Efficient vehicle monitoring system based on vehicle-road-cloud cooperation

Abstract

The invention belongs to the technical field of vehicle monitoring, and discloses an efficient vehicle monitoring system based on vehicle-road-cloud collaboration, which constructs a geofence capable of self-adaptive adjustment based on V2X communication and differential GPS through a V2X geofence dynamic sensing module, can change the range in real time according to time periods, vehicle density and the like, and improves the vehicle monitoring efficiency. If the vehicle contracts to a key area late night, queuing vehicles are covered in congestion; meanwhile, the vehicle-road-cloud resource dynamic scheduling module integrates roadside fixed equipment and surrounding vehicle sensors, and brings moving vehicles into a sensing network, so that the problems of weak dynamic sensing and monitoring blind areas caused by the fact that a traditional system depends on the fixed equipment are solved; a multi-modal event verification module fuses vehicle-mounted radar tracks, camera behavior features, IMU attitude data, roadside sound waves, panoramic pictures and other multi-source information, key features are extracted after time-space synchronization data alignment, and then cross validation is performed through a deep belief network, so that the accuracy and reliability of event qualitation are greatly improved.

Description

Technical Field

[0001] This invention belongs to the field of vehicle monitoring technology, specifically a high-efficiency vehicle monitoring system based on vehicle-road-cloud collaboration. Background Technology

[0002] Existing vehicle monitoring systems suffer from three major defects: information isolation, rigid scheduling, and crude response. Information collection relies excessively on fixed roadside equipment, and dynamic perception capabilities are weak.

[0003] A search revealed that the invention patent CN119155649A discloses signal transmission control for vehicle-to-everything (V2X) communication. Although it achieves basic communication between vehicles and roadside infrastructure through V2X, it does not incorporate surrounding vehicles into the mobile sensing node system. When roadside cameras are obstructed or vehicles are offline, blind spots can easily form in scenarios such as curves in logistics parks and congested urban roads. Furthermore, it lacks a dynamic vehicle-road-cloud collaborative mechanism, making it impossible to select the optimal monitoring node based on the real-time location of vehicles. During peak hours, equipment load imbalances frequently occur (some cameras overload and fail, while others remain idle). Relying on data from a single device to determine events can easily lead to misjudgments when there is insufficient light or obstruction (e.g., misjudging a normal stop as loitering). A search revealed that the invention patent with announcement number CN120179045A discloses a GPU computing power dynamic scheduling system, which focuses on GPU computing power scheduling but does not integrate data from multiple sources of sensors on vehicles and roadsides, resulting in insufficient dynamic perception accuracy in complex scenarios; it only optimizes a single device cluster and does not coordinate the resources of roadside smart streetlights and surrounding vehicle sensors, which can easily lead to overlapping or omissions in monitoring coverage when multiple vehicles are in parallel.

[0004] Neither of them has established a threat classification mechanism, and they adopt a unified response to high- and low-risk events. This results in both excessive alarms that waste emergency resources and difficulties in efficiently handling high-risk behaviors such as violent sabotage, thus restricting the intelligence and practicality of the monitoring system. Summary of the Invention

[0005] The purpose of this invention is to provide a high-efficiency vehicle monitoring system based on vehicle-road-cloud collaboration to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a high-efficiency vehicle monitoring system based on vehicle-road-cloud collaboration, the system comprising: V2X geofence dynamic sensing module: Based on V2X communication and differential GPS, a dynamic geofence is constructed. The range can be adjusted according to time period, event, vehicle density and trajectory. When a vehicle triggers an entry or exit event, a signal is immediately pushed to the vehicle-road-cloud resource dynamic scheduling module to start monitoring and provide precise spatial triggering conditions for the vehicle-road-cloud resource dynamic scheduling module. Vehicle-Road-Cloud Resource Dynamic Scheduling Module: After receiving the signal from the geofence module, it integrates the real-time status of fixed equipment such as roadside smart streetlights and panoramic cameras with the surrounding vehicles, selects the optimal monitoring combination through an improved greedy algorithm, and transmits the data to the multimodal event verification module. Multimodal event verification module: Based on data from vehicle-mounted radar, cameras, IMU, roadside acoustic sensors, and panoramic cameras transmitted from the vehicle-road-cloud resource dynamic scheduling module, the data is spatiotemporally synchronized and feature fused, and the events are cross-verified through a deep belief network. The results are then output to the threat level intelligent assessment module to solve the problem of misjudgment by a single device. Threat Level Intelligent Assessment Module: Receives event verification results, integrates environmental factors such as pedestrian density and time period weighting from roadside cameras, and historical event data of the area, quantifies the threat level using a fuzzy comprehensive evaluation model, and pushes it to the graded response execution module; Tiered Response Execution Module: Based on the results of the threat level module, a three-level response is initiated: low risk triggers vehicle-mounted warnings and roadside light alerts; medium risk pushes alarm packets and dispatches patrol vehicles; high risk links with the public security department to trigger audible and visual alarms and vehicle locking commands. The commands are transmitted via the vehicle-road-cloud collaborative communication module to achieve precise action. Vehicle-Road-Cloud Collaborative Communication Module: Supports the transmission of trigger signals, monitoring data, threat levels, response commands, etc. between various modules. It adopts 5G+C-V2X dual links, and ensures stability through intelligent switching and breakpoint resume transmission, connecting the information flow of each link. System Configuration and Management Module: Configures geofencing rules, scheduling algorithm parameters, and validation model weights for other modules, provides scenario templates for logistics parks, monitors equipment and link status, and provides parameter support and operation and maintenance assurance for efficient system operation.

