A risk early warning coordination system of a two-wheeled vehicle and the vehicle

Abstract

The application discloses a two-wheeled vehicle risk early warning cooperation system and vehicle. Each two-wheeled vehicle is connected to a target vehicle group through a dynamic networking mechanism, the system is mounted on each two-wheeled vehicle of the target vehicle group, and the system comprises a local perception module, a V2V communication module and a group cooperation load reduction module. A single vehicle identifies a high-risk event through the local perception module and calculates a credibility, generates a risk message after the credibility meets a standard, and broadcasts the risk message to the whole group through the V2V module. The V2V module screens verification data collected by vehicles meeting a threshold distance from a risk point, and the group cooperation load reduction module fuses credibility scores of the vehicles and calculates a comprehensive confidence degree based on the verification data. When the comprehensive confidence degree reaches a first score threshold, the group cooperation load reduction module outputs a risk early warning, and simultaneously reduces a running frequency of the local perception module. The application can improve credibility of two-wheeled vehicle risk determination, and simultaneously reduce perception load.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, specifically to a risk warning and coordination system and vehicle for two-wheeled vehicles. Background Technology

[0002] With the increasing demand for short-distance urban travel, the number of motorized two-wheeled vehicles, including electric motorcycles, mopeds, and gasoline-powered motorcycles, continues to rise. These vehicles are generally equipped with cameras, tire pressure monitoring systems, and vehicle-to-everything (V2X) modules, providing basic environmental perception capabilities. Relying on cellular vehicle-to-everything (C-V2X), 4G, or 5G communication channels, they can achieve data interaction between vehicles. In high-frequency scenarios such as food delivery and ride-sharing, the phenomenon of vehicles traveling in convoys is becoming increasingly common, providing a foundation for fleet collaborative management. However, current two-wheeled vehicle-mounted perception solutions require sensors to maintain a high-frequency acquisition state for extended periods, continuously collecting massive amounts of environmental data, placing a heavy burden on onboard computing power and hardware power consumption. Furthermore, single-vehicle perception and identification have low reliability, lack multi-vehicle collaborative verification mechanisms within fleets, and the accuracy of risk assessment results is insufficient.

[0003] Therefore, how to improve the reliability of risk assessment for two-wheeled vehicles and reduce the on-board sensing load is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a risk warning collaborative system and vehicle for two-wheeled vehicles, which can improve the reliability of risk assessment for two-wheeled vehicles while reducing the load on the vehicle-mounted sensing system.

[0005] The embodiments of this application disclose the following technical solutions: A risk warning and coordination system for two-wheeled vehicles, wherein each two-wheeled vehicle connects to a target fleet through a dynamic networking mechanism, the system is mounted on each two-wheeled vehicle in the target fleet, and the system includes a local perception module, a V2V communication module and a group collaborative unloading module; The local perception module is used to collect the first surrounding environment data of the current vehicle, and identify whether a first high-risk event has occurred during the current vehicle's driving process based on the first surrounding environment data; the first high-risk event is a road safety risk event. The group collaborative load reduction module is used to calculate the credibility score of the first high-risk event; if the credibility score is greater than or equal to the credibility threshold, a first risk message is generated based on the first risk data output by the local perception module and the credibility score of the first high-risk event; the first risk message carries the risk location and the credibility score of the first high-risk event; the first risk data includes the first risk event type and the coordinates of the risk location; The V2V communication module is used to broadcast the first risk message to the remaining vehicles in the target convoy and receive the first verification data returned by the first target vehicle in the target convoy; the remaining vehicles are all vehicles in the target convoy except the current vehicle; the first target vehicle is the vehicle among the remaining vehicles whose distance from the risk point is less than or equal to a distance threshold. The group collaborative load reduction module is further configured to calculate the first comprehensive confidence level of the first high-risk event by combining the confidence score of the current vehicle for the first high-risk event and the first verification data of the first target vehicle; if the first comprehensive confidence level is greater than or equal to the first scoring threshold, a risk warning is issued and the operating frequency of the local perception module in the current vehicle is reduced.

