identification code with a single-session-level random communication token; mapping the physical location coordinates of the target vehicle to a road segment hash identifier; and constructing the desensitized data packet based on the target feature vector, the single-session-level random communication token, and the road segment hash identifier.

[0024] Preferably, the specific implementation process for feature separation and extraction of the multimodal sensor stream data is as follows: After the local processing node of the target vehicle receives the multimodal sensor stream data, it first reads the sampling channel identifier, sampling time caliber, sampling data format field, and perception modality source marker from the multimodal sensor stream data, and writes the sampling channel identifier, sampling time caliber, sampling data format field, and perception modality source marker into the multimodal stream parsing bearer record. In this application, the multimodal sensor stream data is a local perception data sequence formed locally on the target vehicle by sampling data streams corresponding to at least two perception modalities in the vehicle physical sensor matrix, according to the same perception request command and the same sampling time caliber; the multimodal sensor stream data is not the original perception content to be directly transmitted across nodes, but rather the reading object for subsequent feature separation and extraction at the local processing node of the target vehicle. The multimodal stream parsing record is used to incorporate the sampling channel identifier, sampling time caliber, sampling data format field, and sensing modality source marker in the multimodal sensing stream data into the same parsing caliber. This allows subsequent feature separation and extraction to proceed item by item according to the sensing modality source marker, rather than mixing and reading sampling data streams corresponding to different sensing modalities. Specifically, based on the sampling channel identifier, the thermal gradient sampling data stream, gas concentration sampling data stream, and environmental vibration sampling data stream are first separated from the multimodal sensing stream data. These data streams are then written into the thermal gradient separation record, gas concentration separation record, and environmental vibration separation record, respectively. The thermal gradient separation record is further used to extract regional thermal gradient distribution attributes, the gas concentration separation record is further used to extract gas concentration abrupt change attributes, and the environmental vibration separation record is further used to extract environmental vibration attributes. Subsequently, the thermal gradient separation load record, the gas concentration separation load record, and the environmental vibration separation load record are time-segment aligned according to the sampling time caliber to form a time-aligned sensing load record. This time-aligned sensing load record ensures that the subsequently obtained thermal gradient distribution numerical sequence, gas concentration abrupt change numerical sequence, and environmental vibration numerical sequence correspond to the same time segment under the same sensing request command. Then, based on the sampling data format fields, the thermal gradient sampling data stream, gas concentration sampling data stream, and environmental vibration sampling data stream in the time-aligned sensing load record are processed to form a multimodal separation feature load record.The technical function of the multimodal separation feature carrying record is to convert the multimodal sensing flow data from the original sampling data stream into a feature reading entry organized according to the sensing mode, so that the subsequent regional thermal gradient distribution attribute, gas concentration change attribute and environmental vibration attribute can respectively form the corresponding thermal gradient distribution numerical sequence, gas concentration change numerical sequence and environmental vibration numerical sequence, and jointly participate in the construction of the target feature vector.

[0025] Preferably, the specific implementation process of the regional thermal gradient distribution attribute is as follows: After obtaining the multimodal separation feature carrying record, the thermal gradient separation carrying record in the multimodal separation feature carrying record is first read, and the image frame fragment corresponding to the vehicle image sampling channel, the temperature sampling fragment corresponding to the vehicle body temperature sampling channel, the sampling time caliber, and the road segment hash identifier are extracted from the thermal gradient separation carrying record to establish a thermal gradient attribute parsing record. The thermal gradient attribute parsing record is used to incorporate the image frame fragment, temperature sampling fragment, sampling time caliber, and road segment hash identifier into the same thermal gradient parsing caliber, so that the subsequent regional thermal gradient distribution attribute can simultaneously retain local spatial changes, temperature changes, and road segment attribution relationships. In this application, the regional thermal gradient distribution attribute is a numerical attribute used to describe the distribution state of heat changes in local space within the area corresponding to the specified geofence parameter. It is not equivalent to a single temperature value, nor is it equivalent to the original image itself, but is formed by the brightness response area in the image frame fragment, the temperature change trend in the temperature sampling fragment, and the spatial attribution relationship corresponding to the road segment hash identifier. In practice, the image frame segments are first divided into local regions according to the pixel arrangement to form thermal response grid segments. These thermal response grid segments convert local thermal response locations within the image frame segments into spatial units that can be numerically read, and they continue to serve as the spatial source for recording local thermal response values. Subsequently, the high-brightness response ratio, high-color-temperature response ratio, and thermal response region continuity are read from each thermal response grid segment, and combined with the temperature change corresponding to the same sampling time caliber in the temperature sampling segment, to form a local thermal response value record. This local thermal response value record is then associated with road segment hash identifiers, ensuring that it not only represents local thermal changes within the target vehicle's field of view but also falls within the road segment caliber corresponding to the specified geofence parameters. To reduce interference from different time periods and meteorological conditions on the local numerical records of the thermal response, a historical environmental benchmark parameter library corresponding to the road segment hash identifier, sampling time caliber, and real-time meteorological condition characteristic parameters can be read. Thermal response benchmark records are obtained from this library, and the local numerical records of the thermal response are compared with these benchmark records using deviation reading processing to form regional thermal gradient distribution attributes. The thermal response benchmark records provide a reference thermal response caliber under the same road segment hash identifier, the same sampling time caliber, and corresponding real-time meteorological condition characteristic parameters. After deviation reading processing, the regional thermal gradient distribution attributes can express the thermal response changes relative to the historical environmental benchmark parameter library. The regional thermal gradient distribution attributes are then written into a thermal gradient distribution numerical sequence, arranged according to the sampling time caliber, and each time segment stores the thermal response deviation value, thermal response area continuity, temperature change, and road segment hash identifier associated fields.Therefore, the thermal gradient distribution numerical sequence can participate in the construction of subsequent numerical sequences in a structured numerical manner, so that the thermal correlation dimension in the target feature vector does not depend on the cross-node transmission of the original image frame, but depends on the regional thermal gradient distribution attribute extracted locally by the target vehicle.

[0026] Preferably, the specific implementation process of the gas concentration mutation attribute is as follows: A gas concentration separation record is read from the multimodal separation feature record, and the volatile organic compound (VOC) sampling segment, fine particulate matter (PM2.5) sampling segment, carbon monoxide (CO) sampling segment, sampling time caliber, and road segment hash identifier corresponding to the in-vehicle gas sampling channel are read from the gas concentration separation record to establish a gas concentration attribute analysis record. This gas concentration attribute analysis record is used to incorporate the VOC sampling segment, PM2.5 sampling segment, CO sampling segment, sampling time caliber, and road segment hash identifier into the same gas concentration analysis caliber, so that subsequent gas concentration mutation attributes can retain the concentration change relationship of different gas types and the road segment attribution relationship. In this application, the gas concentration mutation attribute is a numerical attribute used to describe the rapid change of gas concentration within a specified geofence parameter area between consecutive time segments. It is not equivalent to a single gas concentration reading, but is jointly determined by the direction of concentration change, the magnitude of concentration change, and the spatial positional relationship corresponding to the road segment hash identifier within multiple consecutive sampling time calibers. In practice, the sampling segments for volatile organic compounds (VOCs), fine particulate matter (PM2.5), and carbon monoxide (CO) are first arranged in a time sequence according to the sampling time caliber to form a gas concentration time sequence record. This gas concentration time sequence record is used to preserve the chronological relationship of various gas sampling data within consecutive time segments and continues to serve as the time source for gas concentration change records. Subsequently, the changes in VOC, PM2.5, and CO concentration readings for adjacent time segments in the gas concentration time sequence record are processed to form a gas concentration change record. This gas concentration change record is then bound to a road segment hash identifier, enabling subsequent cloud aggregation nodes to identify the gas concentration change record corresponding to the same road segment hash identifier according to specified geofence parameters. To ensure that the gas concentration abrupt change attributes reflect the gas diffusion trend, the driving direction records of target vehicles within adjacent sampling time calibers can also be read, and these driving direction records are correlated with the gas concentration change records to form a gas diffusion direction record. The gas diffusion direction record is not used to identify the target vehicle, but rather to express the increasing or decreasing relationship of gas concentration changes along the road travel direction. This gas diffusion direction record continues to serve as the directional source of the gas concentration abrupt change attribute. Subsequently, the historical environmental benchmark parameter library corresponding to the road segment hash identifier and real-time meteorological condition characteristic parameters is read. Gas concentration benchmark records are obtained from this library, and the gas concentration change record, gas diffusion direction record, and gas concentration benchmark record are subjected to deviation reading processing to form the gas concentration abrupt change attribute.The gas concentration benchmark record provides a reference gas concentration caliber for the same road segment hash identifier and corresponding real-time meteorological condition characteristic parameters. After participating in deviation reading processing, the gas concentration abrupt change attribute enables it to express gas concentration changes relative to the historical environmental benchmark parameter library. This gas concentration abrupt change attribute is further written into a gas concentration abrupt change numerical sequence, which is arranged according to sampling time caliber. Each time segment stores the changes in volatile organic compound (VOC) concentration, fine particulate matter (PM2.5) concentration, carbon monoxide (CO) concentration, gas diffusion direction, and road segment hash identifier associated fields. Thus, the gas concentration abrupt change numerical sequence can provide a structured numerical source of gas concentration abrupt change attributes to subsequent target feature vectors without uploading the original gas sampling segments.

[0027] Preferably, the specific implementation process of environmental vibration attributes is as follows: The environmental vibration separation load record is read from the multimodal separation feature load record, and the lateral acceleration sampling segment, longitudinal acceleration sampling segment, vertical acceleration sampling segment, sampling time caliber, and road segment hash identifier corresponding to the three-axis acceleration sampling channel are read from the environmental vibration separation load record to establish an environmental vibration attribute analysis record. The environmental vibration attribute analysis record is used to incorporate the lateral acceleration sampling segment, longitudinal acceleration sampling segment, vertical acceleration sampling segment, sampling time caliber, and road segment hash identifier into the same environmental vibration analysis caliber, so that subsequent environmental vibration attributes can retain the joint change relationship between the lateral, longitudinal, and vertical axes, as well as the road segment attribution relationship. In this application, the environmental vibration attribute is a numerical attribute used to describe the amplitude change, direction source, and duration of impact, periodic vibration, or abnormal vibration in the vehicle body sampling coordinates within the area corresponding to the specified geofence parameter. It is not equivalent to a single-axis acceleration reading, but is formed by the joint change relationship between the lateral acceleration sampling segment, longitudinal acceleration sampling segment, and vertical acceleration sampling segment. In practice, the lateral acceleration sampling segments, longitudinal acceleration sampling segments, and vertical acceleration sampling segments are first segmented according to the sampling time caliber to form environmental vibration time segment records. These environmental vibration time segment records are used to preserve the synchronous sampling relationship of the lateral, longitudinal, and vertical directions within the same time segment, and continue to serve as the time source for the three-dimensional vibration amplitude records. Subsequently, the amplitude change readings of the lateral, longitudinal, and vertical acceleration sampling segments in each time segment of the environmental vibration time segment records are performed to form three-dimensional vibration amplitude records. Then, direction assignment processing is performed based on the relative magnitude relationships of the lateral, longitudinal, and vertical directions in the three-dimensional vibration amplitude records to form environmental vibration direction records. These environmental vibration direction records are further bound to road segment hash identifiers, enabling them to be assigned to the corresponding area of ​​the specified geofence parameters during subsequent spatiotemporal aggregation processing. To distinguish between short-term impacts and sustained vibrations, periodic density readings can be performed on the triaxial vibration amplitude records within a continuous time segment to form a sustained environmental vibration record. This sustained environmental vibration record indicates whether environmental vibrations maintain a similar vibration rhythm within a continuous time segment, and it continues to serve as the duration source for environmental vibration attributes. Subsequently, the historical environmental benchmark parameter library corresponding to the road segment hash identifier is read, and environmental vibration benchmark records are obtained from this library. The triaxial vibration amplitude records, environmental vibration direction records, environmental vibration sustained records, and environmental vibration benchmark records are then subjected to deviation readings to form environmental vibration attributes.The environmental vibration benchmark record is used to provide a reference environmental vibration caliber under the same road segment hash identifier. After participating in deviation reading processing, the environmental vibration attribute can express the environmental vibration change relative to the historical environmental benchmark parameter library. The environmental vibration attribute is further written into the environmental vibration numerical sequence, which is arranged according to the sampling time caliber, and stores the three-dimensional vibration amplitude record, environmental vibration direction record, environmental vibration duration record, and road segment hash identifier association field in each time segment. Thus, the environmental vibration numerical sequence can convert the environmental vibration attribute into structured numerical content that can participate in the construction of target feature vectors, and enable the subsequent cloud aggregation node to identify the vibration change trend within the same fenced area based on the environmental vibration numerical sequences of multiple target vehicles after receiving the desensitized data packet.

[0028] Preferably, the specific implementation process of constructing the target feature vector based on the numerical sequence is as follows: After forming the thermal gradient distribution numerical sequence, the gas concentration mutation numerical sequence, and the environmental vibration numerical sequence respectively, the thermal gradient distribution numerical sequence, the gas concentration mutation numerical sequence, and the environmental vibration numerical sequence are first written into the same target feature vector construction record, and the sampling time caliber, target perception mode identifier, and road segment hash identifier are simultaneously written into the target feature vector construction record. In this application, the numerical sequence is the structured numerical content formed by arranging the regional thermal gradient distribution attribute, gas concentration mutation attribute, and environmental vibration attribute through time segment arrangement, spatial attribution binding, and benchmark deviation reading; the numerical sequence is not the original sampling data, nor is it the desensitized perception data packet itself, but rather the direct data source for constructing the target feature vector. In specific implementation, the thermal gradient distribution numerical sequence, gas concentration mutation numerical sequence, and environmental vibration numerical sequence are first aligned according to the sampling time caliber to form a numerical sequence record with the same segment. The numerical sequence records of the same time segment are used to ensure that the regional thermal gradient distribution attribute, gas concentration abrupt change attribute, and environmental vibration attribute within the same time segment are arranged in the same vector position. These numerical sequence records of the same time segment continue to serve as input records for dimensionality sorting processing. Subsequently, according to the vector arrangement rules corresponding to the target perception modality identifier, the numerical sequence records of the same time segment are subjected to dimensionality sorting processing to form the target feature vector. The vector arrangement rules can be pre-written into the local processing node of the target vehicle. These rules determine the order of the thermal gradient distribution numerical sequence, gas concentration abrupt change numerical sequence, and environmental vibration numerical sequence in the target feature vector, ensuring that the target feature vectors generated by different target vehicles have a consistent dimensionality reading caliber. In this application, the target feature vector is a structured feature record extracted by the local processing node of the target vehicle from multimodal sensor stream data, which can characterize the multimodal risk state within the area corresponding to the specified geofence parameters. Different vector elements in the target feature vector correspond to thermal response deviation value, thermal response area continuity, temperature change, volatile organic compound concentration change, fine particulate matter concentration change, carbon monoxide concentration change, gas diffusion direction record, three-dimensional vibration amplitude record, environmental vibration direction record, and environmental vibration duration record, respectively. To enable subsequent cloud aggregation nodes to identify the temporal and spatial attribution of the target feature vector, the sampling time caliber and road segment hash identifier are written as association fields into the target feature vector to construct a carrying record. To enable subsequent cloud aggregation nodes to identify the perception source of the target feature vector, the target perception modality identifier is written as the source field into the target feature vector to construct a carrying record.After the above processing, the target feature vector can provide a structured representation of multimodal risk states to subsequent desensitized data packets without carrying the original image frames, original gas sampling segments, and original acceleration continuous segments, thus making the desensitized data packets suitable for spatiotemporal aggregation processing at cloud aggregation nodes.