[0007] This invention integrates mobile vehicles into the sensing network through the linkage of the V2X geofence dynamic perception module and the vehicle-road-cloud resource dynamic scheduling module, making up for the insufficient coverage of fixed equipment; the multimodal event verification module integrates multi-source data cross-verification to solve the problem of misjudgment by a single device; the threat level intelligent assessment module and the hierarchical response execution module work together to achieve differentiated resource scheduling and overcome the shortcomings of the traditional system's coarse response.

[0008] Preferably, the V2X geofence dynamic sensing module includes: (1) Dynamic fence construction logic: Based on V2X communication and differential GPS positioning technology, dynamic geofences are constructed in combination with cloud GIS system. Through adaptive adjustment of time dimension (time period characteristics, sudden events) and spatial dimension (vehicle density, driving trajectory), the fence range can match the monitoring needs in real time—for example, shrinking to key areas late at night and extending to cover queuing vehicles when congested. (2) Transmission and connection of trigger signals: When a vehicle triggers the "enter or leave" fence event, the module immediately generates a standardized trigger signal and pushes it accurately to the vehicle-road-cloud resource dynamic scheduling module to start the subsequent monitoring process. This is a link design of "vehicle dynamic behavior → trigger signal → scheduling start".

[0009] Preferably, the vehicle-road-cloud resource dynamic scheduling module includes: (1) Multi-dimensional consideration of resource integration: After receiving the trigger signal from the geofence module, the module enters the resource integration stage: On the one hand, it collects parameters such as the viewing angle coverage and network latency of roadside fixed equipment (smart street lights, panoramic cameras); on the other hand, it obtains the status of vehicle-mounted sensors of surrounding vehicles through V2V communication, such as camera angle and IMU data availability, to provide comprehensive data support for scheduling decisions. (2) Link connection of scheduling results: Based on the improved greedy optimization algorithm, the optimal combination of monitoring nodes is selected with the goal of "complete coverage, efficient transmission and balanced load". For example, in the curve scenario, the roadside camera and the front-view camera of the vehicle behind are coordinated. After the scheduling is completed, the multi-source data acquisition channel is directed to the multi-modal event verification module to form a closed loop of trigger signal → resource allocation → data acquisition.

[0010] Preferably, the multimodal event verification module includes: (1) Multi-source data fusion mechanism: Based on the multi-source data allocated by the scheduling module, such as the trajectory speed of the vehicle millimeter-wave radar, the behavioral characteristics of the visual camera, the attitude data of the IMU, the ambient sound waves of the roadside acoustic wave sensor, and the scene images of the panoramic camera, the high and low frequency sampling data are aligned through the spatiotemporal synchronization algorithm to extract key features such as "trajectory shape, action mode, and sound wave frequency". Multimodal feature fusion adopts a weighted summation formula: Comprehensive feature value = trajectory morphology feature × 0.4 + action pattern feature × 0.3 + sound wave frequency feature × 0.3 (weights can be dynamically adjusted through the system configuration and management module), where each feature value is normalized (range 0-1) before participating in the calculation; Feature anomaly detection threshold: Trajectory pattern: A dwell time > 5 minutes and a velocity change rate < 0.5 m / s² is judged as 'abnormal dwell'; Action mode: Steering angle > 90° and no brake signal is judged as 'abnormal steering'; Sound wave frequency: Abnormal sound peak >80dB and not in the engine frequency range (20-2000Hz) is judged as 'abnormal sound'.

[0011] Multimodal feature fusion weights: trajectory shape (0.4), action pattern (0.3), sound frequency (0.3); the weights can be dynamically adjusted through the system configuration and management module, such as increasing the sound frequency weight to 0.4 (default 0.3) in urban arterial road mode to adapt to complex acoustic environments.

[0012] (2) Decision connection of verification results: The extracted features are input into the deep belief network (DBN) for cross-validation. For example, the wandering behavior must simultaneously meet the consistency of radar trajectory, camera attitude and sound wave features to accurately determine the event type. The verification results (including event feature data) will be output to the threat level intelligent assessment module to solve the problem of misjudgment caused by lighting and occlusion of a single device and complete the link connection of resource collection → event characterization.