[0006] In one possible implementation, the V2V communication module is further configured to receive a second risk message broadcast by any remaining vehicle; the second risk message is a risk message generated and broadcast by any remaining vehicle after identifying a second high-risk event. When the V2V communication module receives the second risk message, the group cooperative load reduction module is further used to calculate the distance between the current vehicle and the risk point carried in the second risk message to obtain the risk distance; if the risk distance is greater than the distance threshold, it is determined that the current vehicle is not within the effective perception range, and the second risk message is discarded; if the risk distance is less than or equal to the distance threshold, it is determined that the current vehicle is within the effective perception range, and its own local perception module is activated to collect the second surrounding environment data of the current vehicle, and the second surrounding environment data is used to identify whether the second high-risk event is a real event. If the second high-risk event is determined to be a real event, the group collaborative load reduction module is also used to calculate the credibility score of the second high-risk event; The V2V communication module is also used to receive second verification data returned by a second target vehicle in the target convoy; the second target vehicle is a vehicle among the remaining vehicles that, excluding those that broadcast the second risk message, is less than or equal to the distance threshold from the risk point. The group collaborative load reduction module is further configured to calculate the second comprehensive confidence level of the second high-risk event by combining the current vehicle's confidence score for the second high-risk event, the second risk message, and the second verification data of the second target vehicle; if the second comprehensive confidence level is greater than or equal to the first scoring threshold, the operating frequency of the local perception module in the current vehicle is reduced.

[0007] In one possible implementation, the formula for calculating the overall confidence level is as follows: The formula for calculating the overall confidence level is as follows: FinalConfidence=Σ(w i ×c i ); Where FinalConfidence is the overall confidence level; c i The credibility score output by the i-th vehicle for high-risk events; w i w is the weighting factor for the rating of the i-th vehicle. i =α×(1 / d i )+β×q i +γ×r i ;d i Let q be the distance between the i-th vehicle and the risk point; i For the sensor quality score of vehicle i, r i Let be the historical accuracy of vehicle i; α be the relative distance weighting coefficient of vehicles; β be the sensor quality score weighting coefficient; γ be the vehicle reputation score weighting coefficient; i iterates through all vehicles participating in the confidence calculation of this high-risk event, and each vehicle i is the current vehicle for identifying the high-risk event or the target vehicle that meets the distance condition.

[0008] In one possible implementation, the system is characterized by further including a human-computer interaction module, which includes a visual interaction device, an auditory interaction device, and a smart handle. When the overall confidence level is greater than or equal to the first scoring threshold, the visual interaction device is used to overlay and display the relevant early warning information of high-risk events in the driver's field of vision in an augmented reality form; When the overall confidence level is greater than or equal to the first scoring threshold, the auditory interaction device is used to deliver high-risk event warning information to the driver in the form of voice broadcast; When the overall confidence level is greater than or equal to the first scoring threshold, the smart handlebar is used to transmit a high-risk event warning signal to the driver through vibration feedback.

[0009] In one possible implementation, when the high-risk event is a visual event, the credibility score c... i The calculation formula is as follows: c i =f(Model Score Environmental Factor Model Score The probability value output by the AI ​​object detection model, Environmental Factor Here, f() represents the environmental factor correction coefficient, and f() denotes the confidence score c obtained by weighted fusion of the model output probability value and the environmental correction coefficient. i ; When the high-risk event is a non-visual event, the credibility score c i The calculation formula is as follows: c i =g(IMU Peak TirePressure Change Threshold), IMU Peak TirePressure provides peak vibration data output from the inertial measurement unit (IMU). Change The instantaneous tire pressure change collected by tire pressure monitoring is represented by Threshold, which is a preset risk assessment benchmark threshold. g() represents the confidence score c output by performing a piecewise mapping calculation combining the vibration peak value, tire pressure change, and benchmark threshold. i .

[0010] In one possible implementation, the group-coordinated load reduction module is further configured to: If the overall confidence level is greater than or equal to the first scoring threshold, a sliding window combined with an exponential smoothing update algorithm is used to update the vehicle's historical accuracy r in real time. i ; Among them, the vehicle history accuracy rate r i The update formula is: r(t)=(1-η)×r(t-1)+η×a t r(t) represents the vehicle's historical accuracy at time t, and r(t-1) represents the vehicle's historical accuracy at the previous time step; a t This is the accuracy indicator for this risk alert; η is the preset learning rate; when multiple vehicles in the fleet verify and confirm the existence of a high-risk event, a t Assigning a value of 1 corresponds to a gradual increase in the vehicle's historical accuracy; when a risk event is determined to be a false alarm, a... t Assigning a value of 0 will decrease the accuracy of the vehicle history for the corresponding vehicle.

[0011] In one possible implementation, the verification data includes the confidence score of the target vehicle's output for the high-risk event, the sensor quality score of the target vehicle, the vehicle's historical accuracy, and the distance between the target vehicle and the risk location.

[0012] In one possible implementation, the group-coordinated load reduction module is further configured to: When the overall confidence level is less than the first scoring threshold but greater than or equal to the second scoring threshold, the high-risk event is determined to be a medium-confidence risk event. The event is marked as pending observation, and no formal warning is pushed out. We continue to wait for the perception data of other vehicles in the fleet to complete the secondary verification. When the overall confidence level is less than the second scoring threshold, the high-risk event is determined to be a low-confidence suspected risk, and the risk message is directly discarded, or only a record is kept in the local background.