[0029] Preferably, the specific implementation process of performing privacy isolation processing on the hardware identification code of the target vehicle and replacing the hardware identification code with a single-session-level random communication token is as follows: After the target feature vector is formed, the hardware identification code in the target vehicle's local identity configuration area is read first, and the hardware identification code is written into the token replacement bearer record. The target vehicle's local identity configuration area is used to store identity-type fields that can be read locally by the target vehicle but are not included in cross-node transmission. The hardware identification code is provided by the target vehicle's local identity configuration area and participates in privacy isolation processing as a field to be replaced in the token replacement bearer record. In this application, the hardware identification code is a local identity field used to identify the hardware entity to which the vehicle communication node or vehicle physical sensor matrix belongs locally in the target vehicle. The hardware identification code can come from one or more of the vehicle identification code, vehicle communication node device identification code, or vehicle physical sensor matrix device identification code. It can be used for device management locally, but is not suitable for direct writing into desensitized data packets for cross-node transmission. The token replacement bearer record is used to establish a temporary replacement relationship between the hardware identification code and the single-session-level random communication token without rewriting the target vehicle's local identity configuration area. In specific implementation, the request time period field in the perception request instruction, the certificate body field in the digital certificate, and the random seed field generated locally by the target vehicle are first read. These fields are then written into the token generation bearer record. The random seed field can be generated locally by the target vehicle upon receiving the perception request instruction and is used together with the request time period field and the certificate body field to generate a single-session-level random communication token. The token generation bearer record is not written to the desensitized data packet but is used to generate a single-session-level random communication token locally by the target vehicle. Subsequently, the request time period field, certificate body field, and random seed field in the token generation bearer record are irreversibly hashed and encapsulated to generate a single-session-level random communication token. In this application, the single-session-level random communication token is a temporary communication identifier that replaces the hardware identification code during the session corresponding to a perception request instruction. This token enables the cloud aggregation node to distinguish the source of different desensitized data packets within the same time segment but does not directly expose the hardware identification code. After generating the single-session-level random communication token, the hardware identification code in the token replacement bearer record is marked as a locally reserved field, and the single-session-level random communication token is written into the transmission identifier field of the token replacement bearer record to complete the privacy isolation process. In this application, the privacy isolation process separates the local identity field from the cross-node transmission field at the field level, so that when constructing the desensitized data packet, only the single-session-level random communication token is read as the communication source identifier, and the hardware identification code is not read as the communication source identifier.Therefore, the single-session-level random communication token continues to participate in the construction of desensitized data packets, and the hardware identification code is only retained in the target vehicle's local identity configuration area and token replacement bearer record, thereby reducing the traceability of the target vehicle's identity in subsequent multi-vehicle aggregation processing.

[0030] Preferably, the specific implementation process of mapping the physical location coordinates of the target vehicle to a road segment hash identifier and constructing the desensitized data packet based on the target feature vector, the single-session-level random communication token, and the road segment hash identifier is as follows: First, read the physical location coordinates in the target vehicle's local location sampling record and write the physical location coordinates into the location mapping bearer record. The target vehicle's local location sampling record is used to store the location sampling result formed by the target vehicle at the corresponding time of the sampling time caliber. The physical location coordinates are provided by the target vehicle's local location sampling record and participate in the road segment attribution processing as a field to be mapped in the location mapping bearer record. In this application, the physical location coordinates are the original spatial location field formed by the target vehicle at the corresponding time of the sampling time caliber. They can be used by the target vehicle to determine whether it is located in the area corresponding to the specified geofence parameter, but are not suitable for direct writing into the desensitized data packet for cross-node transmission. In specific implementation, the physical location coordinates are first processed for road segment attribution according to the road segment index caliber to form a road segment attribution record; the road segment attribution record is used to convert the original spatial location field into a road-level spatial location, rather than retaining precise coordinates. Subsequently, the road segment attribution record is hashed to generate the road segment hash identifier. In this application, the road segment hash identifier is an anonymous spatial index field formed by processing the physical location coordinates through road segment attribution and hash mapping. The road segment hash identifier enables the cloud aggregation node to determine whether the desensitized data packet originates from a road segment within the area corresponding to the specified geofence parameters, while simultaneously reducing the exposure range of the physical location coordinates in the transmission link. After generating the road segment hash identifier, the target feature vector, the single-session-level random communication token, and the road segment hash identifier are written into the desensitized data packet bearer record. The sampling time caliber, the target perception modality identifier, and the fence identifier from the specified geofence parameters are also written into the desensitized data packet bearer record. The desensitized knowledge data packet carrying record is used to incorporate multimodal risk status, temporary communication sources, and anonymous spatial locations into the same transmission caliber. The target feature vector carries structured results corresponding to regional thermal gradient distribution attributes, gas concentration abrupt change attributes, and environmental vibration attributes. The single-session-level random communication token replaces hardware identification encoding in cross-node transmission, and the road segment hash identifier replaces physical location coordinates in spatial attribution identification. Finally, the desensitized knowledge data packet carrying record, after incorporating the target feature vector, single-session-level random communication token, road segment hash identifier, sampling time caliber, target perception modality identifier, and fence identifier, undergoes field encapsulation processing to generate the desensitized knowledge data packet.In this application, the desensitized knowledge data packet is a structured perception transmission unit generated by the target vehicle's local processing node for the cloud aggregation node. The desensitized knowledge data packet does not carry hardware identification codes, physical positioning coordinates, original image frames, original gas sampling segments, and original continuous acceleration segments. Instead, it carries target feature vectors, single-session-level random communication tokens, and road segment hash identifiers, enabling the subsequent cloud aggregation node to perform spatiotemporal aggregation processing and generate a regional situational time series profile under the desensitized caliber.

[0031] Optionally, the step of performing spatiotemporal aggregation processing on multiple desensitized data packets from the area corresponding to the specified geofence parameter at the cloud aggregation node to generate a regional situation time-series profile includes: counting the total number of valid responses of the desensitized data packets carrying the same specified geofence parameter within a time slice of preset duration; extracting target feature vectors from each desensitized data packet and calculating the multimodal deviation statistics of the target feature vectors; spatially distributing and mapping the target feature vectors according to grid partitioning rules to generate a spatial heat map; and associating the total number of valid responses, the multimodal deviation statistics, and the spatial heat map in chronological order to construct the regional situation time-series profile.

[0032] Preferably, the specific implementation process for counting the total number of valid responses of the desensitized data packets carrying the same specified geofence parameter within a time slice of a preset duration is as follows: After the cloud aggregation node receives multiple desensitized data packets, it first reads the fence identifier, specified geofence parameter field, sampling time caliber, road segment hash identifier, single-session-level random communication token, and target perception modality identifier from each desensitized data packet, and writes the fence identifier, specified geofence parameter field, sampling time caliber, road segment hash identifier, single-session-level random communication token, and target perception modality identifier into the spatiotemporal aggregation bearer record. In this application, the preset duration is a time window length field pre-written into the time slice configuration record of the cloud aggregation node. The time window length field is not the alarm response time, nor the local sampling duration of the target vehicle, but is used to limit the time boundary for the cloud aggregation node to receive, filter, and assign multiple desensitized data packets within the same time slice. The time slice configuration record is read when a time slice is subsequently generated, and the preset duration is used in the process of determining the time slice start marker and time slice end marker through the time slice configuration record. In this application, the time slice is an aggregated time unit formed by dividing continuously received desensitized data packets into time windows based on the preset duration. Each time slice includes a time slice start marker, a time slice end marker, and a time slice sequence marker. The time slice sequence marker continues to participate in the time sorting of subsequent regional situation time series profiles. The preset duration can be configured according to the normal perception interval parameter, the transmission delay of desensitized data packets between the vehicle communication node and the cloud aggregation node, the regional situation time series profile's requirement for distinguishing continuous time changes, and the cloud aggregation node's receiving capacity for multiple desensitized data packets. This allows a time slice to receive desensitized data packets from multiple target vehicles while reducing the probability of risk signals being overly smoothed in the time dimension due to excessively long time spans. Specifically, in the statistical analysis, desensitized data packets whose sampling time caliber falls within the same time slice are first written into the time slice candidate record based on the time slice start marker and the time slice end marker. Then, the specified geofence parameter field in the time slice candidate record is read, and desensitized data packets carrying the same specified geofence parameter are written into the geofence response candidate record. The specified geofence parameter is used here to provide spatial attribution criteria, so that desensitized data packets from the same region corresponding to the specified geofence parameter can enter the same spatiotemporal aggregation processing flow, while desensitized data packets from different regions corresponding to the specified geofence parameters will not be mixed into the same total number of valid responses.Subsequently, the single-session-level random communication token, road segment hash identifier, and target perception modality identifier in the candidate response records within the fence are subjected to validity screening. Data packets lacking a single-session-level random communication token, lacking a road segment hash identifier, or whose target perception modality identifier does not fall within the allowed range of the perception request instruction are removed to form valid response records. Finally, the deduplicated data packets corresponding to different single-session-level random communication tokens in the valid response records are de-duplicated and counted to obtain the total number of valid responses. In this application, the total number of valid responses refers to the number of anonymous target vehicle responses that can participate in the construction of a regional situation time-series profile under the same time slice, the same specified geofence parameters, and the same perception request instruction constraints. The total number of valid responses is subsequently correlated with multimodal deviation statistics and spatial heat map to characterize the data coverage basis of the regional situation time-series profile within the time slice and to provide a statistical source for the number of valid responses contributing to the subsequent spatial heat map.

[0033] Preferably, the specific implementation process for extracting the target feature vector from each of the desensitized data packets is as follows: After forming the effective response record, the desensitized data packets are read item by item from the effective response record, and the target feature vector, sampling time caliber, road segment hash identifier, target perception modality identifier, and single-session-level random communication token are extracted from each of the desensitized data packets to establish a target feature vector reading record. The target feature vector reading record is used to uniformly incorporate target feature vectors under the same time slice and the same specified geofence parameters into the reading caliber of the cloud aggregation node, so that the subsequent calculation of multimodal deviation statistics can trace the time source, spatial source, and perception modality source of each target feature vector. In this application, the target feature vector is a structured feature record extracted from multimodal sensor stream data by the local processing node of the target vehicle. Its vector elements include thermal response deviation, thermal response region continuity, and temperature change corresponding to the regional thermal gradient distribution attribute; volatile organic compound concentration change, fine particulate matter concentration change, carbon monoxide concentration change, and gas diffusion direction record corresponding to the gas concentration abrupt change attribute; and triaxial vibration amplitude record, environmental vibration direction record, and environmental vibration duration record corresponding to the environmental vibration attribute. When extracting the target feature vector, the cloud aggregation node does not read the original image frames, original gas sampling segments, or original acceleration continuous segments. Instead, it reads the target feature vector after feature separation and extraction by the local processing node of the target vehicle, thus establishing spatiotemporal aggregation processing based on desensitized structured features. Specifically, the vector element arrangement of the target feature vector is first read according to the target perception modality identifier, and the vector element arrangement is written into a vector dimension verification record. This vector dimension verification record is used to determine whether the target feature vectors uploaded by different target vehicles form a consistent dimensional order according to the same target perception modality identifier. Subsequently, target feature vectors with consistent dimensional order are written into the target feature vector aggregation candidate record, while target feature vectors with missing dimensions, inconsistent dimensional order, or missing road segment hash identifiers are written into the abnormal read record. The abnormal read record does not participate in the calculation of the multimodal deviation statistics. The abnormal read record is used to store target feature vectors that did not enter the target feature vector aggregation candidate record and the reasons for their abnormality, enabling the cloud aggregation node to exclude target feature vectors with incomplete fields when subsequently generating the total number of valid responses and the multimodal deviation statistics. The target feature vector aggregation candidate record continues to serve as a common input for the calculation of the multimodal deviation statistics and spatial distribution mapping, allowing the same batch of target feature vectors to be used both for statistical analysis within a time slice and for spatial location attribution within a specified geofence parameter area.Therefore, the extraction of target feature vectors is not a simple reading of fields, but a structured screening of desensitized data packets based on time slices, specified geofence parameters, road segment hash identifiers, and target perception modality identifiers, so that subsequent multimodal deviation statistics can reflect the multimodal state changes of multiple target vehicles within the area corresponding to the same specified geofence parameter.

[0034] Preferably, in the specific technical implementation of calculating the multimodal deviation statistics of the target feature vector, multiple target feature vectors in the target feature vector aggregation candidate record are first read, and each target feature vector is split into a regional thermal gradient distribution dimension, a gas concentration mutation dimension, and an environmental vibration dimension according to the target perception modality identifier, so as to form a multimodal dimension arrangement record. The multimodal dimension arrangement record is used to align the target feature vectors uploaded by different target vehicles laterally according to the same perception modality, so that the regional thermal gradient distribution dimension is only compared with the regional thermal gradient distribution dimension, the gas concentration mutation dimension is only compared with the gas concentration mutation dimension, and the environmental vibration dimension is only compared with the environmental vibration dimension, avoiding the direct mixing and calculation of vector elements with different dimensions. Subsequently, the historical environmental benchmark parameter library corresponding to the specified geofence parameters, road segment hash identifier, sampling time caliber, and real-time meteorological condition feature parameters is read, and thermal response benchmark records, gas concentration benchmark records, and environmental vibration benchmark records are obtained from the historical environmental benchmark parameter library to form a multimodal benchmark reading record. The multimodal benchmark read records are used to provide reference feature calibers for the same region, time slice, and corresponding environmental conditions under the same specified geofence parameters. This allows the multimodal deviation statistics to represent the deviation of the current time slice from the historical environmental benchmark parameter library, rather than simply representing the differences between target feature vectors. Specifically, the regional thermal gradient distribution dimension in the multimodal dimension arrangement record is first processed to deviate from the thermal response benchmark record, forming a thermal gradient distribution deviation record. Then, the gas concentration mutation dimension is processed to deviate from the gas concentration benchmark record, forming a gas concentration mutation deviation record. Finally, the environmental vibration dimension is processed to deviate from the environmental vibration benchmark record, forming an environmental vibration deviation record. The thermal gradient distribution deviation record, the gas concentration mutation deviation record, and the environmental vibration deviation record are then merged within the same time slice according to the time slice sequence marking and road segment hash identifier, forming a merged deviation record for the same time slice. This merged deviation record is then statistically encapsulated according to the specified geofence parameters to form the multimodal deviation statistics. In this application, the multimodal deviation statistics are statistical feature records formed after reading the modal deviations of target feature vectors from multiple target vehicles within the same time slice, performing same-modal statistics, and merging across vehicles. These statistics may include the average deviation degree, dispersion degree, quantile state, and number of effective response contributions corresponding to each sensing modality. The number of effective response contributions is formed by the number of target feature vectors participating in the same-time slice merging process and is used to mark the data coverage basis within the corresponding spatial grid cell when subsequently generating a spatial thermal distribution map.The multimodal deviation statistics are able to reflect multimodality because they separately store records of thermal gradient distribution deviations, gas concentration abrupt change deviations, and environmental vibration deviations, and establish correlations between these three types of deviation records under the same time slice and the same specified geofence parameters. After this processing, the multimodal deviation statistics can serve as a statistical situation field in subsequent regional situation time-series profiling, enabling cloud-based aggregation nodes to determine whether the risk signal within the current time slice is a single sensing mode deviation or a simultaneous correlated deviation across multiple sensing modes.