[0013] Preferably, the threat level intelligent assessment module includes: (1) Input logic of grade quantification: After receiving the qualitative results of the event verification module, start multi-factor fusion analysis: On the one hand, integrate environmental data, such as the pedestrian density and time period risk weight of roadside cameras; on the other hand, call the historical database, such as the consequence correlation characteristics of similar events in the region, and transform the abstract event into quantifiable threat parameters through the fuzzy comprehensive evaluation model, such as the weighted calculation of the lingering time and pedestrian density. Historical database fields include: event type (e.g., 'abnormal parking', 'speeding'), occurrence time, pedestrian density, duration, and consequence level (levels 1-5, where level 1 is no impact and level 5 is significant loss); the mapping relationship between consequence level and threat parameters is: consequence level × 0.2 + historical occurrence frequency × 0.1 as the weight of historical factors; Historical frequency = number of similar events in the last 30 days / 30; if there are no similar events in the region in the last 30 days, the weight of historical factors is 0.05 by default; (2) Execution of assessment results: The model has a built-in online learning mechanism that uses historical event data to continuously optimize the membership function and inference rules, and finally outputs three levels of threat: low, medium and high. The assessment results will be pushed to the graded response execution module in real time to realize the decision-making upgrade from event qualitative to threat quantitative.

[0014] Preferably, the hierarchical response execution module includes: (1) Design of a graded response strategy: Based on the quantitative results of the threat level module, the module formulates a three-level response plan: for low-risk events, the vehicle warning (voice + instrument prompt) and the slow flashing of roadside lights are triggered, and the surrounding vehicles are broadcast simultaneously via V2V; for medium-risk events, an alarm package containing location and video is pushed to the management platform, and patrol vehicles are dispatched; for high-risk events, the public security system is linked to activate the sound and light alarm, and a vehicle locking command is sent via V2X (with brake function retained) to ensure that the response intensity matches the threat. (2) Transmission and connection of response instructions: All response instructions are transmitted in encrypted form through the vehicle-road-cloud collaborative communication module. At the same time, a response rollback mechanism is designed to support remote unlocking in case of misjudgment. The quantitative results of threat level assessment are matched to form an execution link of threat judgment → response action → instruction transmission.

[0015] Preferably, the vehicle-road-cloud collaborative communication module includes: (1) Full-process data transmission architecture: The module undertakes the transmission tasks of multiple types of data, including trigger signals of the geofencing module, resource status data of the scheduling module, multimodal characteristics of the verification module, grade results of the evaluation module, control commands of the response module, etc. The "5G (high bandwidth video / multimodal data) + C-V2X (low latency control commands)" dual-link architecture is adopted to ensure the high efficiency of data transmission; C-V2X communication follows the 3GPPR16 standard, operates on a frequency band of 5.9GHz, has a transmission rate of ≥256kbps, and an end-to-end latency of ≤50ms. It supports broadcast (BSM), multicast (RSM), and unicast (PSM) message types, and uses the national cryptographic algorithm SM4 for message encryption. (2) Cross-module connection of communication links: Dynamic adaptation is achieved through intelligent link switching algorithms, such as automatic switching to C-V2X in tunnel scenarios, and enabling breakpoint resume mechanism, such as RSU / vehicle terminal cached data, and verification and synchronization after network recovery. This design not only supports the entire process of data flow from geofencing to response execution, but also provides device communication status monitoring data for system configuration and management modules.

[0016] Preferably, the system configuration and management module includes: (1) Configuration logic of all module parameters: Provide customized parameter support for the preceding modules: Define spatiotemporal rules for the V2X geofence dynamic perception module, such as the fence expansion range during weekday morning rush hour; Set algorithm weights for the vehicle-road-cloud resource dynamic scheduling module, such as the priority of coverage integrity; Optimize model parameters for the multimodal event verification module, such as the proportion of multimodal features; Customize action details for the hierarchical response execution module, such as the frequency of sound and light warnings, to ensure that each module adapts to the needs of the scenario; (2) Support and connection of system operation and maintenance: Pre-set scenario templates such as logistics parks and urban trunk roads, and support one-click switching configuration, such as the "park mode" with default small fence + sensitive early warning; at the same time, real-time diagnosis of the operating status of each module (equipment failure, algorithm abnormality, communication interruption), and automatic triggering of early warning and repair suggestions.

[0017] The beneficial effects of this invention are as follows: 1. This invention utilizes a V2X geofence dynamic sensing module to construct an adaptively adjustable geofence based on V2X communication and differential GPS. This geofence can be adjusted in real time according to time period, vehicle density, etc., such as shrinking to key areas late at night or covering queued vehicles during congestion. At the same time, the vehicle-road-cloud resource dynamic scheduling module integrates roadside fixed equipment and surrounding vehicle sensors, incorporating moving vehicles into the sensing network. This solves the problems of weak dynamic sensing and blind spots caused by the reliance on fixed equipment in traditional systems, allowing monitoring needs in different scenarios to be accurately matched.