[0013] In one possible implementation, the group-coordinated load reduction module is further configured to: If multiple vehicles report risk messages targeting the same area and with the same risk event type within a preset time period, the multiple risk messages are semantically deduplicated and integrated based on a geographic location clustering algorithm and high-risk event type to obtain a merged message. Subsequent comprehensive confidence calculation is then performed based on the merged message.

[0014] In one possible implementation, reducing the operating frequency of the local sensing module specifically includes: performing differentiated scheduling on various sensing devices in the local sensing module; The differentiated scheduling includes reducing the real-time frame rate of the camera to a preset frame rate, disabling the AI ​​detection model and retaining only the basic image acquisition function of the camera, extending the tire pressure sampling cycle from the first sampling cycle to the second sampling cycle, reducing the sampling rate of the IMU, and turning off the high-frequency vibration analysis function of the IMU.

[0015] A vehicle comprising a two-wheeled vehicle-based risk warning and coordination system as described above.

[0016] Compared with the prior art, this application has the following beneficial effects: This application provides a risk warning and collaborative system for two-wheeled vehicles and a corresponding vehicle. The system is mounted on each two-wheeled vehicle in a fleet, forming a collaborative system through dynamic networking. A local perception module collects data on the vehicle's surrounding environment and initially identifies high-risk events. A group collaborative unloading module then calculates a credibility score. Upon reaching a certain threshold, a risk message is generated and broadcast via a vehicle-to-vehicle (V2V) communication module. Next, the V2V communication module receives verification data from the first target vehicle in the fleet that meets the distance criteria. The group collaborative unloading module then integrates multiple dimensions of data, including the credibility scores of the current vehicle and the first target vehicle, sensor quality scores, and historical accuracy rates, to calculate a first comprehensive confidence level. This multi-vehicle collaborative verification mechanism replaces independent judgment by a single vehicle, effectively filtering false risks and significantly improving the credibility and accuracy of risk identification. Finally, when the first comprehensive confidence level is greater than or equal to the first scoring threshold, confirming the risk's authenticity and validity, the group collaborative load reduction module generates a risk warning through the human-machine interaction module to ensure riding safety. Simultaneously, it proactively reduces the operating frequency of the local sensing module, decreasing the data collection and computational load of the sensing devices, effectively alleviating the onboard computing power pressure and reducing hardware power consumption. This application forms a target fleet through dynamic networking of various two-wheeled vehicles. The current vehicle initiating the risk assessment, combined with verification data returned by the first target vehicle, calculates the first comprehensive confidence level by integrating key data such as multi-party confidence scores, sensor quality scores, and vehicle historical accuracy rates through the group collaborative load reduction module. Relying on a multi-vehicle cross-collaborative verification mechanism, multi-party verification of risk assessment is achieved, effectively filtering false risks and significantly improving the credibility and accuracy of risk identification. Simultaneously, after the comprehensive confidence level is greater than or equal to the first scoring threshold, confirming the risk's authenticity and validity, the group collaborative load reduction module proactively reduces the operating frequency of the local sensing module, decreasing the data collection and computational load of the sensing devices, effectively alleviating the onboard computing power pressure and reducing hardware power consumption. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings used in the description of the embodiment or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic diagram illustrating an exemplary application scenario provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of a risk warning and collaborative system for two-wheeled vehicles provided in an embodiment of this application; Figure 3 This is a schematic diagram of an AR interface provided in an embodiment of this application. Detailed Implementation

[0019] To facilitate understanding of the technical solutions provided in the embodiments of this application, the background technology involved in the embodiments of this application will be described below.

[0020] With the increasing demand for short-distance urban travel, the number of motorized two-wheeled vehicles, including electric motorcycles, light electric motorcycles, and gasoline motorcycles, continues to rise. These vehicles are increasingly equipped with cameras, tire pressure monitoring devices, and vehicle-to-everything (V2X) modules, enabling basic intelligent data perception. C-V2X, 4G, and 5G communication technologies provide underlying support for information exchange between different vehicles. The characteristics of concentrated vehicle platooning in high-frequency usage scenarios such as shared mobility and food delivery are becoming increasingly prominent, supporting the implementation of multi-vehicle collaborative management functions. However, current two-wheeled vehicle-mounted perception systems rely solely on single-vehicle sensor data to determine road risks, resulting in poor reliability of the identification results. Furthermore, the continuous high-frequency collection of environmental data by sensing devices leads to high power consumption and computing power costs for onboard hardware, and there is a lack of solutions for multi-vehicle collaborative risk verification and adaptive load reduction for sensing.