[0035] Preferably, the specific implementation process of spatially distributing and mapping the target feature vector according to the grid partitioning rules is as follows: After obtaining the aggregated candidate records of the target feature vector and the multimodal deviation statistics, the grid partitioning rules corresponding to the specified geofence parameters are first read, and the grid partitioning rules are written into the spatial mapping carrying record. In this application, the grid partitioning rules are spatial coding rules that divide the area corresponding to the specified geofence parameters into multiple computable spatial units. The grid partitioning rules are not drawing rules for displaying maps, but spatial indexing rules for incorporating road segment hash identifiers, target feature vectors, and multimodal deviation statistics into the same spatial attribution caliber. The grid partitioning rules can be written when the specified geofence parameters are generated, or they can be configured after the cloud aggregation node receives the specified geofence parameters according to the fence type field, preset coverage radius, road segment hash identifier distribution, and target vehicle traffic density. The configured grid partitioning rules include spatial grid unit identifiers, spatial grid unit boundary fields, road segment hash identifier attribution fields, and grid resolution caliber fields. The spatial grid unit identifier is used to distinguish different spatial units within the area corresponding to the specified geofence parameters. The spatial grid unit boundary field is used to limit the anonymous spatial range corresponding to each spatial grid unit. The road segment hash identifier attribution field is used to indicate which road segment hash identifiers belong to the same spatial grid unit. The grid resolution field is used to limit the coarseness of the spatial distribution mapping. The spatial mapping carrying record continues to store the grid resolution field, so that subsequent spatial heat map can align spatial grid units of different time slices according to the same grid resolution field. In specific implementation, the road segment hash identifiers in the target feature vector aggregation candidate record are first read, and the spatial grid unit identifier corresponding to each target feature vector is determined according to the road segment hash identifier attribution field to form a spatial grid attribution record. The spatial grid attribution record is used to map each target feature vector from the road segment attribution to the spatial grid unit attribution, so that the cloud aggregation node can compare the risk status of different spatial grid units within the area corresponding to the specified geofence parameters. Subsequently, the thermal gradient distribution deviation records, gas concentration abrupt change deviation records, environmental vibration deviation records, and effective response contribution quantities from the multimodal deviation statistics are written into the spatial deviation carrying records under the corresponding spatial grid cell identifiers to complete the spatial distribution mapping. The spatial deviation carrying records are used to bind the multimodal deviation statistics and effective response contribution quantities under each spatial grid cell identifier to the same spatial calculation caliber. The technical function of this spatial distribution mapping is to convert the target feature vectors, originally organized by target vehicle and road segment hash identifiers, into regional distribution data organized by spatial grid cell identifiers. This allows subsequent spatial thermal distribution maps to reflect the multimodal deviation distribution of different spatial grid cells within the area corresponding to the specified geofence parameters, rather than simply generating a single overall fence statistical value.

[0036] Preferably, the specific implementation process for generating the spatial thermal distribution map is as follows: After completing the spatial distribution mapping, the spatial grid cell identifier, thermal gradient distribution deviation record, gas concentration mutation deviation record, environmental vibration deviation record, and effective response contribution quantity are read from the spatial deviation carrying record, and then the spatial grid cell identifier, thermal gradient distribution deviation record, gas concentration mutation deviation record, environmental vibration deviation record, and effective response contribution quantity are written into the spatial thermal map carrying record. In this application, the spatial thermal distribution map is a structured regional distribution data formed by using the spatial grid cell identifier as an index, multimodal deviation statistics as numerical content, and the effective response contribution quantity as data coverage basis; the spatial thermal distribution map is not simply a display of images, nor is it a direct display of the target vehicle location, but rather a computable map of the spatial distribution of risk signals within the area corresponding to specified geofence parameters, expressed by cloud aggregation nodes under anonymized conditions. In the specific generation process, firstly, the thermal gradient distribution deviation records under each spatial grid cell identifier are merged within the grid to form a spatial thermal gradient value; then, the gas concentration abrupt change deviation records under each spatial grid cell identifier are merged within the grid to form a gas concentration value; and finally, the environmental vibration deviation records under each spatial grid cell identifier are merged within the grid to form an environmental vibration value. The spatial thermal gradient values, gas concentration values, and environmental vibration values ​​are then written into the spatial thermal map carrying record according to the same spatial grid cell identifier, enabling each spatial grid cell to store the deviation intensity corresponding to different sensing modes. Subsequently, based on the number of effective response contributions, the spatial thermal gradient values, gas concentration values, and environmental vibration values ​​within the spatial grid cell are subjected to credibility marking processing to form a spatial thermal credibility marker. The spatial thermal reliability marker is used to indicate whether the spatial thermal gradient values, gas concentration values, and environmental vibration values ​​within the corresponding spatial grid cell are formed by a large number of desensitized data packets or by a small number of desensitized data packets. This spatial thermal reliability marker participates in the situation field association along with the total number of valid responses in the subsequent regional situation time-series profile. Finally, the spatial thermal map carrying record, after being written with the spatial thermal gradient values, gas concentration values, environmental vibration values, and spatial thermal reliability marker, undergoes map encapsulation processing to generate the spatial thermal distribution map. This spatial thermal distribution map is not used for reverse reconstruction of the target vehicle's physical location coordinates, but rather to express the multimodal deviation aggregation locations of different spatial grid cells in the regional situation time-series profile, enabling the generation of graded early warning event sequences to be based on spatial distribution.

[0037] Preferably, in the specific technical implementation of associating the total number of valid responses, the multimodal deviation statistics, and the spatial thermal distribution map in chronological order, the time slice sequence marker of each time slice is first read, and the total number of valid responses, the multimodal deviation statistics, and the spatial thermal distribution map corresponding to the same specified geofence parameter are written into the situation time-bound loading record according to the time slice sequence marker. In this application, the chronological order refers to arranging the results of different time slices generated by the cloud aggregation node according to the time slice sequence marker, rather than arbitrarily arranging them according to the order in which the desensitized data packets arrive at the cloud aggregation node; this processing can reduce the interference caused by network transmission delay on the temporal relationship of the regional situation time-series profile. The situation time-bound loading record includes a fence identifier, a time slice sequence marker, a total number of valid responses field, a multimodal deviation statistics field, and a spatial thermal distribution map field. The fence identifier is used to limit the situation time-bound loading record to belonging to the same specified geofence parameter; the total number of valid responses field is used to characterize the data coverage basis of the corresponding time slice; the multimodal deviation statistics field is used to characterize the statistical situation change of the corresponding time slice; and the spatial thermal distribution map field is used to characterize the spatial distribution state of the corresponding time slice. In practice, the total number of valid responses under the same time slice sequence is first written into the "Total Valid Responses" field. Then, the multimodal deviation statistics under the same time slice sequence are written into the "Multimodal Deviation Statistics" field. Finally, the spatial thermal distribution map under the same time slice sequence is written into the "Spatial Thermal Distribution Map" field, forming a single-slice situational correlation record. This single-slice situational correlation record expresses the correspondence between data coverage basis, statistical situational changes, and spatial distribution status within the same time slice. For example, when the total number of valid responses is low but the spatial thermal distribution map of a certain spatial grid cell shows a high thermal value, subsequent regional situational time-series profiling can simultaneously read the risk signal location and data coverage basis, reducing the probability of making situational judgments based solely on a single spatial thermal value. Subsequently, multiple single-slice situational correlation records are sequentially spliced ​​according to the time slice sequence to form a situational time-series record. This situational time-series record continues to participate in the construction of the regional situational time-series profiling, enabling changes in multimodal deviation statistics, spatial thermal distribution map migration, and changes in the total number of valid responses between different time slices to be read and compared on the same time axis.

[0038] Preferably, the specific implementation process for constructing the regional situation time-series profile is as follows: After forming the situation time-series record, the fence identifier, time slice sequence marker, total number of effective responses, multimodal deviation statistics, and spatial heat distribution map in the situation time-series record are read first, and the fence identifier, time slice sequence marker, total number of effective responses, multimodal deviation statistics, and spatial heat distribution map are written into the regional situation profile carrying record. In this application, the regional situation time-series profile is a spatiotemporal situation data structure composed of the total number of effective responses, multimodal deviation statistics, and spatial heat distribution map in multiple consecutive time slices for the same specified geographic fence parameter area; the regional situation time-series profile is not a single alarm tag, nor a single spatial heat distribution map, but a continuous spatiotemporal situation data structure that can be read by the subsequent dedicated risk judgment rule base. Specifically, during construction, the single-slice situation association records in the regional situation profile carrying record are first sorted chronologically according to the time slice sequence marker to form a time-ordered situation record; the time-ordered situation record is used to retain the sequential relationship between multiple time slices. Subsequently, the total number of valid responses in the time-ordered situation record is processed to read the response quantity change, forming a response quantity change field; the multimodal deviation statistics in the time-ordered situation record are processed to read the deviation trend, forming a multimodal deviation trend field; and the spatial thermal distribution map in the time-ordered situation record is processed to read spatial migration, forming a spatial thermal migration field. The response quantity change field describes the temporal change in the number of target vehicle responses within the area corresponding to the same specified geofence parameter. The multimodal deviation trend field describes the direction of change of thermal gradient distribution deviation records, gas concentration abrupt change deviation records, and environmental vibration deviation records in continuous time slices. The spatial thermal migration field describes the distribution change of multimodal deviations among different spatial grid cells. Then, the response quantity change field, multimodal deviation trend field, and spatial thermal migration field are written into the regional situation profile carrying record and associated with the fence identifier and time slice sequence marker to generate the regional situation time-series profile. The aforementioned regional situation time-series profile allows for the reading of feature surge records for the current time slice by a dedicated risk assessment rule base. It also enables cloud-based aggregation nodes to determine, before generating a tiered warning event sequence, whether deviations are independent deviations of single-modal features, joint deviations of multi-modal features, or continuous deviations across consecutive time slices. Thus, the regional situation time-series profile organizes the data coverage provided by the effective response total, the statistical situational changes provided by multi-modal deviation statistics, and the spatial distribution status provided by the spatial heat map into a unified spatiotemporal situational data structure. This allows the regional request-based vehicle multi-modal perimeter perception and risk warning method to form continuous data support for regional risk assessment based on desensitized data packets.

[0039] Optionally, generating a corresponding graded early warning event sequence based on the regional situation time-series profile and a pre-configured dedicated risk assessment rule base corresponding to the specified geofence parameters includes: extracting feature surge records of the current time slice from the regional situation time-series profile; querying trigger condition tags matching the feature surge records in the dedicated risk assessment rule base; generating candidate observation-level event tags if the trigger condition tags satisfy the single-modal feature independent deviation condition; generating emergency high-risk event tags if the trigger condition tags satisfy the multi-modal feature joint deviation condition; and constructing the graded early warning event sequence in a standardized format based on the generated candidate observation-level event tags or emergency high-risk event tags.

[0040] Preferably, the specific implementation process for extracting the feature burst record of the current time slice from the regional situation time series profile is as follows: After the regional situation time series profile has been generated at the cloud aggregation node, the fence identifier, time slice sequence marker, total number of valid responses, multimodal deviation statistics, spatial heat distribution map, response quantity change field, multimodal deviation trend field, and spatial heat migration field in the regional situation time series profile are read first, and the fence identifier, time slice sequence marker, total number of valid responses, multimodal deviation statistics, spatial heat distribution map, response quantity change field, multimodal deviation trend field, and spatial heat migration field are written into the feature burst parsing carrying record. In this application, the current time slice is the current aggregated time unit to be judged in the regional situation time series profile determined according to the time slice sequence marker. The current time slice is not determined by the order in which the desensitized data packets arrive at the cloud aggregation node, but by the time slice sequence marker that the sampling time caliber falls into, so as to reduce the impact of network transmission delay on the risk judgment time relationship. Specifically, when determining the current time slice, the single-slice situation association record with the largest time slice sequence mark in the regional situation time series profile can be read first, and this single-slice situation association record can be written into the current time slice determination record. If the total number of valid responses in the single-slice situation association record is lower than the lower limit field of the number of valid responses pre-configured in the dedicated risk determination rule base, then the single-slice situation association records of the adjacent previous time slice can be read, and the single-slice situation association records of the adjacent previous time slice can be supplemented and associated with the current time slice determination record, so that the current time slice determination record has both time relevance and data coverage. In this application, the feature surge record is a mutation-type risk state record extracted from the multimodal deviation statistics and spatial thermal distribution map corresponding to the current time slice determination record. It is used to describe the rapid deviation of regional thermal gradient distribution attributes, gas concentration mutation attributes, or environmental vibration attributes in the current time slice relative to the adjacent previous time slice or the historical environmental benchmark parameter library. Specifically, when extracting the feature abrupt increase record, first read the thermal gradient distribution deviation record, gas concentration abrupt change deviation record, and environmental vibration deviation record in the current time slice determination record, and perform same-mode change reading processing with the thermal gradient distribution deviation record, gas concentration abrupt change deviation record, and environmental vibration deviation record in the adjacent previous time slice to form thermal gradient abrupt increase field, gas concentration abrupt increase field, and environmental vibration abrupt increase field; then read the spatial thermal distribution map in the current time slice determination record, and read the grid-in-grid change state of thermal gradient spatial thermal value, gas concentration spatial thermal value, and environmental vibration spatial thermal value according to the spatial grid cell identifier to form spatial abrupt increase location field.Subsequently, the fields for sudden increases in thermal gradient, gas concentration, environmental vibration, spatial location, total number of effective responses, and time slice sequence are written into the same feature abrupt increase record. This feature abrupt increase record continues to serve as input for querying the dedicated risk assessment rule base, ensuring that subsequent risk assessments do not directly rely on a single desensitized data packet, but rather on the current time slice assessment record after spatiotemporal aggregation processing, deviation trend reading processing, and spatial distribution mapping.

[0041] Preferably, the specific implementation process of querying the trigger condition tags matching the feature surge record in the dedicated risk judgment rule base is as follows: After the feature surge record is formed, the fence identifier, time slice sequence marker, thermal gradient surge field, gas concentration surge field, environmental vibration surge field, spatial surge location field, total number of effective responses, and road segment hash identifier association field in the feature surge record are read first, and the fence identifier, time slice sequence marker, thermal gradient surge field, gas concentration surge field, environmental vibration surge field, spatial surge location field, total number of effective responses, and road segment hash identifier association field are written into the rule query carrying record. In this application, the dedicated risk judgment rule base is a risk judgment data set that establishes a corresponding relationship with specified geofence parameters. It is not a simple management list, but is read by the cloud aggregation node after the regional situation time series profile is generated, and is used as the judgment basis for feature surge record matching, event level determination, and hierarchical early warning event sequence construction. The dedicated risk assessment rule base can be established during the generation stage of specified geofence parameters, or after the specified geofence parameters are received by the cloud aggregation node. During establishment, the fence identifier, fence type field, risk type field, trigger condition label, single-modal feature independent deviation condition, multi-modal feature joint deviation condition, hazard level identifier, event sequence format field, and preset receiving endpoint field are written into the same rule base record. This ensures that the area corresponding to the same specified geofence parameter has a judgment caliber that matches its spatial range, perception modality, and risk type. In this application, the trigger condition label is a condition index field in the dedicated risk assessment rule base used to map feature surge records to risk assessment branches. It is neither the warning event itself nor the final hazard level, but rather an intermediate judgment field used to mark which judgment entry point the feature surge record meets. During the specific query, firstly, based on the rules, the fence identifier in the carrying record is queried, and the rule base record corresponding to the specified geofence parameter is read from the dedicated risk judgment rule base, and this rule base record is written into the fence rule reading record. Then, based on the rules, the thermal gradient surge field, gas concentration surge field, and environmental vibration surge field in the carrying record are queried, and the trigger condition label candidates corresponding to thermal gradient anomalies, gas concentration abrupt anomalies, environmental vibration anomalies, and multimodal joint anomalies are read item by item from the fence rule reading record. Subsequently, the trigger condition label candidates are matched with the spatial surge location field and the total number of valid responses in the feature surge record to form a trigger condition label matching record.The trigger condition label matching record is used to express the matching relationship between the feature surge record and the dedicated risk judgment rule base. When any one of the fields—thermal gradient surge, gas concentration surge, or environmental vibration surge—independently satisfies the corresponding judgment entry, the trigger condition label matching record is written into a single-modal trigger condition label. When at least two of the fields—thermal gradient surge, gas concentration surge, and environmental vibration surge—simultaneously satisfy the corresponding judgment entry in the same time slice and adjacent spatial grid cells, the trigger condition label matching record is written into a multi-modal trigger condition label. The trigger condition label matching record continues to participate in the judgment of single-modal feature independent deviation conditions and multi-modal feature joint deviation conditions, enabling subsequent candidate observation-level event labels or emergency high-risk event labels to be generated based on the same dedicated risk judgment rule base.