[0018] 2. This invention integrates multi-source information such as vehicle radar trajectory, camera behavior characteristics, IMU attitude data, roadside sound waves, and panoramic images through a multi-modal event verification module. After spatiotemporal synchronization and data alignment, key features are extracted and then cross-validated through a deep belief network. For example, when determining loitering behavior, the radar trajectory, camera attitude, and sound wave characteristics must be consistent, effectively avoiding misjudgments caused by a single device due to lighting, occlusion, etc., such as preventing normal parking from being misjudged as abnormal, and greatly improving the accuracy and reliability of event characterization.

[0019] 3. This invention uses a threat level intelligent assessment module that combines environmental factors and regional historical data to quantify events into low, medium, and high threats using a fuzzy comprehensive evaluation model. The graded response execution module then initiates appropriate measures accordingly: low-risk events trigger vehicle-mounted and roadside warnings, medium-risk events push alarm packets and dispatch patrol vehicles, and high-risk events involve police intervention and vehicle locking. This mechanism avoids the waste of resources in a unified response, ensures rapid handling of high-risk events, and further improves response efficiency and system usability through the optimization of scenario templates and parameters in the system configuration module. Attached Figure Description

[0020] Figure 1 This is a flowchart of the efficient vehicle monitoring system based on vehicle-road-cloud collaboration of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, this embodiment of the invention provides a high-efficiency vehicle monitoring system based on vehicle-road-cloud collaboration, the system comprising: V2X geofence dynamic sensing module: Based on V2X communication and differential GPS, a dynamic geofence is constructed. The range can be adjusted according to time period, event, vehicle density and trajectory. When a vehicle triggers an entry or exit event, a signal is immediately pushed to the vehicle-road-cloud resource dynamic scheduling module to start monitoring and provide precise spatial triggering conditions for the vehicle-road-cloud resource dynamic scheduling module. Vehicle-Road-Cloud Resource Dynamic Scheduling Module: After receiving the signal from the geofence module, it integrates the real-time status of fixed equipment such as roadside smart streetlights and panoramic cameras with the surrounding vehicles, selects the optimal monitoring combination through an improved greedy algorithm, and transmits the data to the multimodal event verification module. Multimodal event verification module: Based on data from vehicle-mounted radar, cameras, IMUs, roadside acoustic sensors, and panoramic cameras transmitted from the vehicle-road-cloud resource dynamic scheduling module, the event is cross-verified through deep belief networks after spatiotemporal synchronization and feature fusion, and the results are output to the threat level intelligent assessment module to solve the problem of misjudgment by a single device; Threat level intelligent assessment module: Receives the event verification results, integrates environmental factors such as pedestrian density and time period weights from roadside cameras and regional historical event data, quantifies the threat level through a fuzzy comprehensive evaluation model, and pushes it to the graded response execution module; Tiered Response Execution Module: Based on the results of the threat level module, a three-level response is initiated: low risk triggers vehicle-mounted warnings and roadside light alerts; medium risk pushes alarm packets and dispatches patrol vehicles; high risk links with the public security department to trigger audible and visual alarms and vehicle locking commands. The commands are transmitted via the vehicle-road-cloud collaborative communication module to achieve precise action. Vehicle-Road-Cloud Collaborative Communication Module: Supports the transmission of trigger signals, monitoring data, threat levels, response commands, etc. between various modules. It adopts 5G+C-V2X dual links, and ensures stability through intelligent switching and breakpoint resume transmission, connecting the information flow of each link. System Configuration and Management Module: Configures geofencing rules, scheduling algorithm parameters, and validation model weights for other modules, provides scenario templates for logistics parks, monitors equipment and link status, and provides parameter support and operation and maintenance assurance for efficient system operation.

[0023] The V2X geofence dynamic perception module is based on V2X communication and differential GPS positioning technology, and combines a cloud GIS system to build a dynamic geofence. Through adaptive adjustment of the time dimension (time period characteristics, sudden events) and the spatial dimension (vehicle density, driving trajectory), the fence range can match the monitoring needs in real time—for example, shrinking to key areas late at night and extending to cover queuing vehicles when there is congestion. Differential GPS positioning accuracy: static error ≤0.1m, dynamic error ≤0.5m (when vehicle speed ≤120km / h); the coordinate synchronization frequency with the cloud GIS system is 1Hz, ensuring that the fence range adjustment delay is ≤1s; Specific thresholds for fence adjustment: Time period (automatically expands to 1.5 times the default range on weekdays 7:00-9:00); Vehicle density (expands the fence when ≥20 vehicles / 100m², shrinks to 0.5 times the default range when ≤5 vehicles / 100m²); Emergency events (temporarily expands the fence by 500m when a traffic accident is reported). When a vehicle triggers an "enter or leave" fence event, the module immediately generates a standardized trigger signal, which is accurately pushed to the vehicle-road-cloud resource dynamic scheduling module to start the subsequent monitoring process. This is a link design of "vehicle dynamic behavior → trigger signal → scheduling start".