[0021] To address this issue, this application provides a risk warning and collaborative system for two-wheeled vehicles. Each two-wheeled vehicle connects to the target fleet via a dynamic networking mechanism. The system is mounted on every two-wheeled vehicle in the target fleet and includes a local perception module, a V2V communication module, and a group collaborative load reduction module. Each vehicle relies on its local perception module to identify high-risk events and calculate a confidence score. Upon reaching a certain threshold, a first risk message carrying the coordinates of the risk location and event type is generated and broadcast across the network via the V2V communication module. After broadcasting, the V2V communication module filters verification data from vehicles that meet the threshold based on distance from the risk location. The group collaborative load reduction module integrates single-vehicle and multi-vehicle verification information to calculate a comprehensive confidence score. If the comprehensive confidence score is greater than or equal to the first score threshold, a risk warning is issued, and the operating frequency of the local perception module in the current vehicle is reduced. This application overcomes the shortcomings of existing solutions that rely solely on single-vehicle perception, have low confidence in risk assessment, and insufficient identification accuracy by utilizing a multi-vehicle cross-collaborative verification mechanism, significantly filtering out false risks. Meanwhile, after confirming the authenticity and effectiveness of the risk by meeting the comprehensive confidence level, the operating frequency of the local sensing module was proactively reduced to address the problem of excessive burden on vehicle computing power and hardware power consumption caused by the high-frequency acquisition of massive environmental data by existing sensing devices.

[0022] To facilitate understanding of the risk warning and collaborative system for two-wheeled vehicles provided in the embodiments of this application, the following is combined with... Figure 1 The example scenario is shown below. See also... Figure 1 This figure is a schematic diagram of an exemplary application scenario provided in the embodiments of this application.

[0023] This embodiment uses a target convoy consisting of four two-wheeled vehicles (A, B, C, and D) as an example to fully illustrate the entire process of multi-vehicle collaborative risk warning: First, vehicle A in the fleet, acting as the risk initiator, uses its local perception module equipped with tire pressure sensors and an IMU unit to collect vehicle vibration data, identifying a P1-level high-risk event—a series of potholes on the road surface. Vehicle A's group-coordinated load reduction module calculates the credibility score of this high-risk event. If the calculated credibility score is greater than or equal to a preset credibility threshold (e.g., 70 points), it indicates that the high-risk event warrants further verification and early warning. A standardized risk message is generated and simultaneously broadcast to vehicles B, C, and D in the fleet via the V2V communication module. Vehicles B, C, and D simultaneously receive the risk broadcast message and perform local perception verification in parallel.

[0024] Among them, after receiving the risk broadcast message, the group collaborative unloading module of vehicles B, C and D synchronously calculates the risk distance between the vehicle and the risk point carried in the risk message in order to identify whether the risk distance is greater than or equal to the distance threshold.

[0025] If the risk distance of vehicle B is less than the distance threshold, vehicle B immediately calls the camera in its local perception module to collect image data of the road ahead. Combining the collected image information with the preset credibility scoring formula for visual high-risk events, vehicle B calculates its own credibility score for this high-risk event. If the risk distance of vehicle C is greater than the distance threshold, vehicle C's own group collaborative unloading module immediately marks the risk event as "to be observed," and it no longer calculates its own credibility score for this event, nor does it report credibility-related data to the fleet; it only retains a record of no effective perception result. If the risk distance of vehicle D is less than the distance threshold, vehicle C immediately calls the camera in its local perception module to collect image data of the road ahead. Combining the collected image information with the preset credibility scoring formula for visual high-risk events, vehicle C calculates its own credibility score for this high-risk event.

[0026] Then, vehicles B and D respectively transmit their locally perceived credibility scores, sensor quality scores, vehicle historical accuracy, and distance between the vehicle and the risk point as verification data back to vehicle A's group collaborative load reduction module, while vehicle C only transmits relevant records that it has no effective perception data for and that have been marked "to be observed".