[0042] Preferably, if the trigger condition label satisfies the independent deviation condition of a single modality feature, the pre-judgment process for generating candidate observation-level event labels is as follows: After obtaining the trigger condition label matching record, the single modality trigger condition label in the trigger condition label matching record is first read, and the thermal gradient surge field, gas concentration surge field, environmental vibration surge field, spatial surge location field, total number of effective responses, and time slice sequence marker are simultaneously read from the feature surge record to establish a single modality deviation judgment record. In this application, the independent deviation condition of a single modality feature is a combination of conditions used to determine whether only one sensing modality has an independent abnormal deviation relative to the historical environmental benchmark parameter library or the adjacent previous time slice within the current time slice; the independent deviation condition of a single modality feature is not a single threshold comparison, but a judgment criterion composed of deviation amplitude condition, duration condition, spatial consistency condition, and effective response quantity condition. The deviation magnitude condition is used to determine whether one of the fields—thermal gradient surge, gas concentration surge, or environmental vibration surge—reaches the observation-level deviation boundary of the corresponding risk type field in the dedicated risk assessment rule base. The duration condition is used to determine whether the field maintains the same direction of deviation in the current time slice and the adjacent previous time slice. The spatial consistency condition is used to determine whether the spatial surge location field corresponding to the field is concentrated in the same spatial grid cell or adjacent spatial grid cells. The effective response quantity condition is used to determine whether the total number of effective responses participating in the formation of the field reaches the lower limit field of the effective response quantity pre-configured in the dedicated risk assessment rule base. Specifically, the thermal gradient surge, gas concentration surge, and environmental vibration surge fields are first processed independently for deviation reading to form independent deviation records for thermal gradient, gas concentration, and environmental vibration. Then, the independent deviation records for thermal gradient, gas concentration, and environmental vibration are processed for modal quantity verification to form a single-modal quantity verification record. When the single-modal quantity verification record characterization shows that only one of the independent deviation records of thermal gradient, gas concentration, or environmental vibration satisfies the deviation amplitude condition, the duration condition, the spatial consistency condition, and the effective response quantity condition, the single-modal deviation judgment record is written with a single-modal feature independent deviation condition satisfaction flag. This single-modal feature independent deviation condition satisfaction flag does not directly form a hierarchical early warning event sequence, but rather serves as the triggering basis for subsequently generating candidate observation-level event labels. This ensures that the candidate observation-level event labels can correspond to the independent deviation state of a single sensing mode, avoiding the direct escalation of risk signals that have not yet demonstrated multimodal correlation to higher-level events.

[0043] Preferably, the specific implementation process for generating candidate observation-level event tags is as follows: After the single-mode deviation judgment carrier record is written with the single-mode feature independent deviation condition satisfaction mark, the single-mode trigger condition tag, thermal gradient independent deviation record, gas concentration independent deviation record, environmental vibration independent deviation record, spatial abrupt change location field, total number of effective responses, and time slice sequence mark in the single-mode deviation judgment carrier record are read first. Then, the single-mode trigger condition tag, the thermal gradient independent deviation record, the gas concentration independent deviation record, the environmental vibration independent deviation record, the spatial abrupt change location field, the total number of effective responses, and the time slice sequence mark are written into the candidate observation-level event carrier record. In this application, the candidate observation-level event tag is a low-level early warning event identifier formed after a single sensing mode independent deviation is identified in the current time slice. It is used to enable the cloud aggregation node to enter the continuous observation and subsequent time slice review process of the current risk status, rather than directly determining the risk status as an emergency high-risk level. In the specific generation process, the risk type field of the candidate observation-level event label is first determined based on the single-modal trigger condition label. When the independent deviation condition of the single-modal feature is met by a thermal gradient independent deviation record, the risk type field is written into the thermal gradient anomaly observation type. When the independent deviation condition of the single-modal feature is met by a gas concentration independent deviation record, the risk type field is written into the gas concentration sudden change anomaly observation type. When the independent deviation condition of the single-modal feature is met by an environmental vibration independent deviation record, the risk type field is written into the environmental vibration anomaly observation type. Subsequently, the spatial abrupt change location field is read and written into the event location field in the candidate observation-level event carrier record. The time slice sequence marker is also read and written into the event time field in the candidate observation-level event carrier record, ensuring that the candidate observation-level event label retains the source relationship between the spatial grid cell and the current time slice in subsequent hierarchical early warning event sequences. Then, data overlay marking processing is performed on the candidate observation-level event carrier record based on the total number of valid responses to form an observation-level trusted marker. This observation-level trusted marker is used to express how many anonymous target vehicle response data are used to form the single-modal deviation state corresponding to the candidate observation-level event label. Then, the risk type field, event location field, event time field, observation level credibility marker, and the independent deviation records of thermal gradient, gas concentration, or environmental vibration that triggered the candidate observation level event label are written into the candidate observation level event label. The independent deviation records of thermal gradient, gas concentration, or environmental vibration that triggered the candidate observation level event label are then written into the single-modal evidence summary field. The single-modal evidence summary field is used to record the source of deviation in a single sensing mode corresponding to the candidate observation level event label, and serves as the source field for the evidence summary field when constructing a standardized format of hierarchical early warning event sequences.The candidate observation-level event labels continue to participate in the construction of the standardized format of the hierarchical early warning event sequence, and can be read by the subsequent dedicated risk judgment rule base in adjacent time slices to determine whether the single-sensory modality deviation corresponding to the candidate observation-level event label has been transformed into a continuous deviation in the continuous time slice or a joint deviation of multi-modal features. Thus, the candidate observation-level event labels retain the source of risk signals while avoiding directly using short-term deviations of a single-sensory modality as emergency high-risk event labels, thereby providing a clearer hierarchical generation basis for the hierarchical early warning event sequence.

[0044] Preferably, if the trigger condition label satisfies the multimodal feature joint deviation condition, the pre-judgment process for generating an emergency high-risk event label is as follows: After obtaining the trigger condition label matching record, the multimodal trigger condition label in the trigger condition label matching record is read first, and the thermal gradient surge field, gas concentration surge field, environmental vibration surge field, spatial surge location field, total number of effective responses, multimodal deviation statistics, and spatial thermal distribution map are simultaneously read from the feature surge record to establish a multimodal joint judgment carrying record. In this application, the multimodal feature joint deviation condition is used to determine whether there is a combination of conditions for at least two sensing modalities to deviate in time and space within the same specified geographic fence parameter area in the current time slice; the multimodal feature joint deviation condition is not a direct addition of vector elements of different dimensions, nor is it a simple arrangement of multiple single-modal feature independent deviation conditions, but a joint judgment based on the correspondence between the same time slice, adjacent spatial grid cells, and multimodal deviation statistics. In the specific determination, firstly, the thermal gradient distribution deviation record, gas concentration abrupt change deviation record, and environmental vibration deviation record in the multimodal deviation statistics are read, and these records are written into the same-modal deviation verification record. The same-modal deviation verification record is used to retain the dimensionality of each sensing mode. Subsequently, the spatial grid cell identifier, thermal gradient spatial thermal value, gas concentration spatial thermal value, environmental vibration spatial thermal value, and spatial thermal credibility mark in the spatial thermal distribution map are read, and these records are written into the spatial joint verification record. The spatial joint verification record is used to determine whether the deviations corresponding to different sensing modes are concentrated in the same spatial grid cell or adjacent spatial grid cells. The spatial thermal credibility mark is used to participate in the formation of the high-risk credibility mark when generating emergency high-risk event labels in the future. Then, based on the time slice sequence markers, the multimodal deviation statistics in the current time slice and the adjacent previous time slice are read to form a time joint verification record. This time joint verification record is used to determine whether the deviations of at least two perception modes have a unidirectional relationship in continuous time slices. When the unimodal deviation verification record indicates that at least two perception modes reach the joint deviation boundary of the corresponding risk type field in the dedicated risk judgment rule base, and the spatial joint verification record indicates that the deviation positions of at least two perception modes are associated with spatial grid cells, and the time joint verification record indicates that the deviations of at least two perception modes have a unidirectional relationship in continuous time slices, the multimodal joint judgment record is marked as satisfying the multimodal feature joint deviation condition.The joint deviation condition of the multimodal features is satisfied and the marker continues to serve as the trigger for generating emergency high-risk event tags, so that emergency high-risk event tags come from the combined judgment results of multimodal, time and space.

[0045] Preferably, the specific implementation process for generating an emergency high-risk event label is as follows: After the multimodal joint judgment carrier record is written with the multimodal feature joint deviation condition satisfaction flag, the multimodal trigger condition label, same-modal deviation verification record, spatial joint verification record, temporal joint verification record, total number of effective responses, spatial surge location field, and time slice sequence flag in the multimodal joint judgment carrier record are read first, and then the multimodal trigger condition label, same-modal deviation verification record, spatial joint verification record, temporal joint verification record, total number of effective responses, spatial surge location field, and time slice sequence flag are written into the emergency high-risk event carrier record. In this application, the emergency high-risk event label is a high-level early warning event identifier formed after the multimodal feature joint deviation condition is identified within the current time slice. It is used to enable subsequent graded early warning event sequences to distinguish the multimodal correlation deviation state from the candidate observation-level event label. In the specific generation process, the risk type field of the emergency high-risk event label is first determined based on the multimodal trigger condition label. When the thermal gradient distribution deviation record and the gas concentration mutation deviation record deviate simultaneously within the same or adjacent spatial grid cells, the risk type field is written into the thermal gradient and gas concentration mutation composite anomaly type. When the gas concentration mutation deviation record and the environmental vibration deviation record deviate simultaneously within the same or adjacent spatial grid cells, the risk type field is written into the gas concentration mutation and environmental vibration composite anomaly type. When the thermal gradient distribution deviation record, the gas concentration mutation deviation record, and the environmental vibration deviation record all meet the joint deviation boundary within the same time slice, the risk type field is written into the multimodal composite anomaly type. Subsequently, the spatial grid cell identifier in the spatial joint verification record is read and written into the event location field of the emergency high-risk event carrier record. The time slice sequence mark in the time joint verification record is read and written into the event time field of the emergency high-risk event carrier record. Then, based on the total number of valid responses and the spatial thermal credibility marker, the records of high-risk emergency events are subjected to data overlay marking to form high-risk credibility markers. These high-risk credibility markers express how many anonymous target vehicle response data and how many spatial grid cells jointly form the multimodal joint deviation state corresponding to the high-risk emergency event label. Next, the risk type field, event location field, event time field, high-risk credibility marker, trigger sensing modality field, and multimodal evidence summary field are written into the high-risk emergency event label. The trigger sensing modality field records the regional thermal gradient distribution attribute, gas concentration mutation attribute, and environmental vibration attribute participating in the multimodal feature joint deviation condition. The multimodal evidence summary field records the determination source of thermal gradient distribution deviation records, gas concentration mutation deviation records, environmental vibration deviation records, spatial joint verification records, and temporal joint verification records.The emergency high-risk event tag continues to participate in the construction of the standardized format graded early warning event sequence, and serves as the source field for the hazard level identifier when the dynamic event fence extension processing mechanism is subsequently triggered, so that high-risk events can be technically connected with the updated geofence parameter generation process.

[0046] Preferably, the specific implementation process of constructing the standardized format of the graded early warning event sequence based on the generated candidate observation-level event tags or emergency high-risk event tags is as follows: After obtaining the candidate observation-level event tags or emergency high-risk event tags, first read the risk type field, event location field, event time field, observation-level trust mark, and single-modal evidence summary field from the candidate observation-level event tags, or read the risk type field, event location field, event time field, high-risk trust mark, trigger perception modality field, and multimodal evidence summary field from the emergency high-risk event tags, and write the read fields into the graded early warning event carrying record. In this application, the standardized format is an event data structure used to enable early warning events corresponding to different fence identifiers, different risk type fields, and different hazard level identifiers to be parsed by a preset receiving endpoint according to the same field arrangement caliber; the standardized format is not a display format, but rather a field encapsulation caliber that the graded early warning event sequence must satisfy before cross-node distribution. The standardized format can consist of a fence identification field, a hazard level identification field, a trigger sensing modality field, a comprehensive anomaly score field, an anonymous response quantity field, a trigger time window field, a risk type field, and an evidence summary field. The fence identification field is used to characterize the specified geographic fence parameters that trigger the warning; the hazard level identification field is used to distinguish between candidate observation-level event labels and emergency high-risk event labels; the trigger sensing modality field is used to record the sensing modalities participating in the trigger condition label matching; the comprehensive anomaly score field is used to record the degree of deviation formed by reading multimodal deviation statistics; the anonymous response quantity field is used to record the total number of valid responses; the trigger time window field is used to record the current time slice; the risk type field is used to record the thermal gradient anomaly observation type, gas concentration mutation anomaly observation type, environmental vibration anomaly observation type, thermal gradient and gas concentration mutation combined anomaly type, gas concentration mutation and environmental vibration combined anomaly type, or multimodal combined anomaly type; and the evidence summary field is used to record thermal gradient distribution deviation records, gas concentration mutation deviation records, environmental vibration deviation records, spatial abrupt increase location field, spatial thermal credibility marker, single-modal evidence summary field, or multimodal evidence summary field. In the specific construction process, first, the danger level identifier field is written based on the candidate observation-level event label or the emergency high-risk event label; then, the fence identifier field and risk type field are written based on the fence identifier and event location field; then, the anonymous response quantity field is written based on the total number of valid responses; then, the trigger time window field is written based on the time slice sequence mark; then, the evidence summary field is written based on the single-modal evidence summary field or the multi-modal evidence summary field; subsequently, based on the single-modal deviation degree corresponding to the single-modal evidence summary field, or based on the multi-modal joint deviation degree corresponding to the multi-modal evidence summary field, the comprehensive anomaly score field is written to form a single standardized early warning event record.The comprehensive anomaly score field is not used to replace the hazard level identifier field, but rather to sort multiple standardized early warning event records under the same hazard level identifier field. Subsequently, the multiple standardized early warning event records are arranged sequentially according to the time slice order marker, and the standardized early warning event records within the same time slice are sorted by hazard level identifier to construct the graded early warning event sequence. In this application, the graded early warning event sequence is a distributable event data sequence formed by encapsulating candidate observation-level event tags and emergency high-risk event tags in a standardized format. The graded early warning event sequence is further read by the early warning distribution module and distributed to the preset receiving endpoint, enabling the preset receiving endpoint to parse the risk type, hazard level, trigger perception modality, number of anonymous responses, trigger time window, and evidence summary without receiving the original image frames, original gas sampling segments, original acceleration continuous segments, hardware identification codes, and physical positioning coordinates.