[0024] Trigger signal field description: vehicle_id: Vehicle identification code conforming to ISO3779 standard (17 characters); coords: Uses the WGS84 coordinate system, with an accuracy of 6 decimal places (unit: degrees). fence_id: Unique fence identifier (format: F + 6-digit number, e.g., F000001). Upon receiving the trigger signal from the geofence module, the vehicle-road-cloud resource dynamic scheduling module enters the resource integration phase: on one hand, it collects parameters such as the viewing angle coverage and network latency of roadside fixed equipment (smart streetlights, panoramic cameras); on the other hand, it obtains the status of onboard sensors of surrounding vehicles through V2V communication, such as camera angles and IMU data availability, providing comprehensive data support for scheduling decisions. Based on the improved greedy optimization algorithm, the optimal combination of monitoring nodes is selected with the goal of "complete coverage, efficient transmission, and balanced load". For example, in the case of a curve, the roadside camera and the front-view camera of the vehicle behind are coordinated. After the scheduling is completed, the multi-source data acquisition channel is directed to the multi-modal event verification module to form a closed loop of trigger signal → resource allocation → data acquisition. The objective function of the improved greedy algorithm is 'coverage completeness × 0.6 + transmission efficiency × 0.3 + load balancing × 0.1' (coverage completeness refers to the coverage ratio of the monitoring node to the target area, transmission efficiency is the reciprocal of the data transmission delay, and load balancing is the standard deviation of the occupancy rate of each device). Load balancing calculation method: Utilization rate of each device = number of current tasks / maximum number of tasks that can be carried (roadside cameras can carry a maximum of 8 video feeds, and vehicle-mounted sensors can carry a maximum of 4 data feeds); Standard deviation = √[Σ(utilization rate - mean)² / n], where n is the total number of devices, and a standard deviation < 0.2 is considered 'load balancing'. The iterative steps are as follows: ① Filter candidate nodes with an overlap rate of <10%; ② Calculate the objective function value of each node and select the node with the highest score first; ③ Repeat the filtering until the target area is covered and the transmission delay is <50ms. Candidate node coverage overlap rate = (intersection area of ​​two node monitoring areas / union area of ​​two node monitoring areas) × 100%. When this value is <10%, it is included in the candidate set. Initial candidate node range: roadside fixed equipment includes all equipment within the fenced area; surrounding vehicles are those ≤500m from the fence boundary and whose onboard sensors are online, with priority given to cameras with an angle ≤30° to the target area (to reduce the probability of occlusion). The multimodal event verification module, based on multi-source data allocated by the scheduling module, such as the trajectory and speed of the onboard millimeter-wave radar, the behavioral characteristics of the visual camera, the attitude data of the IMU, and the ambient sound waves from the roadside acoustic sensor and the scene images from the panoramic camera, aligns high and low frequency sampling data using a spatiotemporal synchronization algorithm to extract key features such as "trajectory morphology, action mode, and sound wave frequency." Trajectory morphology characteristics include: trajectory curvature (unit m⁻¹), rate of change of speed (unit m / s²), and dwell time (unit s); motion pattern characteristics include: vehicle steering angle (unit °), light status (on / off), and braking frequency (times / min); sound wave frequency characteristics include: engine sound frequency band (20-2000Hz) and abnormal sound peak value (unit dB). The spatiotemporal synchronization algorithm uses 'GPS timestamp + interpolation compensation' to achieve time alignment (error ≤ 10ms); spatial alignment is achieved by 'differential GPS coordinate transformation + roadside RSU (roadside unit) calibration' to unify the spatial coordinates of vehicle-mounted and roadside equipment to the WGS84 coordinate system (error ≤ 0.5m). The extracted features are input into a deep belief network (DBN) for cross-validation. For example, loitering behavior must simultaneously satisfy the consistency of radar trajectory, camera attitude, and acoustic features to accurately determine the event type. The validation results (including event feature data) will be output to the threat level intelligent assessment module to solve the problem of misjudgment caused by lighting or occlusion of a single device, and complete the link connection between resource collection and event characterization. The Deep Belief Network (DBN) structure is as follows: the input layer contains 128 neurons (corresponding to 128-dimensional features such as trajectory morphology and action pattern), the hidden layers have 3 layers (containing 64, 32, and 16 neurons respectively), and the output layer has 3 neurons (corresponding to 'normal', 'low-risk abnormality', and 'high-risk abnormality'); the training data consists of 100,000 labeled multi-source data (including vehicle radar trajectory, roadside sound waves, etc.); the cross-validation rule is: when the features from at least 3 data sources match the same event label, it is considered a valid result.

[0025] Deep Belief Network (DBN) training parameters: ReLU activation function is used, initial learning rate is 0.001, decays to 0.9 of the previous value every 1000 iterations; batch size is set to 64, total number of training iterations is 50000; Adam optimizer is used, weight decay coefficient is 1e-5; validation set accounts for 20%, training stops when the validation set accuracy does not improve for 1000 consecutive iterations.