[0027] The group collaborative load reduction module of vehicle A, acting as the core processing unit, collects its own initial credibility score, complete verification data returned by vehicles B and D (including their respective credibility scores, sensor quality scores, vehicle historical accuracy rates, and distances to risk points), and the no-valid-perception records and "to be observed" markings returned by vehicle C. Since the verification data of vehicles B and D already includes their distances to risk points, sensor quality scores, and vehicle historical accuracy rates, there is no need to retrieve them repeatedly. The module directly integrates vehicle A's own relevant parameters (distances to risk points, sensor quality scores, and vehicle historical accuracy rates) with the verification data of vehicles B and D, and the relevant records of vehicle C, uniformly substituting them into the comprehensive confidence calculation formula to complete the fusion calculation. The comprehensive confidence score corresponding to the high-risk event is calculated to be 92 points. Subsequently, the group collaborative load reduction module of vehicle A, based on the preset credibility grading rules, determines that the risk belongs to a high-credibility event. The credibility grading rules include: a comprehensive confidence score of 85 or higher is considered a high-credibility event; a comprehensive confidence score of 60 or higher but less than 85 is considered a medium-credibility event; and a comprehensive confidence score of less than 60 is considered a low-credibility event.

[0028] After completing the full-domain data fusion and judgment, vehicle A's group collaborative load reduction module confirms that the current risk comprehensive confidence level has reached the high-confidence standard, generates a risk warning message, and broadcasts it to all two-wheeled vehicles in the fleet. The unified message is: "A pothole section is ahead, x meters away, confirmed by 3 vehicles. Please slow down." At the same time, vehicle A issues a risk warning through its own human-machine interaction module and reduces the operating frequency of its local perception module. Vehicles B and D, upon receiving the risk warning message, simultaneously issue risk warnings through their own human-machine interaction modules and reduce the operating frequency of their local perception modules. Although vehicle C receives the risk warning message, because no matching risk has been detected locally and the event has been previously marked as "to be observed," it only receives and stores the warning message, does not trigger a human-machine interaction warning, and does not execute the local perception module frequency reduction operation.

[0029] This fully demonstrates the dynamic collaborative operation logic of this application, which includes single-point reporting, simultaneous message reception by multiple vehicles, parallel hierarchical cross-verification, unified calculation of comprehensive confidence level by summarizing all vehicle data, and unified and simplified early warning. It can solve the problems of redundant perception computing power of single vehicles and excessive cognitive load on drivers caused by repeated alarms from multiple vehicles. At the same time, it relies on cross-verification of full data from four vehicles to reduce the probability of false alarms and missed alarms of road risks, and improve the overall safety early warning capability of two-wheeled vehicle groups.

[0030] 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 this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0031] See Figure 2 , Figure 2 This is a structural diagram of a risk warning and collaborative system for two-wheeled vehicles provided in an embodiment of this application. The risk warning and collaborative system for two-wheeled vehicles proposed in this application is applicable to various types of motorized two-wheeled vehicles such as electric motorcycles, light electric motorcycles, and fuel motorcycles. It can effectively solve the problems of existing two-wheeled vehicle-mounted sensing solutions that rely solely on single-vehicle sensor data to determine risks, resulting in low reliability, and the excessive burden on vehicle-mounted computing power and hardware power consumption caused by long-term high-frequency data collection by sensing devices.

[0032] Specifically, each two-wheeled vehicle connects to the target fleet via a dynamic networking mechanism. This system is installed on every two-wheeled vehicle in the target fleet, such as... Figure 2 As shown, the system mainly includes a local sensing module, a V2V communication module, and a group collaborative load reduction module. The three modules work together to achieve risk collaborative early warning and sensing load optimization within the fleet.

[0033] Among them, the local perception module, as the basic perception unit of the system, is mainly used to collect the first surrounding environment data of the current vehicle, and based on the collected first surrounding environment data, it is combined with the AI ​​detection model for analysis and processing to identify whether the current vehicle has encountered the first high-risk event during driving.

[0034] It should be noted that the surrounding environment data in this embodiment includes, but is not limited to, image data collected by cameras, distance data collected by lidar, speed data collected by millimeter-wave radar, vehicle operating status data collected by accelerometers, and tire status data collected by tire pressure sensors. This raw data is transmitted to an artificial intelligence (AI) detection model, which performs fusion analysis, target identification, and risk assessment on the various data types. This allows for comprehensive capture of the surrounding environment and the vehicle's own operating status during operation, accurately identifying potential risks such as road obstacles, road surface anomalies, and vehicle malfunctions, providing comprehensive and accurate data and technical support for risk identification. Furthermore, the first high-risk event in this embodiment specifically refers to a P0 or P1 level road safety risk event. Such risk events have a significant impact on the driving safety of two-wheeled vehicles and require timely warning and handling. Among them, P0 level is an extremely high-risk event, including situations that may directly lead to safety accidents such as vehicle collisions, sudden obstacles entering the driving lane, and road collapses. P1 level is a high-risk event, including situations that may indirectly lead to safety accidents such as sudden vehicle malfunctions, slippery road surfaces, and illegal lane changes by surrounding vehicles. By classifying and defining high-risk events, early warning priorities can be accurately matched, improving the pertinence and timeliness of early warning and response.