[0047] Optionally, after generating the graded early warning event sequence, the method further includes: parsing the hazard level identifier carried by the graded early warning event sequence; if the hazard level identifier corresponds to an emergency high-risk hazard level, triggering a dynamic event fence expansion processing mechanism; modifying the initial coverage radius corresponding to the specified geofence parameter to an expanded coverage radius through the dynamic event fence expansion processing mechanism to obtain the updated geofence parameter; and sending a perception request instruction carrying the updated geofence parameter to all regional nodes located within the expanded coverage radius.

[0048] Preferably, the specific implementation process for parsing the hazard level identifier carried by the graded early warning event sequence is as follows: After the graded early warning event sequence has been generated by encapsulating candidate observation-level event tags or emergency high-risk event tags in a standardized format, the fence identifier field, hazard level identifier field, trigger perception modality field, comprehensive anomaly score field, anonymous response quantity field, trigger time window field, risk type field, and evidence summary field in the graded early warning event sequence are read first, and the fence identifier field, the hazard level identifier field, the trigger perception modality field, the comprehensive anomaly score field, the anonymous response quantity field, the trigger time window field, the risk type field, and the evidence summary field are written into the hazard level parsing carrying record. In this application, the hazard level identifier is an event level field generated by a dedicated risk judgment rule base based on the feature surge record, trigger condition tag, candidate observation-level event tag, or emergency high-risk event tag in the regional situation time series profile. It is used to indicate the processing branch of the current graded early warning event sequence in subsequent perception range adjustment, perception request command reconstruction, and preset receiving endpoint distribution. The hazard level identifier is not displayed by color or as a separate text prompt, but rather as a data field that can be further read by the cloud aggregation node. When the tiered early warning event sequence was generated, the hazard level identifier had already established a field association relationship within the same event record with the risk type field, trigger perception modality field, trigger time window field, and evidence summary field. During specific analysis, the hazard level identifier field is first located from the hazard level analysis record according to the field arrangement criteria in the standardized format. Then, the hazard level identifier value corresponding to the hazard level identifier field is read, and the hazard level identifier value is compared with the single-modal evidence summary field or multi-modal evidence summary field in the evidence summary field to form a hazard level verification record. The hazard level verification record is used to determine whether the hazard level identifier is consistent with the source of the candidate observation-level event label or the emergency high-risk event label. When the hazard level identifier field represents a candidate observation-level event label, the hazard level verification record continues to read the single-modal evidence summary field to confirm that the hazard level identifier originates from the single-modal feature independent deviation condition. When the hazard level identifier field represents an emergency high-risk event label, the hazard level verification record continues to read the multimodal evidence summary field to confirm that the hazard level identifier originates from the multimodal feature joint deviation condition. After completing the above analysis, the hazard level verification record is written into the event follow-up processing record, so that whether the dynamic event fence expansion processing mechanism is triggered subsequently does not depend on the overall text judgment of the graded warning event sequence, but on the hazard level identifier that has been parsed by standardized fields and verified by evidence source.

[0049] Preferably, if the hazard level identifier corresponds to an emergency high-risk level, the specific implementation process of triggering the dynamic event fence extension processing mechanism is as follows: After forming the hazard level verification record, first read the hazard level identifier, risk type field, trigger perception modality field, trigger time window field, multimodal evidence summary field, spatial abrupt location field, and fence identifier field from the hazard level verification record, and write the hazard level identifier, the risk type field, the trigger perception modality field, the trigger time window field, the multimodal evidence summary field, the spatial abrupt location field, and the fence identifier field into the dynamic event fence extension trigger record. In this application, the emergency high-risk level is a high-level event status field written into the graded early warning event sequence after the dedicated risk judgment rule base identifies that the joint deviation condition of multimodal features is met; the emergency high-risk level is used to indicate that the risk signal in the area corresponding to the currently specified geofence parameter is no longer stuck in the short-term deviation of a single perception modality, but forms a correlated deviation between at least two perception modalities, the same trigger time window, and the same spatial abrupt location field. When determining whether the hazard level identifier corresponds to the emergency high-risk level, the hazard level identifier in the dynamic event fence extension trigger record is first read, and then matched with the hazard level mapping record in the dedicated risk judgment rule base to form an emergency high-risk level matching record. The hazard level mapping record is a pre-written correspondence of level fields in the dedicated risk judgment rule base, used to map candidate observation-level event tags and emergency high-risk event tags to their corresponding hazard level identifiers, enabling different graded early warning event sequences to use the same hazard level comparison criterion. Next, the thermal gradient distribution deviation record, gas concentration mutation deviation record, environmental vibration deviation record, spatial joint verification record, and temporal joint verification record in the multimodal evidence summary field are read, and these records are compared with the emergency high-risk level matching record for evidence verification to form a dynamic event fence extension access record. In this application, the dynamic event fence expansion processing mechanism is a set of data processing steps that adjust the event-driven range of the original designated geofence parameters and reconstruct the perception request command after the emergency high-risk level is confirmed. This mechanism does not simply expand the map display area, nor does it unconditionally broadcast requests externally. Instead, it takes the hazard level identifier, multimodal evidence summary field, spatial abrupt location field, and fence identifier field as inputs, and sequentially executes expansion triggering, radius rewriting, fence parameter repackaging, regional node matching, and perception request command issuance. After completing the dynamic event fence expansion access record, the record is written to the event expansion queue, ensuring that subsequent processing that modifies the initial coverage radius to the expanded coverage radius has a trigger source directly corresponding to the emergency high-risk level.

[0050] Preferably, the specific implementation process for determining the initial coverage radius and extended coverage radius corresponding to the specified geofence parameter through the dynamic event fence extension processing mechanism is as follows: After the dynamic event fence extension admission record has been written into the event extension queue, the specified geofence parameter is first read according to the fence identifier field, and the center point hash coordinates, preset coverage radius, buffer radius, fence type field, vehicle matching refresh field, and authorization level field are read from the specified geofence parameter. The center point hash coordinates, the preset coverage radius, the buffer radius, the fence type field, the vehicle matching refresh field, and the authorization level field are then written into the fence extension basic record. In this application, the initial coverage radius is a distance boundary parameter used by the specified geofence parameter to limit the core perception range before the dynamic event fence extension processing mechanism is triggered; the initial coverage radius can be derived from the preset coverage radius in the specified geofence parameter, or it can be derived from the extended coverage radius read at the current trigger time in the geofence parameter after the previous round of updates. To prevent the same fenced area from repeatedly expanding when the emergency high-risk level is triggered consecutively, the fence type field and vehicle matching refresh field in the fence expansion basic record are first read, and a fence expansion version record is established by combining the trigger time window field. The fence expansion version record is used to record the version order of the fence expansion processing mechanism for this dynamic event relative to the previous geofence parameter update. Subsequently, the spatial abrupt location field, spatial thermal credibility marker, multimodal evidence summary field, risk type field, and real-time meteorological condition feature parameters are read, and the above fields are written into the expansion radius determination record. In this application, the expansion coverage radius is an event-driven distance boundary parameter re-determined based on the initial coverage radius, according to the spatial abrupt location field, multimodal joint deviation degree, and risk type field corresponding to the emergency high-risk level. The expansion coverage radius is used to include adjacent road segments that may continue to show associated deviations within the spatial range of the next round of perception request instructions. When determining the extended coverage radius, the expansion direction can be determined first based on the risk type field. For example, for thermal anomalies, the expansion direction can be determined by combining the wind direction and road connectivity fields in the real-time meteorological condition characteristic parameters; for atmospheric vibration anomalies, the expansion direction can be determined by combining the spatial abrupt increase location field and adjacent spatial grid cells; and for multimodal anomalies, the expansion direction can be determined by simultaneously reading the spatial thermal migration field and the road segment hash identifier distribution. Then, based on the expansion direction, the expansion radius increment field is read from the radius expansion configuration record, and the boundary of the expansion radius increment field and the initial coverage radius is rewritten to form the extended coverage radius. The radius expansion configuration record can be pre-configured according to the fence type field, risk type field, road segment density, and available sensing distance of the vehicle-mounted physical sensor matrix, and is read and used when the dynamic event fence expansion processing mechanism is triggered.After the above processing, the extended coverage radius is not a fixed increase of the initial coverage radius, but is determined by the fence extension base record, the extension radius determination record, and the radius extension configuration record, and continues to participate in the generation of updated geofence parameters.

[0051] Preferably, the specific implementation process of modifying the initial coverage radius corresponding to the specified geofence parameter to the extended coverage radius through the dynamic event fence extension processing mechanism to obtain the updated geofence parameter is as follows: After forming the extended coverage radius, first read the fence identifier, center point hash coordinates, initial coverage radius, buffer radius, fence type field, vehicle matching refresh field, and authorization level field from the fence extension base record, and read the spatial burst location field, risk type field, trigger time window field, and extended coverage radius from the extended radius determination record, and write the above fields into the geofence parameter update carrier record. In this application, the updated geofence parameter is an event-driven spatial range parameter formed by rewriting the initial coverage radius to the extended coverage radius while retaining the original fence identifier, center point hash coordinates, authorization level field, and vehicle matching refresh field; the updated geofence parameter is not a newly created fence object unrelated to the specified geofence parameter, but rather spatial constraint data after versioning and rewriting the specified geofence parameter. In the specific modification, firstly, the initial coverage radius in the geofence parameter update record is written into the original radius field, and the extended coverage radius is written into the event extension radius field to form a radius rewriting record. This radius rewriting record is used to retain the correspondence between the initial coverage radius and the extended coverage radius, enabling subsequent cloud aggregation nodes to identify that the spatial range change originates from an emergency high-risk level. Then, based on the risk type field, the fence type field is rewritten to the event extension area, and the trigger time window field is written into the extension trigger time field. The spatial surge location field is written into the extension source location field to form a fence event extension record. This fence event extension record continues to undergo permission inheritance processing with the authorization level field, ensuring that the updated geofence parameters maintain the same permission verification standard as the digital certificate and the lowest response level certificate identifier when generating subsequent perception request commands. Then, the preheating range corresponding to the buffer radius is recalculated based on the extended coverage radius, and the recalculated buffer radius is written into the updated buffer radius field. This updated buffer radius field is used to distinguish the range corresponding to the event extension radius field and the range corresponding to the updated buffer radius field during subsequent regional node matching. Finally, the geofence parameter update record containing the fence identifier, center point hash coordinates, event extension radius field, updated buffer radius field, event extension area, extension trigger time field, extension source location field, authorization level field, and vehicle matching refresh field is processed by field aggregation and encapsulation to obtain the updated geofence parameters. The updated geofence parameters continue to serve as the spatial range field for the next round of perception request instructions and are used to subsequently filter all area nodes located within the extended coverage radius.

[0052] Preferably, the specific implementation process for determining all regional nodes within the extended coverage radius is as follows: After obtaining the updated geofence parameters, first read the fence identifier, center point hash coordinates, event extension radius field, updated buffer radius field, event extension area, vehicle matching refresh field, and authorization level field from the updated geofence parameters, and write the above fields into the regional node matching bearer record. In this application, the regional node is a communication node located within the spatial range defined by the updated geofence parameters and capable of receiving or forwarding perception request instructions; the regional node may include target vehicles within the extended coverage radius, candidate vehicles within the preheating range corresponding to the updated buffer radius field, roadside communication nodes capable of forwarding perception request instructions, and edge cache nodes capable of saving the perception request instruction issuance status. All regional nodes do not represent directly reading the real identity list, but rather represent all reachable communication nodes filtered through the anonymous location field, road segment hash identifier, and single-session-level random communication token in the anonymous vehicle location reporting record; wherein, the anonymous location field is formed by processing the physical location coordinates of the target vehicle through road segment attribution, and continues to participate in regional node matching through the road segment hash identifier. Specifically, the process begins by reading the road segment hash identifier and single-session-level random communication token from the anonymous vehicle location reporting record. These road segment hash identifiers and single-session-level random communication tokens are then compared with the center point hash coordinates in the regional node matching record under the same road segment index caliber for distance relationship reading, thus forming a regional node distance relationship record. Subsequently, the road segment hash identifier in the regional node distance relationship record is matched with the event expansion radius field to form node records within the expanded coverage radius. When a regional node distance relationship record indicates that the corresponding single-session-level random communication token is within the range corresponding to the event expansion radius field, this single-session-level random communication token is written into the core distribution field of the node record within the expanded coverage radius. When a regional node distance relationship record indicates that the corresponding single-session-level random communication token is within the updated buffer radius field but has not yet entered the range corresponding to the event expansion radius field, this single-session-level random communication token is written into the preheating distribution field of the node record within the expanded coverage radius. For roadside communication nodes and edge cache nodes, the updated geofence parameters associated with the road segment hash identifier can be read, and the node forwarding capability record can be read based on the road segment hash identifier. Then, the nodes that can cover the event extension area in the node forwarding capability record are written into the forwarding and distribution field of the node record within the extension coverage radius.The node records within the extended coverage radius continue to be used to issue perception request commands carrying the updated geofence parameters, enabling the perception request commands to simultaneously cover the target vehicles in the core issuance field, the candidate vehicles in the preheating issuance field, and the communication nodes in the forwarding issuance field, thereby expanding the acquisition range of the next round of multimodal sensor stream data after the emergency high-risk level is triggered.

[0053] Preferably, the specific implementation process of issuing a perception request instruction carrying the updated geofence parameters to all regional nodes within the extended coverage radius is as follows: After forming node records within the extended coverage radius, first read the fence identifier, center point hash coordinates, event extension radius field, updated buffer radius field, event extension area, authorization level field, vehicle matching refresh field, and extension trigger time field from the updated geofence parameters, and then read the core delivery field, preheating delivery field, and forwarding delivery field from the node records within the extended coverage radius. At the same time, read the danger level identifier, risk type field, trigger perception modality field, and evidence summary field from the previous round of graded early warning event sequence, and write the above fields into the extended perception request bearer record. In this application, the perception request instruction carrying the updated geofence parameters is structured request data repackaged with the updated geofence parameters after an emergency high-risk level is triggered. The perception request instruction carrying the updated geofence parameters is used to instruct all regional nodes within the extended coverage radius to perform the next round of perception matching, vehicle-mounted physical sensor matrix activation, target feature vector extraction, and desensitized perception data packet generation, according to the spatial range corresponding to the event extension area, the alarm-state perception interval parameter, and the target perception modality identifier. The alarm-state perception interval parameter is read from the alarm-state perception interval field in the perception frequency configuration record. This parameter is used to adjust the issuance rhythm of the perception request instruction carrying the updated geofence parameters after an emergency high-risk level is triggered, and to distinguish it from the normal perception interval parameter. Before the specific issuance, the issuance range of the target sensing modality identifier is determined based on the risk type field and the trigger sensing modality field. For example, the sensing request instruction corresponding to the thermal gas composite anomaly type should at least carry the target sensing modality identifier corresponding to the regional thermal gradient distribution attribute and the gas concentration mutation attribute; the sensing request instruction corresponding to the gas vibration composite anomaly type should at least carry the target sensing modality identifier corresponding to the gas concentration mutation attribute and the environmental vibration attribute; and the sensing request instruction corresponding to the multimodal composite anomaly type should carry the target sensing modality identifier corresponding to the regional thermal gradient distribution attribute, the gas concentration mutation attribute, and the environmental vibration attribute. Subsequently, the alarm state sensing interval parameter is read according to the hazard level identifier and written into the time interval field of the extended sensing request bearer record, so that all regional nodes within the extended coverage radius adopt the sampling time caliber corresponding to the emergency high-risk level within the event extension area. Then, the digital certificate, the minimum response level certificate identifier, the updated geofence parameters, the target sensing modality identifier, the alarm state sensing interval parameter, and the extended trigger time field are written into the extended sensing request bearer record, and the extended sensing request bearer record is encapsulated to generate a sensing request instruction carrying the updated geofence parameters.Finally, based on the node types of the core distribution field, preheating distribution field, and forwarding distribution field, the perception request instructions carrying the updated geofence parameters are grouped and distributed. Upon receiving the perception request instruction carrying the updated geofence parameters, the target vehicle in the core distribution field enters the permission verification and vehicle physical sensor matrix activation process. The candidate vehicle in the preheating distribution field enters the vehicle matching refresh and sensor preheating process. The roadside communication node or edge cache node in the forwarding distribution field saves the distribution status and continues to forward the perception request instruction carrying the updated geofence parameters to the corresponding single-session-level random communication token. Through the above processing, a continuous data processing relationship is formed between the updated geofence parameters, node records within the extended coverage radius, and the perception request instructions carrying the updated geofence parameters. This enables the emergency high-risk level to drive the generation of desensitized data packets over a larger spatial range and the construction of the next round of regional situational time-series profiles from cloud-aggregated nodes.