[0026] The threat level intelligent assessment module receives the qualitative results from the event verification module and then initiates a multi-factor fusion analysis: on the one hand, it integrates environmental data, such as pedestrian density and time-period risk weights from roadside cameras; on the other hand, it calls historical databases, such as the consequence correlation characteristics of similar events in the region, and transforms abstract events into quantifiable threat parameters through a fuzzy comprehensive evaluation model, such as the weighted calculation of loitering time and pedestrian density. The calculation steps for fuzzy comprehensive evaluation are as follows: ① Determine the weights of the factors (people density 0.4, time period risk 0.3, event duration 0.3); ② Substitute the quantitative values ​​of each factor (e.g., people density of 15 people / ㎡ corresponds to a membership degree of 0.7 for 'medium risk') into the membership function; ③ Obtain the comprehensive threat value by weighted summation (e.g., 0.7×0.4+0.8×0.3+0.6×0.3=0.68), which corresponds to the 'medium risk' level (threshold 0.5-0.8). The factor set of the fuzzy comprehensive evaluation model is {people density, time period risk, event duration}; the membership function is defined as follows: when people density is <5 people / m², the membership degree of 'low risk' is 0.8; when people density is 5-20 people / m², the membership degree of 'medium risk' is 0.7; when people density is >20 people / m², the membership degree of 'high risk' is 0.8; the time period risk weight is: 1.2 for weekday morning peak (7:00-9:00) and 0.8 for late night (0:00-5:00); when the event duration is >30 minutes, the level is upgraded by 1 level. The membership function of pedestrian density is a piecewise linear function: In the formula: x is the population density, unit: people / m².

[0027] The model has a built-in online learning mechanism that uses historical event data to continuously optimize membership functions and inference rules, and finally outputs three levels of threat: low, medium and high. The assessment results will be pushed to the graded response execution module in real time, realizing the decision-making upgrade from event qualitative to threat quantitative.

[0028] The specific logic of the online learning mechanism is as follows: At 3:00 AM every day, historical event data from the past 7 days (sample size ≥ 100) is automatically retrieved, and the membership function parameters are updated using the gradient descent method (learning rate 0.01). When the misclassification rate of a certain type of event is > 5% for 3 consecutive days, an immediate update is triggered to ensure that the model adapts to changes in the scenario.

[0029] Membership function parameter update formula: Parameter update amount = learning rate (0.01) × historical event error gradient, where error gradient = (predicted threat level - actual consequence level) × corresponding factor weight (such as population density weight 0.4, time period risk weight 0.3, etc.).

[0030] The graded response execution module formulates a three-level response plan based on the quantitative results of the threat level module: for low-risk events, it triggers vehicle warnings (voice + instrument prompts) and slow flashing of roadside lights, and simultaneously broadcasts to surrounding vehicles via V2V; for medium-risk events, it pushes an alarm package containing location and video to the management platform and dispatches patrol vehicles; for high-risk events, it links with the public security system to activate a strong audible and visual alarm and sends a vehicle locking command via V2X (while retaining braking function), ensuring that the response intensity matches the threat. Audible and visual alarm parameters: The audible alarm uses a 110dB buzzer (with a continuous interval of 2 seconds), and the visual alarm is a red flashing light (frequency 5Hz), ensuring clear perception within a 50-meter range; the vehicle locking command is transmitted via V2X encryption, restricting only the vehicle starting function (basic braking and steering operations are retained).

[0031] All response commands are transmitted encrypted through the vehicle-road-cloud collaborative communication module. A response rollback mechanism is also designed to support remote unlocking in case of misjudgment. The quantitative results of the threat level assessment are matched to form an execution chain of threat determination → response action → command transmission.

[0032] Remote unlocking process: The management platform sends an unlocking command (encrypted with AES-256) to the vehicle terminal via a 5G link. The command contains a unique unlocking code (valid for 5 minutes). After the vehicle terminal verifies the code, the vehicle is unlocked and the unlocking result is simultaneously fed back to the management platform.

[0033] The vehicle-road-cloud collaborative communication module undertakes the transmission of multiple types of data, including trigger signals from the geofencing module, resource status data from the scheduling module, multimodal characteristics from the verification module, grade results from the evaluation module, and control commands from the response module. It employs a dual-link architecture of "5G (high-bandwidth video / multimodal data) + C-V2X (low-latency control commands)" to ensure efficient data transmission. Dynamic adaptation is achieved through intelligent link switching algorithms, such as automatically switching to C-V2X in tunnel scenarios, and enabling breakpoint resumption mechanisms, such as RSU / vehicle terminal caching data and verifying synchronization after network recovery. This design supports the entire data flow from geofencing to response execution, and also provides device communication status monitoring data for the system configuration and management modules. The intelligent link switching algorithm is triggered as follows: when the 5G link latency is >100ms or the signal strength is <-90dBm, it automatically switches to the C-V2X link; when the C-V2X link packet loss rate is >5%, it switches back to the 5G link (prioritizing control command transmission). Resuming interrupted transmission uses a 'RSU / vehicle terminal local caching + CRC check' mechanism. Cache data is retained for 30 minutes, and synchronization is verified via timestamp alignment after network recovery.