[0035] It should also be noted that the AI ​​detection model adapted to the low computing power of two-wheeled vehicle hardware uses lightweight convolutional neural networks such as YOLOv5-Nano and MobileNet-V3 Small. The network is divided into three parts: feature extraction layer, neck network and detection head. The feature extraction layer introduces depthwise separable convolution to reduce the number of model parameters.

[0036] The AI ​​detection model takes the raw video stream from the camera as input. It first adjusts the image resolution to 320x320 or 640x640 pixels, then enhances the contrast through histogram equalization, and normalizes pixel values ​​to the 0-1 range for standardized preprocessing. In multimodal perception scenarios, it also stitches together the peak vehicle vibration data and tire pressure monitoring values ​​from the IMU (Integrated Device Unit) and converts them into feature vectors via a fully connected layer, fusing them with image visual features for joint analysis. If the system is configured only in pure visual perception mode, or if IMU / tire pressure sensor data is unavailable, the IMU and tire pressure data branches are ignored, and risk assessment is based solely on the camera video stream.

[0037] The output of the AI ​​detection model includes risk type ID (such as potholes, pedestrians, sudden braking), bounding box coordinates, object presence confidence, and classification confidence. The final risk assessment score is obtained by multiplying the object presence confidence and classification confidence. If the score exceeds the risk score threshold (such as 0.6), it is determined that there is a potential road risk in the current scene (i.e., there is a first-highest risk event).

[0038] The AI ​​model is trained on a dataset of real cycling videos covering various lighting conditions and road surface conditions. After manually labeling road risk targets, data augmentation is performed using methods such as random rotation, brightness adjustment, and adding Gaussian noise. During the training phase, CIoULoss is used as the bounding box regression loss and cross-entropy loss is used as the classification loss. It is then trained iteratively with a stochastic gradient descent (SGD) optimizer until the model accuracy converges, ensuring the accuracy of risk identification in complex cycling scenarios.

[0039] The group-coordinated unloading module, as the core processing unit of the system, is responsible for assessing the credibility of the first high-risk event identified by the local perception module. This involves calculating a credibility score for the first high-risk event. If the calculated credibility score is greater than or equal to a preset credibility threshold (e.g., 70 points), it indicates that the high-risk event requires further verification and early warning. At this point, the group-coordinated unloading module will generate a first risk message based on the first risk data output by the local perception module and the credibility score of the first high-risk event. This first risk message carries key information such as the risk location and the credibility score of the first high-risk event. Specifically, the first risk data includes the type of the first risk event and the coordinates of the risk location, enabling other vehicles in the convoy to accurately locate the risk position and identify the risk type.

[0040] It should be noted that risk points are textual and concrete descriptions of the locations of safety hazards on the road, used to intuitively inform the convoy of which section of the road is at risk. For example, in this scenario, the road section with potholes 80 meters ahead.

[0041] The first risk event type is a classification label for high-risk hazards, used to distinguish different road hazards, such as potholes, water accumulation, and road obstacles, which all belong to different event types.

[0042] Risk point coordinates are the digital positioning parameters corresponding to risk points. They are generally based on latitude and longitude coordinates from the Global Positioning System (GPS). These are machine-readable data. The vehicle positioning module can read these coordinates and automatically calculate its distance from the risk location, providing numerical basis for subsequent distance threshold judgment and comprehensive confidence calculation.

[0043] The V2V communication module, as the information interaction unit of the system, mainly undertakes the data transmission function between vehicles in the fleet. On the one hand, it is used to broadcast the first risk message generated by the current vehicle to the remaining vehicles in the target fleet, which specifically refers to all vehicles in the target fleet except the current vehicle. On the other hand, it is used to receive the first verification data returned by the first target vehicle in the target fleet. The first target vehicle is the vehicle among the remaining vehicles whose distance from the risk point is less than or equal to a preset distance threshold (e.g., 200 meters). This type of vehicle is within the effective perception range of the risk point, and its verification data has high reference value. The first verification data specifically includes the credibility score output by the first target vehicle for the first high-risk event, the sensor quality score of the first target vehicle, the vehicle's historical accuracy, and the distance between the first target vehicle and the risk point, providing multi-dimensional reference for subsequent comprehensive confidence calculation.