[0054] Optionally, the step of extracting a target feature vector from the multimodal sensor stream data in the local processing node of the target vehicle, generating a desensitized data packet based on the target feature vector, and transmitting the desensitized data packet to the cloud aggregation node further includes: monitoring a single abnormal amplitude feature value in the target feature vector; if the single abnormal amplitude feature value exceeds a preset extreme abnormal amplitude baseline, intercepting the instruction for the desensitized data packet to enter the regular aggregation waiting queue, and activating the independent direct connection reporting transmission link of the vehicle operating system; and sending an instant alarm data packet representing the single abnormal amplitude feature value to the cloud aggregation node through the independent direct connection reporting transmission link to trigger an early warning.

[0055] Preferably, the specific implementation process for monitoring the single abnormal amplitude feature value in the target feature vector is as follows: After the local processing node of the target vehicle has constructed the target feature vector based on the numerical sequence, the vector element arrangement caliber, sampling time caliber, target perception modality identifier, road segment hash identifier, and fence identifier in the specified geofence parameters of the target feature vector are read first, and the above fields are written into the single abnormal amplitude monitoring record. In this application, the target feature vector is a structured feature record formed by feature separation and extraction of regional thermal gradient distribution attributes, gas concentration abrupt change attributes, and environmental vibration attributes on the target vehicle. The target feature vector is not equivalent to the original image frame, the original gas sampling segment, or the original continuous acceleration segment, but is the common data source for subsequent desensitization perception data packet construction and rapid response of the single abnormal amplitude feature value. In this application, the single abnormal amplitude feature value is an abnormal amplitude characterization field formed by selecting the vector element with the highest deviation degree in the target feature vector. The single abnormal amplitude feature value is used to describe whether a certain perception modality has reached an extreme deviation state under the current sampling time caliber, which can be reported immediately without waiting for multi-vehicle aggregation. In specific monitoring, the target feature vector is first decomposed into three dimensions—thermal gradient distribution, gas concentration mutation, and environmental vibration—according to the vector element arrangement of the target sensing modality identifier. The thermal response deviation, thermal response region continuity, and temperature change are read from the thermal gradient distribution dimension; the volatile organic compound (VOC) concentration change, fine particulate matter (PM2.5) concentration change, carbon monoxide (CO) concentration change, and gas diffusion direction record are read from the gas concentration mutation dimension; and the three-dimensional vibration amplitude record, environmental vibration direction record, and environmental vibration duration record are read from the environmental vibration dimension. Subsequently, based on the sampling time, road segment hash identifier, and fence identifier, normal fluctuation boundary records corresponding to each sensing modality are read from the historical environmental benchmark parameter library. Each vector element in the target feature vector is then compared with a normal fluctuation boundary record of the same dimension to generate single-modality abnormal amplitude candidate records. These single-modality abnormal amplitude candidate records are further sorted by amplitude to determine the abnormal amplitude candidate field with the highest deviation. This abnormal amplitude candidate field is then written into the single abnormal amplitude monitoring record to form the single abnormal amplitude feature value. After the above processing, the single abnormal amplitude feature value is not an alarm field generated out of thin air, but is determined by the target feature vector, the historical environmental benchmark parameter library, the sampling time caliber and the road segment hash identifier, and continues to be used as an input field for comparison with the preset extreme abnormal amplitude benchmark.

[0056] Preferably, in the specific technical implementation of determining whether the single anomaly amplitude feature value exceeds the preset extreme anomaly amplitude baseline, the single anomaly amplitude feature value, target perception modality identifier, sampling time caliber, road segment hash identifier, and fence identifier in the single anomaly amplitude monitoring record are first read, and the extreme anomaly baseline configuration record corresponding to the specified geofence parameter is read according to the fence identifier. In this application, the preset extreme anomaly amplitude baseline is a modal-level extreme anomaly amplitude baseline field pre-written locally on the target vehicle or synchronously issued with the perception request command. The preset extreme anomaly amplitude baseline is used to distinguish between normal deviation state and extreme deviation state. The preset extreme anomaly amplitude baseline is not a fixed universal value, nor is it an alarm level text, but a data field that is associated with the fence identifier, target perception modality identifier, sampling time caliber, historical environmental benchmark parameter library, and dedicated risk judgment rule library. When configuring the preset extreme anomaly amplitude baseline, the basic baseline caliber can first be determined based on the fence type field of the area corresponding to the specified geofence parameters. Then, the basic baseline caliber can be corrected based on the current seasonal label, real-time meteorological condition characteristic parameters, and historical environmental baseline parameter library to form preset extreme anomaly amplitude baselines corresponding to thermal gradient distribution attributes, gas concentration mutation attributes, and environmental vibration attributes, respectively. The extreme anomaly baseline configuration record may include a fence identifier field, a target sensing mode identifier field, a baseline source field, a normal fluctuation boundary field, an extreme anomaly amplitude baseline field, and a baseline validity time field. The baseline source field indicates whether the preset extreme anomaly amplitude baseline originates from a locally pre-stored configuration or from the mapping result of a dedicated risk judgment rule library carried by the sensing request command. The baseline validity time field limits the time range within which the preset extreme anomaly amplitude baseline can participate in the comparison. In specific comparison, the preset extreme anomaly amplitude baseline corresponding to the same sensing mode is first read from the extreme anomaly baseline configuration record based on the target sensing mode identifier. The single anomaly amplitude feature value is then compared with the preset extreme anomaly amplitude baseline corresponding to the same sensing mode using the same dimensions to form an extreme anomaly amplitude comparison record. When the extreme anomaly amplitude comparison record indicates that the value of the single anomaly amplitude feature does not exceed the preset extreme anomaly amplitude baseline, the desensitized data packet continues to enter the regular aggregation waiting queue; when the extreme anomaly amplitude comparison record indicates that the value of the single anomaly amplitude feature exceeds the preset extreme anomaly amplitude baseline, the extreme anomaly amplitude comparison record writes a fast response trigger flag, and the fast response trigger flag continues to serve as the instruction to intercept the desensitized data packet from entering the regular aggregation waiting queue and as the triggering basis for activating the independent direct-connection reporting transmission link of the vehicle operating system.

[0057] Preferably, if the value of the single abnormal amplitude feature exceeds the preset extreme abnormal amplitude baseline, the specific implementation process of the instruction to intercept the desensitized data packet into the regular convergence waiting queue is as follows: After the extreme abnormal amplitude comparison record is written with the fast response trigger flag, the desensitized data packet bearer record that has completed field encapsulation processing is first read, and the target feature vector, single session-level random communication token, road segment hash identifier, sampling time caliber, target perception modality identifier, and fence identifier are read from the desensitized data packet bearer record. Then, the above fields and the fast response trigger flag are written into the fast response interception bearer record. In this application, the regular convergence waiting queue is a local queue used by the target vehicle to wait for batch transmission according to the normal perception interval parameter, the available bandwidth of the vehicle communication node, and the time slice caliber of the cloud convergence node after the target vehicle has completed the construction of the desensitized data packet; the regular convergence waiting queue corresponds to the multi-vehicle aggregation path of the desensitized data packet that has not triggered the fast response trigger flag, which enables the desensitized data packet to participate in the effective response total count, multimodal deviation statistics calculation, and spatial heat distribution map generation in the cloud convergence node together with other desensitized data packets in the area corresponding to the same specified geofence parameter. For cases where a single abnormal amplitude value exceeds a preset extreme abnormal amplitude baseline, if the desensitized data packet is still written into the regular aggregation waiting queue, the desensitized data packet needs to wait for other desensitized data packets within the same time slice to arrive at the cloud aggregation node before participating in spatiotemporal aggregation processing. This waiting process will prolong the reporting path for extreme deviation states. Specifically, during interception, the reporting queue status record in the vehicle operating system is first read. This record includes a regular aggregation waiting queue identifier, a queue write permission field, a fast response interception field, and a queue rollback field. Then, based on the fast response trigger flag in the fast response interception bearer record, the queue write permission field is rewritten from the regular write state to the fast response interception state to form a regular aggregation waiting queue interception record. This regular aggregation waiting queue interception record does not delete the desensitized data packet, but rather prevents it from entering the multi-vehicle aggregation path according to the waiting criteria of the regular aggregation waiting queue. The target feature vector, single-session-level random communication token, road segment hash identifier, and fence identifier in the desensitized data packet bearer record are rewritten as the entry point for constructing an instant alarm data packet. The queue backoff field is used to rewrite the desensitized data packet into the regular aggregation waiting queue when the independent direct-connection reporting transmission link of the vehicle operating system fails to be established within the effective time of the link, so as to avoid the loss of the desensitized data packet due to the link switching failure.Therefore, the instruction to intercept the desensitized data packets into the regular aggregation waiting queue is to change the transmission path of the desensitized data packets while retaining the target feature vector, single-session-level random communication token and road segment hash identifier, so that extreme anomaly amplitude comparison records can directly drive subsequent independent direct connection reporting processing.

[0058] Preferably, the specific implementation process of activating the independent direct-connection reporting transmission link of the vehicle operating system is as follows: After forming the conventional aggregation waiting queue interception record, the fast response trigger flag, fence identifier, target perception modality identifier, single-session-level random communication token, road segment hash identifier, and sampling time caliber in the conventional aggregation waiting queue interception record are read first, and the above fields are written into the independent direct-connection reporting transmission link activation bearer record. In this application, the independent direct-connection reporting transmission link of the vehicle operating system is an instant reporting transmission link established by the vehicle operating system outside the conventional aggregation waiting queue. The independent direct-connection reporting transmission link of the vehicle operating system is used to prioritize sending the instant alarm data packets corresponding to extreme deviation states to the cloud aggregation node. The independent direct-connection reporting transmission link of the vehicle operating system is not an open channel for uploading raw sensor data, nor is it an uncontrolled channel that bypasses digital certificates and specified geofence parameters. Instead, it is a controlled transmission link activated by the fast response trigger flag after the prior permission verification, perception restriction interception removal, target perception modality identifier matching, and desensitized perception data packet construction have all been completed. In the specific construction process, the link policy configuration record in the vehicle operating system is first read. This record includes a regular aggregation link identifier, an independent direct-connection reporting transmission link identifier, a link priority field, a retransmission count control field, a link validity time field, and a link destination address field. Then, based on the fence identifier and the target perception modality identifier, the independent direct-connection reporting transmission link identifier and the link destination address field corresponding to the cloud aggregation node are read from the link policy configuration record to form a candidate record for independent direct-connection reporting transmission links. Subsequently, the candidate record for independent direct-connection reporting transmission links is compared with the permission verification result corresponding to the digital certificate, the lowest response level certificate identifier, and the perception restriction removal record to form an access record for independent direct-connection reporting transmission links. This access record confirms that the current immediate report still falls within the scope of authorized perception request instructions and confirms that the target perception modality identifier is consistent with the target perception modality identifier written into the perception modality whitelist field. After completing the link access verification, the link priority field corresponding to the independent direct-connection reporting transmission link identifier is rewritten to the immediate reporting priority based on the independent direct-connection reporting transmission link access record. A temporary transmission session is then established for the single-session-level random communication token to form an independent direct-connection reporting transmission link activation record. This activation record continues to be used for field encapsulation, transmission destination location, and transmission status recording of immediate alarm data packets, enabling subsequent immediate alarm data packets to complete rapid reporting without carrying hardware identification codes and physical location coordinates.

[0059] Preferably, the specific implementation process of sending the real-time alarm data packet representing the single abnormal amplitude characteristic value to the cloud aggregation node through the independent direct connection reporting transmission link is as follows: After the independent direct connection reporting transmission link activation record is formed, the single abnormal amplitude characteristic value, target perception modality identifier, sampling time caliber, and road segment hash identifier in the single abnormal amplitude monitoring bearer record are read first, and the target feature vector, single session-level random communication token, and fence identifier in the desensitized knowledge data packet bearer record are read. Then, the above fields are written into the real-time alarm data packet bearer record. In this application, the real-time alarm data packet is a fast response transmission unit generated by the local processing node of the target vehicle to the cloud aggregation node after the single abnormal amplitude characteristic value exceeds the preset extreme abnormal amplitude baseline. The real-time alarm data packet is different from the desensitized knowledge data packet that enters the conventional spatiotemporal aggregation processing. The real-time alarm data packet retains the fields directly related to the extreme deviation state and reduces the queuing fields related to conventional aggregation waiting, so that the cloud aggregation node can prioritize the warning trigger judgment after receiving the real-time alarm data packet. In the specific encapsulation process, the single abnormal amplitude feature value is first written into the extreme abnormal amplitude field of the instant alarm data packet carrying record, and the preset extreme abnormal amplitude baseline is written into the baseline source field of the instant alarm data packet carrying record, so that the cloud aggregation node can identify the triggering basis of the instant alarm data packet. Subsequently, the target perception modality identifier is written into the trigger perception modality field of the instant alarm data packet carrying record, the vector element in the target feature vector corresponding to the single abnormal amplitude feature value is written into the instant feature summary field, the road segment hash identifier is written into the anonymous spatial index field, the single session-level random communication token is written into the anonymous communication source field, the sampling time caliber is written into the instant trigger time field, and the fence identifier is written into the fence affiliation field to form the instant alarm data packet. The instant alarm data packet does not carry hardware identification code, physical positioning coordinates, original image frames, original gas sampling segments, and original acceleration continuous segments. Instead, it expresses the extreme deviation state corresponding to the single abnormal amplitude feature value through the extreme abnormal amplitude field, the instant feature summary field, the anonymous spatial index field, and the anonymous communication source field. After the real-time alarm data packet is encapsulated, it is written into the real-time reporting buffer of the vehicle communication node according to the link destination address field in the independent direct-connection reporting transmission link activation record, and then sent to the cloud aggregation node through the vehicle communication node. The real-time reporting buffer is used to temporarily store the real-time alarm data packet before it is sent, and releases the corresponding buffer field after the vehicle communication node receives the reception confirmation status from the cloud aggregation node.During transmission, the vehicle-mounted communication node synchronously generates an instant reporting status record. The instant reporting status record records the transmission time, link identifier, retransmission status, and cloud aggregation node reception confirmation status of the instant alarm data packet. It is also used to trigger the queue backoff field when transmission fails, and rewrite the corresponding desensitized data packet into the regular aggregation waiting queue.