[0034] The system configuration and management module provides customized parameter support for the preceding modules: it defines spatiotemporal rules for the V2X geofence dynamic perception module, such as the fence expansion range during weekday morning rush hours; it sets algorithm weights for the vehicle-road-cloud resource dynamic scheduling module, such as the priority of coverage integrity; it optimizes model parameters for the multimodal event verification module, such as the proportion of multimodal features; and it customizes action details for the hierarchical response execution module, such as the frequency of sound and light warnings, to ensure that each module adapts to the needs of the scenario. It has pre-set scene templates such as logistics parks and urban main roads, and supports one-click switching of configurations. For example, the "park mode" defaults to small fences and sensitive early warnings. At the same time, it diagnoses the operating status of each module in real time (equipment failure, algorithm abnormality, communication interruption) and automatically triggers early warnings and repair suggestions.

[0035] Recommended repair measures: Equipment malfunction: When the camera goes offline, it automatically switches to the nearest backup camera (distance <50m); when the radar packet loss rate exceeds 10%, the device restarts and cached data is synchronized. Link anomaly: When the 5G latency exceeds 200ms, it will automatically switch to C-V2X and compress the video frame rate (from 25fps to 15fps). Algorithm anomaly: When the improved greedy algorithm fails to converge, a backup scheduling scheme is activated (prioritizing roadside equipment). Specific parameters for status diagnosis include: device fault determination (camera offline for more than 5 minutes, radar data packet loss rate >10%); link status indicators (5G / C-V2X packet loss rate >10%, latency >200ms); and algorithm anomaly determination (improved greedy algorithm failing to converge after more than 100 iterations). Diagnostic results are pushed to the management platform via the vehicle-road-cloud collaborative communication module, triggering audible and visual warnings (such as a solid red server indicator light).

[0036] Example of scene template parameters: Weather adaptation parameters: Rain mode: Roadside camera exposure time is extended to 50ms, and radar filter threshold is increased by 20%; Fog mode: The weight of acoustic frequency in multimodal features is increased to 0.5 (default 0.3), and the fence range is expanded by 20%; Scene switching trigger conditions: Rain mode: Automatically activated when the roadside rain sensor detects rainfall ≥ 5mm / h; Fog mode: Automatically activated when the visibility sensor detects visibility ≤500m; Switching delay ≤3s to ensure real-time scene adaptation; Additional parameters for urban arterial road mode: roadside light warning frequency is 2Hz (low risk) / 5Hz (medium risk); Additional parameters for logistics park mode: patrol vehicle dispatch response time threshold is 3 minutes (medium risk) / 1 minute (high risk). Logistics park mode: The default geofence range is 500m×500m, and the trigger density threshold is 10 vehicles / 100m; the coverage integrity weight in the scheduling algorithm is 0.7; the proportion of vehicle radar trajectory in multimodal features is 0.4. Urban arterial road mode: The default geofence range is 1km×1km, and the trigger density threshold is 20 vehicles / 100m; the transmission efficiency weight in the scheduling algorithm is 0.4; the proportion of roadside panoramic images in multimodal features is 0.5.

[0037] Taking the morning rush hour (7:00-9:00) traffic congestion scenario on urban arterial roads as an example: V2X geofence dynamic sensing module: When the detected vehicle density is ≥20 vehicles / 100m, the fence range expands from 500m to 1km to cover the queued vehicles; Vehicle-Road-Cloud Resource Dynamic Scheduling Module: After receiving the trigger signal, it integrates 3 roadside panoramic cameras (coverage overlap rate <8%) and 2 forward-looking cameras of surrounding vehicles (transmission delay <40ms), and selects the optimal combination through an improved greedy algorithm; Multimodal event verification module: acquire data from vehicle radar (speed <5km / h), camera (vehicle stationary), and roadside acoustic sensor (no engine noise), extract features after spatiotemporal synchronization, and input DBN to determine 'abnormal parking'; Threat Level Intelligent Assessment Module: Based on a population density of 15 people / m² (medium-risk membership degree of 0.7) and a morning peak weight of 1.2, it is classified as medium-risk. The tiered response execution module pushes alarm packages (including location and 30-second video) to the management platform and dispatches patrol vehicles. It should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0038] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A high-efficiency vehicle monitoring system based on vehicle-road-cloud collaboration, characterized in that: The system includes: V2X geofence dynamic sensing module: Based on V2X communication and differential GPS, a dynamic geofence is constructed. It can adjust the range according to time period, event, vehicle density, and trajectory. When a vehicle triggers an entry or exit event, it outputs a signal and starts monitoring. Vehicle-Road-Cloud Resource Dynamic Scheduling Module: After receiving signals from the geofencing module, it integrates the real-time status of fixed equipment and surrounding vehicles, selects the optimal monitoring combination through an improved greedy algorithm, and outputs data; Multimodal event verification module: Based on data from vehicle radar, cameras, IMU, roadside acoustic sensors, and panoramic cameras, the module performs spatiotemporal synchronization and feature fusion, cross-verifies events through a deep belief network, and outputs the event verification results. The intelligent threat level assessment module receives event verification results, integrates environmental factors and regional historical event data, and quantifies the threat level using a fuzzy comprehensive evaluation model. Tiered response execution module: Initiates a three-level response based on threat outcome: Low risk triggers vehicle warning and roadside light alert; Medium risk pushes alarm package and dispatches patrol vehicle; High risk links with public security to trigger audible and visual alarms and vehicle locking command; Vehicle-Road-Cloud Collaborative Communication Module: Supports the transmission of trigger signals, monitoring data, threat levels, and response commands between various modules. It adopts 5G+C-V2X dual links and ensures stability through intelligent switching and breakpoint resume. System Configuration and Management Module: Configures geofencing rules, scheduling algorithm parameters, and verification model weights for other modules.