[0044] In addition, the group collaborative load reduction module is also used to fuse multi-vehicle data. Specifically, it combines the credibility score of the first high-risk event generated locally by the vehicle, the distance between the vehicle and the risk point, its own sensor quality score, the vehicle's historical accuracy, and the first verification data returned by the first target vehicle in the fleet (including the credibility score, distance, sensor quality score, and historical accuracy of each first target vehicle). It then summarizes the parameters of all vehicles that meet the distance condition and substitutes them into the comprehensive confidence calculation formula to solve for the first comprehensive confidence score of the first high-risk event. If the calculated first comprehensive confidence score is greater than or equal to the preset first score threshold (e.g., 85 points), it indicates that the high-risk event is real and valid. At this time, the group collaborative load reduction module will generate a corresponding risk warning signal and push the risk warning to the driver of the current vehicle through the human-machine interaction module. At the same time, it will actively reduce the operating frequency of the local perception module in the current vehicle. Under the premise of ensuring driving safety, it will reduce the computing power and power consumption of the on-board perception equipment, and achieve an organic combination of accurate risk warning and lightweight equipment operation.

[0045] The above describes the complete workflow of a vehicle as the risk initiator. Below, this application further provides an implementation scheme for a vehicle as the risk receiving and verification endpoint: Specifically, the V2V communication module installed on each two-wheeled vehicle can not only broadcast risk messages it identifies, but also continuously listen to and receive second risk messages sent by any other vehicle in the fleet. These messages are generated by other vehicles after they perceive the second high-risk event, and are encapsulated according to a unified standard.

[0046] After the V2V communication module successfully captures the second risk message, it synchronizes the message data to the group collaborative unloading module. The group collaborative unloading module combines the vehicle positioning information to calculate the straight-line distance between the vehicle and the risk point marked in the message as the risk distance. Based on the preset distance threshold, it executes different processing logics in layers: if the calculated risk distance is greater than the distance threshold, it is determined that the vehicle has left the effective perception range of the risk point. Without conducting local verification, the second risk message is directly discarded to save onboard computing resources. If the risk distance is less than or equal to the distance threshold, it is determined that the vehicle is within the effective perception range where the risk can be verified. Then, the local perception module is activated to collect the second surrounding environment data. Based on the collected multi-source sensor data, the authenticity of the second high-risk event is identified and judged.

[0047] If the local perception results confirm that the second highest risk event actually exists, the group collaborative load reduction module will then call the built-in credibility calculation logic to solve for the credibility score of the vehicle for the second highest risk event.

[0048] At the same time, the V2V communication module continuously collects the second verification data transmitted back by the other second target vehicles in the convoy. Here, the second target vehicle specifically refers to the vehicle in the convoy that is excluding the vehicle that sent the second risk message this time and whose distance from the risk point does not exceed the distance threshold. The verification data uploaded by each second target vehicle includes multi-dimensional reference indicators such as its own credibility, sensor quality, historical recognition accuracy, and distance from the risk point.

[0049] Subsequently, the group collaborative load reduction module integrates three types of core information for fusion calculation. The required information includes three categories: first, the original risk information carried in the second risk message, from which the credibility score of the risk-initiating vehicle and the coordinates of the risk point are extracted; second, the credibility score of the second high-risk event output after local perception verification by this vehicle, the corresponding distance to this vehicle, sensor quality, and historical accuracy parameters; and third, the second verification data uploaded by all second target vehicles in the fleet (including the credibility score, distance, sensor quality score, and historical accuracy of each second target vehicle). This involves all vehicles participating in this verification... i d i q i r i Substituting into the unified comprehensive confidence score calculation formula: FinalConfidence=Σ(w i ×c i The second comprehensive confidence level corresponding to the second highest risk event is obtained by solving the equation.

[0050] When the second comprehensive confidence level value is greater than or equal to the preset first scoring threshold, it means that the road risk is sufficiently credible after cross-verification by multiple vehicles. The group collaborative load reduction module then issues a scheduling instruction to reduce the overall operating frequency of the local perception module of the vehicle, thereby reducing the power consumption and computing power overhead caused by the continuous high-frequency acquisition of the sensing equipment on the premise that the risk is confirmed to be real and reliable.

[0051] In one possible implementation, after the group collaborative load reduction module completes the comprehensive confidence calculation, it will perform confidence level processing according to the preset score range, and execute differentiated processing logic for different confidence levels: When the overall confidence score is greater than or equal to 85, it is judged as a high-confidence risk event. The group collaborative load reduction module directly pushes the risk to the early warning queue, triggering a full set of multimodal human-machine early warnings. When the overall confidence level is in the range of 60 to 85 points, it is judged as a medium-risk event. The group collaborative load reduction module marks the event as "to be observed" and does not push a formal warning for the time being, and continues to wait for the perception data of other vehicles in the fleet to complete the secondary verification. When the overall confidence score is below 60, it is judged as a low-confidence suspected risk. The group collaborative load reduction module directly discards the risk message, or only keeps a record in the local background, without broadcasting it to the outside world or triggering any riding prompts.