[0060] Preferably, in the specific technical implementation of triggering the early warning, after receiving the real-time alarm data packet, the cloud aggregation node first reads the fence attribution field, trigger perception modality field, extreme anomaly amplitude field, baseline source field, real-time feature summary field, anonymous spatial index field, anonymous communication source field, and real-time trigger time field from the real-time alarm data packet, and writes the above fields into the real-time early warning parsing bearer record. In this application, the triggering of the early warning is a priority determination process performed by the cloud aggregation node for a fast response path based on a single abnormal amplitude characteristic value. It is used to form a low-latency early warning event entry based on the extreme anomaly amplitude field in the real-time alarm data packet before the total number of effective responses, multimodal deviation statistics, and spatial heat map of the multi-vehicle aggregation path have been formed. In the specific determination process, the dedicated risk determination rule base is first read based on the fence attribution field, and then the instant warning determination record corresponding to the trigger perception modality field is read from the dedicated risk determination rule base. The instant warning determination record includes an extreme anomaly amplitude baseline field, an instant warning level field, a review time window field, and a subsequent aggregation association field. The review time window field is used to limit the time range within which the cloud aggregation node continues to receive desensitized perception data packets or instant alarm data packets under the same fence identifier after triggering the warning. The subsequent aggregation association field is used to include the instant alarm data packets in the feature surge record source of the next round of regional situation time-series profiling. Subsequently, the extreme anomaly amplitude field is checked against the extreme anomaly amplitude baseline field in the instant warning determination record for the same dimension, and the instant feature summary field is matched with the risk type field corresponding to the target perception modality identifier to form an instant warning verification record. When the instant warning verification record indicates that the extreme anomaly amplitude field is consistent with the baseline source field, and the instant feature summary field matches the instant warning determination record in the dedicated risk determination rule base, the cloud aggregation node generates a rapid response event label based on the instant warning level field and writes the rapid response event label into the candidate event carrier record of the hierarchical warning event sequence. The candidate event carrying record is used to temporarily store the rapid response event tags triggered by real-time alarm data packets, and continues to associate them with the regional situation time series profiles generated within the same review time window for event source association. The rapid response event tags do not replace the subsequent regional situation time series profiles, but rather serve as the real-time entry field for the graded early warning event sequence, continuing to be associated with the total number of subsequent effective responses, multimodal deviation statistics, and spatial heat map. When multiple desensitized data packets or multiple real-time alarm data packets appear within the same review time window, the cloud aggregation node continues to perform grade review of the rapid response event tags based on the regional situation time series profiles and the dedicated risk judgment rule base.Through the above processing, a continuous data processing relationship is formed between the single abnormal amplitude characteristic value, the preset extreme abnormal amplitude baseline, the regular convergence waiting queue, the independent direct connection reporting transmission link of the vehicle operating system, the real-time alarm data packet, and the cloud convergence node, so that the target vehicle can still provide early warning triggering basis to the cloud convergence node based on the desensitized field even when the number of responses is low or the deviation is extreme.

[0061] Optionally, before generating the graded early warning event sequence based on the regional situation time-series profile and the dedicated risk judgment rule base, the method further includes: obtaining the current seasonal label and real-time meteorological condition feature parameters of the region corresponding to the specified geofence parameters; querying a pre-built historical environmental benchmark parameter library based on the current seasonal label and the real-time meteorological condition feature parameters to obtain the corresponding environmental compensation coefficient; using the environmental compensation coefficient to modify the preset static anomaly judgment rules in the dedicated risk judgment rule base to obtain dynamic anomaly judgment rules; and writing the dynamic anomaly judgment rule into the dedicated risk judgment rule base to replace the static anomaly judgment rules.

[0062] Preferably, the specific implementation process for obtaining the current seasonal label of the area corresponding to the specified geofence parameter is as follows: Before generating the graded early warning event sequence based on the regional situation time series profile and the dedicated risk judgment rule base, the fence identifier, time slice sequence marker, trigger time window field in the regional situation time series profile, and the center point hash coordinates, preset coverage radius, and fence type fields in the specified geofence parameter are read first, and the fence identifier, the time slice sequence marker, the trigger time window field, the center point hash coordinates, the preset coverage radius, and the fence type field are written into the seasonal label parsing carrying record. In this application, the current seasonal label is an environmental index field that converts the seasonal environmental stage of the area corresponding to the specified geofence parameter in the current time slice into an environmental index field that can be read by the dedicated risk judgment rule base; the current seasonal label is not a simple date text, nor is it a display-oriented season name, but a rule index field used to limit the query scope of the historical environmental benchmark parameter base, the preset static anomaly judgment rule correction scope, and the dynamic anomaly judgment rule effective scope. Specifically, during the acquisition process, the current sampling date and current sampling period are first read based on the trigger time window field, and then written into the seasonal time determination record. Next, the regional seasonal segmentation configuration record associated with the area corresponding to the specified geofence parameter is read based on the center point hash coordinates and the preset coverage radius, and then written into the seasonal tag parsing carrying record. The regional seasonal segmentation configuration record can be pre-written during the specified geofence parameter generation stage, or synchronously written during the dedicated risk determination rule base establishment stage. The regional seasonal segmentation configuration record includes a fence identifier field, a regional climate zone field, a seasonal segment boundary field, a seasonal tag encoding field, and a seasonal tag validity period field. The seasonal segment boundary field represents the date range corresponding to different seasonal environmental stages, the seasonal tag encoding field converts the seasonal environmental stage into a queryable index field, and the seasonal tag validity period field limits the time range within which the current seasonal tag can participate in the historical environmental benchmark parameter database query. Subsequently, the current sampling date in the seasonal time determination record is matched with the seasonal segment boundary field in the regional seasonal segmentation configuration record to form a seasonal tag matching record; when the current sampling date falls into the corresponding seasonal segment boundary field, the current seasonal tag corresponding to the current sampling date is read from the seasonal tag encoding field, and the current seasonal tag is written into the seasonal tag parsing carrying record.The current season label continues to participate in the historical environmental benchmark parameter database query together with the real-time meteorological condition feature parameters. This enables the dedicated risk judgment rule database to distinguish the background impact of environmental stages such as high summer temperature, low winter temperature, humid season or dry season on regional thermal gradient distribution attributes, gas concentration mutation attributes and environmental vibration attributes before subsequent risk judgment. This avoids directly using the same preset static anomaly judgment rule, which could lead to environmental background differences being mistakenly identified as risk mutations.

[0063] Preferably, in the specific technical implementation of obtaining the real-time meteorological condition feature parameters of the area corresponding to the specified geofence parameters, the fence identifier, time slice sequence marker, spatial thermal distribution map, road segment hash identifier associated field, and center point hash coordinates and preset coverage radius in the regional situation time series profile are first read. The fence identifier, time slice sequence marker, spatial thermal distribution map, road segment hash identifier associated field, center point hash coordinates, and preset coverage radius are then written into the real-time meteorological condition parsing record. In this application, the real-time meteorological condition feature parameters are environmental parameter records that express the temperature, humidity, and air pressure states of the area corresponding to the specified geofence parameters within the current time slice, as well as the sampling time relationship corresponding to these states, in a field-based manner. The real-time meteorological condition feature parameters are not risk assessment results, nor are they a single weather description, but rather environmental input fields for subsequent queries of the historical environmental benchmark parameter library and correction of preset static anomaly judgment rules. Specifically, the meteorological sampling records are first read based on the fence identifier and the center point hash coordinates, and then the temperature, humidity, and air pressure sampling fields matching the current time slice are extracted from these records to form meteorological sampling candidate records. These records can originate from a pre-accessed regional meteorological data cache in the cloud aggregation node, or from structured feature fields related to environmental conditions in desensitized data packets uploaded by target vehicles within the fence. The regional meteorological data cache stores external environment sampling fields labeled according to the fence identifier and time slice order, while the structured feature fields store environmental condition fields extracted from multimodal sensor stream data by the target vehicle's local processing node. Regardless of whether the records originate from the regional meteorological data cache or the structured feature fields, the regional meteorological sampling records must establish a field correspondence with the fence identifier, time slice order label, and road segment hash identifier, ensuring that the real-time meteorological condition feature parameters correspond to the same region corresponding to the specified geofence parameter and the same current time slice. Subsequently, the meteorological sampling candidate records are time-aligned to remove meteorological sampling fields that do not belong to the current time slice. The remaining temperature, humidity, and air pressure sampling fields are then spatially assigned to form fence meteorological assignment records. These fence meteorological assignment records are further associated with spatial grid cell identifiers in the spatial thermal distribution map, enabling the real-time meteorological condition feature parameters to represent not only the overall meteorological background of the fence but also, when needed, the meteorological background corresponding to different spatial grid cells.The temperature, humidity, and air pressure sampling fields in the fence meteorological attribution record are then aggregated and encapsulated to generate the real-time meteorological condition feature parameters, which are then written into the real-time meteorological condition parsing record. After this processing, the real-time meteorological condition feature parameters can be queried in the historical environmental baseline parameter library along with the current season label. This allows the generation of dynamic anomaly judgment rules to not only rely on risk deviations in the regional situation time series profile but also to adjust the judgment boundaries of risk deviations based on the current environmental background.

[0064] Preferably, the specific implementation process of querying the pre-built historical environmental benchmark parameter library based on the current seasonal label and the real-time meteorological condition feature parameters is as follows: After the current seasonal label and the real-time meteorological condition feature parameters are written into the corresponding carrier records, the current seasonal label in the seasonal label parsing carrier record is read first, and the real-time meteorological condition feature parameters, fence identifier, time slice sequence marker, and road segment hash identifier association fields in the real-time meteorological condition parsing carrier record are read. The current seasonal label, the real-time meteorological condition feature parameters, the fence identifier, the time slice sequence marker, and the road segment hash identifier association fields are written into the historical environmental benchmark query carrier record. In this application, the historical environmental benchmark parameter library is a collection of historical environmental benchmark data organized according to fence identifiers, current season labels, meteorological condition segments, sampling periods, and sensing modal types. This historical environmental benchmark parameter library is not a regular log library, nor is it simply a record of historical alarms. Instead, it describes the normal fluctuation range of regional thermal gradient distribution attributes, gas concentration mutation attributes, and environmental vibration attributes in the area corresponding to the specified geofence parameters under the same or similar seasonal environmental stages, meteorological backgrounds, and sampling periods. When pre-constructing the historical environmental benchmark parameter library, historical target feature vectors, historical road segment hash identifiers, historical sampling time calibers, and historical fence identifiers can be read from historical desensitized sensing data packets. The regional thermal gradient distribution attributes, gas concentration mutation attributes, and environmental vibration attributes from the historical target feature vectors can then be written into thermal gradient historical benchmark records, gas concentration historical benchmark records, and environmental vibration historical benchmark records, respectively. These thermal gradient historical benchmark records, gas concentration historical benchmark records, and environmental vibration historical benchmark records are used to form thermal gradient environmental benchmark records, gas concentration environmental benchmark records, and environmental vibration environmental benchmark records in subsequent queries, respectively. Next, the historical seasonal labels and historical meteorological condition feature parameters corresponding to the historical sampling time caliber are read. The historical seasonal labels, historical meteorological condition feature parameters, historical road segment hash identifiers, and historical fence identifiers are written into the same historical environmental benchmark index record, enabling the historical environmental benchmark parameter database to be queried according to season, meteorology, time period, and spatial affiliation. Specifically, during the query, the fence benchmark record corresponding to the area corresponding to the specified geographic fence parameter is first read from the historical environmental benchmark parameter database based on the fence identifier. Then, seasonal benchmark records for the same seasonal environmental stage are filtered from the fence benchmark records based on the current seasonal label. Subsequently, the temperature sampling field, humidity sampling field, and air pressure sampling field in the real-time meteorological condition feature parameters are matched with the meteorological condition segmentation field in the seasonal benchmark record to form a meteorological condition matching record.The meteorological condition matching records are then jointly queried with the time slice sequence marker and road segment hash identifier fields to obtain thermal gradient environmental benchmark records, gas concentration environmental benchmark records, and environmental vibration environmental benchmark records. These records continue to serve as the benchmark sources for calculating environmental compensation coefficients, ensuring that the environmental compensation coefficients are established around the same specified geofence parameter corresponding area, the same current season label, and the same real-time meteorological condition characteristic parameters, rather than using a generic correction caliber unrelated to the current environmental background.

[0065] Preferably, the specific implementation process for obtaining the corresponding environmental compensation coefficient is as follows: After obtaining the thermal gradient environmental benchmark record, gas concentration environmental benchmark record, and environmental vibration environmental benchmark record from the historical environmental benchmark parameter library, the temperature sampling field, humidity sampling field, and air pressure sampling field in the real-time meteorological condition characteristic parameters are read first, and the normal fluctuation boundary of the thermal gradient in the thermal gradient environmental benchmark record, the normal fluctuation boundary of the gas concentration in the gas concentration environmental benchmark record, and the normal fluctuation boundary of the environmental vibration in the environmental vibration environmental benchmark record are read respectively. The temperature sampling field, the humidity sampling field, the air pressure sampling field, the normal fluctuation boundary of the thermal gradient, the normal fluctuation boundary of the gas concentration, and the normal fluctuation boundary of the environmental vibration are written into the environmental compensation coefficient calculation bearing record. In this application, the environmental compensation coefficient is a set of coefficient fields used to correct the anomaly judgment boundaries of each sensing mode in the preset static anomaly judgment rules under the current seasonal label and real-time meteorological condition characteristic parameters. The environmental compensation coefficient does not directly represent the risk level, nor is it a simple superposition of temperature, humidity, and air pressure into a single value. Instead, it includes thermal gradient compensation fields, gas concentration compensation fields, and environmental vibration compensation fields, and each compensation field is only used to correct the anomaly judgment boundary of its corresponding sensing mode. Specifically, during generation, the temperature sampling field is first subjected to deviation reading processing with the thermal gradient meteorological reference field in the thermal gradient environmental benchmark record to form a thermal gradient meteorological deviation record. The thermal gradient meteorological deviation record is used to describe the degree of deviation of the current temperature background from the temperature background of the same season and time period in the historical environmental benchmark parameter library, and continues to participate in the generation of the thermal gradient compensation field. Subsequently, the humidity sampling field and air pressure sampling field are subjected to deviation reading processing with the gas diffusion meteorological reference field in the gas concentration environmental benchmark record to form a gas concentration meteorological deviation record. This gas concentration meteorological deviation record describes the influence direction of the current humidity and air pressure states on the gas concentration abrupt change attribute benchmark boundary and continues to participate in the generation of the gas concentration compensation field. Next, the temperature sampling field, humidity sampling field, and air pressure sampling field are subjected to deviation reading processing with the environmental vibration meteorological reference field in the environmental vibration environmental benchmark record to form an environmental vibration meteorological deviation record. This environmental vibration meteorological deviation record describes the influence relationship of the current meteorological background on road surface vibration transmission, vehicle vibration response, and the normal fluctuation range of environmental vibration attributes, and continues to participate in the generation of the environmental vibration compensation field. Then, the thermal gradient meteorological deviation record, the gas concentration meteorological deviation record, and the environmental vibration meteorological deviation record are written into the environmental compensation coefficient carrying record according to the sensing mode source, and the thermal gradient compensation field, the gas concentration compensation field, and the environmental vibration compensation field are formed in the environmental compensation coefficient carrying record to generate the environmental compensation coefficient.The environmental compensation coefficient continues to participate in the preset static anomaly judgment rule correction, so that the subsequent dynamic anomaly judgment rule can update the anomaly judgment boundary under the same dimension, the same mode and the same fence caliber, reducing the interference of seasonal temperature rise, humidity change or air pressure change on the judgment results of regional thermal gradient distribution attributes, gas concentration abrupt change attributes and environmental vibration attributes.