2. The efficient vehicle monitoring system based on vehicle-road-cloud collaboration according to claim 1, characterized in that: The V2X geofence dynamic sensing module includes: (1) Dynamic fence construction logic: Based on V2X communication and differential GPS positioning technology, dynamic geofences are constructed in combination with cloud GIS system. Through adaptive adjustment of time and space dimensions, the fence range can match the monitoring needs in real time. (2) Transmission and connection of trigger signals: When a vehicle triggers an event of entering or leaving the fence, the module immediately generates a standardized trigger signal and accurately pushes it to the vehicle-road-cloud resource dynamic scheduling module to start the subsequent monitoring process.

3. The efficient vehicle monitoring system based on vehicle-road-cloud collaboration according to claim 1, characterized in that: The vehicle-road-cloud resource dynamic scheduling module includes: (1) Multi-dimensional consideration of resource integration: After receiving the trigger signal from the geofence module, the module enters the resource integration stage: on the one hand, it collects the view coverage and network delay parameters of roadside fixed equipment; on the other hand, it obtains the status of vehicle-mounted sensors of surrounding vehicles through V2V communication. (2) Link connection of scheduling results: Based on the improved greedy optimization algorithm, the optimal combination of monitoring nodes is selected. After the scheduling is completed, the multi-source data acquisition channel is directed to the multi-modal event verification module.

4. The efficient vehicle monitoring system based on vehicle-road-cloud collaboration according to claim 1, characterized in that: The multimodal event verification module includes: (1) Multi-source data fusion mechanism: Based on the multi-source data allocated by the scheduling module, high and low frequency sampling data are aligned through a spatiotemporal synchronization algorithm to extract key features of trajectory morphology, action mode, and sound wave frequency; (2) Decision connection of verification results: The extracted features are input into a deep belief network for cross-validation, and the verification results are output to the threat level intelligent assessment module.

5. The efficient vehicle monitoring system based on vehicle-road-cloud collaboration according to claim 1, characterized in that: The threat level intelligent assessment module includes: (1) Input logic for level quantification: After receiving the qualitative results from the event verification module, start multi-factor fusion analysis: on the one hand, integrate environmental data; on the other hand, call the historical database and transform the abstract event into quantifiable threat parameters through the fuzzy comprehensive evaluation model; (2) Execution of assessment results: The model has a built-in online learning mechanism that uses historical event data to continuously optimize the membership function and inference rules, and finally outputs three threat levels: low, medium and high.

6. The efficient vehicle monitoring system based on vehicle-road-cloud collaboration according to claim 1, characterized in that: The hierarchical response execution module includes: (1) Design of graded response strategy: Based on the quantitative results of the threat level module, the module formulates a three-level response plan: low-risk events trigger vehicle warning and slow flashing of roadside lights; medium-risk events push alarm packages to the management platform and dispatch patrol vehicles; high-risk events link the public security system to activate sound and light alarms. (2) Transmission of response instructions: All response instructions are transmitted in encrypted form through the vehicle-road-cloud collaborative communication module. At the same time, a response rollback mechanism is designed. This step is matched with the quantitative results of the threat level assessment.

7. The efficient vehicle monitoring system based on vehicle-road-cloud collaboration according to claim 1, characterized in that: The vehicle-road-cloud collaborative communication module includes: (1) Transmission architecture of the whole process data: The module undertakes the transmission tasks of multiple types of data, including trigger signals, multimodal characteristics of resource status data, level results, and control instructions, and adopts a 5G+C-V2X dual-link architecture; (2) Cross-module connection of communication links: Dynamic adaptation is achieved through intelligent link switching algorithm and breakpoint resume mechanism is enabled.

8. The efficient vehicle monitoring system based on vehicle-road-cloud collaboration according to claim 1, characterized in that: The system configuration and management module includes: (1) Configuration logic of all module parameters: Define spatiotemporal rules for the V2X geofence dynamic perception module; set algorithm weights for the vehicle-road-cloud resource dynamic scheduling module; optimize model parameters for the multimodal event verification module; customize action details for the hierarchical response execution module; (2) Support and connection of system operation and maintenance: Pre-set logistics park and urban trunk road scenario templates, supporting one-click configuration switching; at the same time, real-time diagnosis of the operating status of each module, and automatic triggering of early warning and repair suggestions.

Images