[0066] Preferably, the specific implementation process of using the environmental compensation coefficient to modify the preset static anomaly judgment rules in the dedicated risk judgment rule base to obtain dynamic anomaly judgment rules is as follows: After the environmental compensation coefficient is formed, the environmental compensation coefficient is first read from the environmental compensation coefficient carrying record, and the rule base record corresponding to the area corresponding to the specified geofence parameter is read from the dedicated risk judgment rule base according to the fence identifier. Then, the preset static anomaly judgment rule is read from the rule base record, and the preset static anomaly judgment rule and the environmental compensation coefficient are written into the anomaly judgment rule modification carrying record. In this application, the preset static anomaly judgment rule is a combination of anomaly judgment boundaries pre-configured during the establishment stage of the dedicated risk judgment rule base. It includes the regional thermal gradient anomaly static boundary corresponding to the regional thermal gradient distribution attribute, the gas concentration anomaly static boundary corresponding to the gas concentration mutation attribute, the environmental vibration anomaly static boundary corresponding to the environmental vibration attribute, and the trigger condition label matching caliber corresponding to the single-modal feature independent deviation condition and the multi-modal feature joint deviation condition. The preset static anomaly determination rules can be pre-configured based on the fence type field, risk type field, thermal gradient environmental benchmark records, gas concentration environmental benchmark records, and environmental vibration environmental benchmark records in the historical environmental benchmark parameter library, as well as the target sensing modal identifier. These rules serve as the initial anomaly determination boundary before being combined with the current seasonal label and real-time meteorological condition characteristic parameters. Specifically, during correction, the thermal gradient compensation field in the environmental compensation coefficient is first read, and the regional thermal gradient anomaly static boundary is corrected using the same modal boundary based on this field to form the regional thermal gradient anomaly dynamic boundary. Then, the gas concentration compensation field in the environmental compensation coefficient is read, and the gas concentration anomaly static boundary is corrected using the same modal boundary based on this field to form the gas concentration anomaly dynamic boundary. Finally, the environmental vibration compensation field in the environmental compensation coefficient is read, and the environmental vibration anomaly static boundary is corrected using the same modal boundary based on this field to form the environmental vibration anomaly dynamic boundary. All of the above corrections are performed within the same sensing modality, without directly mixing the different dimensional values ​​of the thermal gradient distribution attribute, gas concentration abrupt change attribute, and environmental vibration attribute in the calculation. Subsequently, the dynamic boundary of the regional thermal gradient anomaly, the dynamic boundary of the gas concentration anomaly, and the dynamic boundary of the environmental vibration anomaly are associated with trigger condition labels, single-modal feature independent deviation conditions, multi-modal feature joint deviation conditions, current season labels, real-time meteorological condition feature parameters, and time slice sequence markers to generate dynamic anomaly judgment rules.In this application, the dynamic anomaly determination rule is based on the rule structure and risk type field of the preset static anomaly determination rule. It is formed by using environmental compensation coefficients to correct the anomaly determination boundaries of each perception mode and then creating a valid rule record for the current time slice. The dynamic anomaly determination rule is further used to extract feature surge records from the regional situation time series profile, match trigger condition labels, and generate graded early warning event sequences, so that risk determination can adapt to the current seasonal label and real-time meteorological condition feature parameters.

[0067] Preferably, in the specific technical implementation of writing the dynamic anomaly determination rules into the dedicated risk determination rule library to replace the preset static anomaly determination rules, the following are first read from the dynamic anomaly determination rules: regional thermal gradient anomaly dynamic boundary, gas concentration anomaly dynamic boundary, environmental vibration anomaly dynamic boundary, trigger condition label, single-modal feature independent deviation condition, multi-modal feature joint deviation condition, current season label, real-time meteorological condition feature parameter, and time slice sequence mark. Then, the regional thermal gradient anomaly dynamic boundary, the gas concentration anomaly dynamic boundary, the environmental vibration anomaly dynamic boundary, the trigger condition label, the single-modal feature independent deviation condition, the multi-modal feature joint deviation condition, the current season label, the real-time meteorological condition feature parameter, and the time slice sequence mark are written into the rule library to replace the carrying record. The step of writing the dynamic anomaly determination rule into the dedicated risk determination rule library to replace the preset static anomaly determination rule does not involve deleting the original preset static anomaly determination rule from the dedicated risk determination rule library. Instead, within the determination period corresponding to the current fence identifier, current season tag, current time slice, and current real-time meteorological condition feature parameter, the dynamic anomaly determination rule is set as the currently effective rule with priority reading, while the preset static anomaly determination rule is retained as a rule rollback field and a rule source field. Specifically, during the writing process, the corresponding rule library record is first located from the dedicated risk determination rule library based on the fence identifier, and a dynamic rule version field, a rule effective time field, a rule expiration time field, a rule source field, and a rule rollback field are established in the rule library record. Then, the dynamic anomaly determination rule is written into the dynamic rule version field, the time slice sequence mark is written into the rule effective time field, the next rule update trigger time or the end time of the determination period corresponding to the current time slice is written into the rule expiration time field, the current season tag and the real-time meteorological condition feature parameter are written into the rule source field, and the preset static anomaly determination rule is written into the rule rollback field. Subsequently, the trigger condition label matching entry in the dedicated risk assessment rule base is read, and the trigger condition label matching entry is rewritten from reading preset static anomaly assessment rules to prioritizing reading dynamic anomaly assessment rules, thus forming a risk assessment entry rewrite record. This risk assessment entry rewrite record continues to participate in subsequent feature surge record matching processing, enabling the cloud aggregation node to first read the dynamic boundaries of regional thermal gradient anomalies, gas concentration anomalies, and environmental vibration anomalies according to dynamic anomaly assessment rules before generating a tiered early warning event sequence, and then determine whether the trigger condition label meets the independent deviation condition of a single-modal feature or the joint deviation condition of multiple-modal features.After the writing is completed, a consistency check is performed on the rule base replacement record to confirm that the fence identifier, current season label, real-time meteorological condition feature parameters, and time slice sequence marker in the dynamic anomaly judgment rule are consistent with the current regional situation time series profile. When the consistency check passes, the dynamic anomaly judgment rule is used as the currently effective rule to participate in the generation of graded early warning event sequences. Through the above processing, a continuous data processing relationship is formed between the current season label, the real-time meteorological condition feature parameters, the historical environmental benchmark parameter library, the environmental compensation coefficient, the preset static anomaly judgment rule, and the dynamic anomaly judgment rule, enabling the dedicated risk judgment rule base to update the anomaly judgment boundary according to changes in the environmental background under the same fence scenario.

[0068] like Figure 3 As shown in the figure, an embodiment of this application discloses a region-request-based vehicle multimodal perimeter perception and risk warning device, comprising: a request parsing module, used to receive a perception request instruction containing specified geofence parameters and a digital certificate, and extract a target perception modality identifier and a minimum response level certificate identifier from the perception request instruction; a sensing activation module, used to perform permission verification on the digital certificate, and when it is determined that the permission of the digital certificate meets the minimum response level certificate identifier, to activate the on-board physical sensor matrix corresponding to the target perception modality identifier on the target vehicle to acquire multimodal sensor stream data; and a local processing module, used for local processing on the target vehicle. In the node, a target feature vector is extracted from the multimodal sensor stream data, and a desensitized data packet is generated based on the target feature vector. The desensitized data packet is then transmitted to the cloud aggregation node. The cloud aggregation module is used at the cloud aggregation node to perform spatiotemporal aggregation processing on multiple desensitized data packets from the area corresponding to the specified geofence parameter, generating a regional situation time series profile. The early warning distribution module is used to generate a corresponding graded early warning event sequence based on the regional situation time series profile and a pre-configured exclusive risk judgment rule base corresponding to the specified geofence parameter, and distribute the graded early warning event sequence to a preset receiving endpoint.

[0069] like Figure 4 As shown, an electronic device according to an embodiment of this application includes a memory and a processor. Figure 5 As shown, this is a computer-readable storage medium according to an embodiment of this application. Figure 6 As shown, this is a vehicle according to an embodiment of the present application, which includes an onboard physical sensor matrix, a local processing node, and an onboard communication node. Figure 7 As shown in the figure, this is an embodiment of a regional request-based vehicle multimodal perimeter perception and risk warning system, which includes a perception request node, a target vehicle, a cloud aggregation node, and a preset receiving endpoint.

[0070] The above Figures 3-7 For an exemplary description, please refer to the above. Figure 1 This will not be elaborated upon here.

Claims

1. A region-request-based multimodal vehicle perimeter perception and risk warning method, characterized in that, include: Receive a perception request instruction containing specified geofence parameters and a digital certificate, and extract the target perception modality identifier and the lowest response level certificate identifier from the perception request instruction; The digital certificate is validated for permissions, and when it is determined that the permissions of the digital certificate meet the minimum response level certificate identifier, the on-board physical sensor matrix corresponding to the target perception modality identifier is activated on the target vehicle to acquire multimodal sensor stream data. In the local processing node of the target vehicle, the target feature vector is extracted from the multimodal sensor stream data, and a desensitized data packet is generated based on the target feature vector. The desensitized data packet is then transmitted to the cloud aggregation node. At the cloud aggregation node, spatiotemporal aggregation processing is performed on multiple desensitized data packets from the area corresponding to the specified geofence parameter to generate a temporal profile of the regional situation. Based on the regional situation time-series profile and the pre-configured exclusive risk assessment rule library corresponding to the specified geofence parameters, a corresponding graded early warning event sequence is generated and distributed to a preset receiving endpoint.

2. The method according to claim 1, characterized in that, Before receiving the perception request instruction, the method also includes: Obtain the center point hash coordinates and preset coverage radius of the initial monitoring area, and obtain the pre-configured normal perception interval parameter, which is used to set the normal perception interval of the perception request command. A mobile tracking fence is established based on the center point hash coordinates and the preset coverage radius, and the specified geofence parameters are generated. The sensing request instruction is generated based on the specified geofence parameters, the normal sensing interval parameters, and the digital certificate.

3. The method according to claim 1, characterized in that, The step of verifying the permissions of the digital certificate, and when it is determined that the permissions of the digital certificate meet the minimum response level certificate identifier, activating the on-board physical sensor matrix corresponding to the target perception modality identifier on the target vehicle side to acquire multimodal sensor stream data, includes: Parse the Certificate Issuance Level (CRE) tag of the digital certificate; The certificate issuance level label is compared with the lowest response level certificate identifier pre-stored locally in the target vehicle to determine whether the permission priority of the certificate issuance level label reaches the permission priority of the lowest response level certificate identifier. If the permission priority of the certificate issuance level label reaches the permission priority of the lowest response level certificate identifier, then the perception restriction interception of the target vehicle's in-vehicle operating system is lifted. After the perception restriction interception is lifted, the corresponding vehicle physical sensor matrix is ​​activated according to the target perception modality identifier to obtain the multimodal sensor stream data.

4. The method according to claim 1, characterized in that, The step of extracting a target feature vector from the multimodal sensor stream data in the local processing node of the target vehicle, and generating a desensitized data packet based on the target feature vector, includes: Feature separation and extraction are performed on the multimodal sensing flow data to obtain a numerical sequence reflecting the regional thermal gradient distribution attributes, gas concentration abrupt change attributes, and environmental vibration attributes. The target feature vector is constructed based on the numerical sequence. The hardware identification code of the target vehicle is subjected to privacy isolation processing, and the hardware identification code is replaced with a single-session-level random communication token; Map the physical location coordinates of the target vehicle to a road segment hash identifier; The desensitized data packet is constructed based on the target feature vector, the single-session-level random communication token, and the road segment hash identifier.

5. The method according to claim 1, characterized in that, At the cloud aggregation node, spatiotemporal aggregation processing is performed on multiple desensitized data packets from the area corresponding to the specified geofence parameters to generate a regional situation time-series profile, including: Within a time slice of preset duration, count the total number of valid responses to the desensitized data packets carrying the same specified geofence parameter; Extract the target feature vector from each of the desensitized data packets, and calculate the multimodal bias statistic of the target feature vector; The target feature vectors are spatially distributed and mapped according to the grid division rules to generate a spatial thermal distribution map. The total number of effective responses, the multimodal deviation statistics, and the spatial thermal distribution map are correlated in chronological order to construct a temporal profile of the regional situation.

6. A region-request-based vehicle multimodal perimeter perception and risk warning device, characterized in that, include: The request parsing module is used to receive a perception request instruction containing specified geofence parameters and a digital certificate, and extract the target perception modality identifier and the lowest response level certificate identifier from the perception request instruction; The sensing activation module is used to verify the permissions of the digital certificate, and when it is determined that the permissions of the digital certificate meet the minimum response level certificate identifier, it activates the on-board physical sensor matrix corresponding to the target perception modality identifier on the target vehicle end to acquire multimodal sensing stream data. The local processing module is used to extract target feature vectors from the multimodal sensor stream data in the local processing node of the target vehicle, generate desensitized data packets based on the target feature vectors, and transmit the desensitized data packets to the cloud aggregation node. The cloud aggregation module is used at the cloud aggregation node to perform spatiotemporal aggregation processing on multiple desensitized data packets from the area corresponding to the specified geofence parameter, and generate a regional situation time series profile. The early warning distribution module is used to generate a corresponding graded early warning event sequence based on the regional situation time-series profile and the pre-configured exclusive risk judgment rule library corresponding to the specified geofence parameters, and distribute the graded early warning event sequence to a preset receiving endpoint.

7. An electronic device, characterized in that, The system includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the area-request vehicle multimodal perimeter perception and risk warning method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the area-request vehicle multimodal perimeter perception and risk warning method according to any one of claims 1 to 5.

9. A vehicle, characterized in that, This includes an onboard physical sensor matrix, local processing nodes, and onboard communication nodes. The vehicle-mounted communication node is used to receive a perception request instruction containing specified geofence parameters and a digital certificate, and to extract the target perception modality identifier and the lowest response level certificate identifier from the perception request instruction. The local processing node is used to verify the permissions of the digital certificate, and when it is determined that the permissions of the digital certificate meet the minimum response level certificate identifier, it activates the vehicle physical sensor matrix corresponding to the target perception modality identifier. The on-board physical sensor matrix is ​​used to acquire multimodal sensor stream data; The local processing node is also used to extract target feature vectors from the multimodal sensor stream data and generate desensitized data packets based on the target feature vectors; The vehicle-mounted communication node is also used to transmit the desensitized data packets to the cloud aggregation node, so that the cloud aggregation node performs spatiotemporal aggregation processing on multiple desensitized data packets from the area corresponding to the specified geofence parameter, generates a regional situation time series profile, and generates a corresponding graded early warning event sequence based on the regional situation time series profile and the pre-configured exclusive risk judgment rule library corresponding to the specified geofence parameter.

10. A region-request-based multimodal vehicle perimeter perception and risk warning system, characterized in that, This includes the sensing request node, the target vehicle, the cloud aggregation node, and the preset receiving endpoint; The perception request node is used to generate a perception request instruction containing specified geofence parameters and a digital certificate, and send the perception request instruction to the target vehicle; The target vehicle is used to receive the perception request instruction, extract the target perception modality identifier and the minimum response level certificate identifier from the perception request instruction, perform permission verification on the digital certificate, and when it is determined that the permission of the digital certificate meets the minimum response level certificate identifier, activate the on-board physical sensor matrix corresponding to the target perception modality identifier on the target vehicle end to acquire multimodal sensor stream data; The target vehicle is also used to extract target feature vectors from the multimodal sensor stream data in the local processing node, generate desensitized data packets based on the target feature vectors, and transmit the desensitized data packets to the cloud aggregation node. The cloud aggregation node is used to perform spatiotemporal aggregation processing on multiple desensitized data packets from the area corresponding to the specified geofence parameter to generate a temporal profile of the regional situation. The cloud aggregation node is also used to generate a corresponding graded early warning event sequence based on the regional situation time-series profile and the pre-configured exclusive risk judgment rule library corresponding to the specified geofence parameters, and to distribute the graded early warning event sequence to the preset receiving endpoint. The preset receiving endpoint is used to receive the graded early warning event sequence.

Images