[0052] Through a three-tiered handling mechanism, false alarms with insufficient credibility can be accurately filtered out, and only high-reliability risks that have been fully verified by multiple vehicles are pushed to the driver, effectively reducing riding interference caused by invalid prompts and reducing the cognitive load on the driver.

[0053] It should be noted that performing another round of weighted fusion confidence calculation for the same area and similar risks is called secondary verification. Secondary verification will add the c of other remaining vehicles in the target convoy (excluding the first or second target vehicle). i and weight w i This will be used to participate in the calculation, thereby supplementing more perceptual samples and further verifying whether the risk is real.

[0054] In one possible implementation, when performing fusion verification of multi-vehicle perception data and solving for the comprehensive risk confidence (i.e., solving for the first comprehensive confidence or the second comprehensive confidence), this application uses a weighted summation model to calculate FinalConfidence. The core calculation formula is as follows: FinalConfidence=Σ(w i ×c i ); Wherein, FinalConfidence is the overall confidence level; in the formula, i traverses all vehicles participating in this high-risk event verification. The participating vehicles include not only the current vehicle that first perceives and reports the risk, but also all target vehicles that meet the preset distance conditions from the risk point.

[0055] c i The confidence score of a high-risk event, calculated independently by the i-th vehicle based on its own local sensor data, is the basic input value for multi-vehicle fusion calculation.

[0056] w i The formula for calculating the comprehensive weighting coefficient corresponding to the i-th vehicle is w. i =α×(1 / d i )+β×q i +γ×r i The weighting coefficient is obtained by coupling three dimensions: distance, hardware device, and historical reputation.

[0057] Where, d iThis refers to the spatial distance between the i-th vehicle and the risk point, expressed in meters. The effective range is limited to 0.1 meters to 200 meters, corresponding to the effective coverage area of ​​V2V communication. To avoid the anomaly of a denominator of zero in division operations, the system sets a minimum distance of 0.1 meters. This distance can be calculated using the GPS latitude and longitude of the two vehicles combined with the Havesing formula, or by directly reading the relative vehicle position data output by V2V communication protocols such as C-V2X. The smaller the distance value, the higher the risk point. i The larger the calculation result, the higher the reference weight of the corresponding vehicle observation result, which fully reflects the spatial logic that the near-range vehicle perception information is more reliable.

[0058] q i The system assigns a standardized sensor quality score to the i-th vehicle, ranging from 0 to 100 points. These scores represent static hardware parameters determined at the time of vehicle manufacturing and network access. The system pre-defines scoring rules for various sensor types. For cameras, scores are based on resolution and frame rate: 720p cameras receive a base score of 60 points, 1080p 80 points, and 4K 100 points; a frame rate below 15fps deducts 10 points, while 30fps or higher adds 10 points. For tire pressure sensors, scores are based on measurement accuracy and sampling frequency: accuracy ±0.1 bar earns 90 points, ±0.2 bar earns 70 points, and accuracy greater than ±0.3 bar earns only 50 points; sampling frequency of 1Hz earns 70 points, 5Hz earns 90 points, and 10Hz earns 100 points. For inertial measurement units (IMUs)... The Inertial Unit (IMU) is scored based on accelerometer accuracy and data update frequency. Accelerometer accuracy of ±0.05g is scored as 90 points, ±0.1g as 70 points, and update frequency of 50Hz as 70 points, 100Hz as 90 points, and 200Hz as 100 points. When a vehicle first connects to the fleet network, it uploads a complete list of hardware devices, and the system automatically matches the rules to generate a score. i The score is then stored in the vehicle's file, and can be updated synchronously when subsequent camera and other sensor hardware are upgraded via Over-the-Air (OTA) technology.

[0059] r i It is the historical risk identification accuracy of the i-th vehicle dynamically updated, which is also the vehicle credit score. The score is also in the range of 0 to 100. The system defaults to an initial baseline score of 80 points and iterates in real time by relying on a feedback-driven exponential smooth sliding window mechanism.

[0060] α, β, and γ are the distance dimension weighting coefficient, sensor quality weighting coefficient, and vehicle reputation weighting coefficient, respectively. They satisfy the constraint α + β + γ = 1, and all three are greater than or equal to 0. These three sets of coefficients are customizable parameters that can be calibrated according to the different operating scenarios of two-wheeled vehicles. For urban food delivery fleet scenarios, priority is given to short-range vehicle perception, with α=0.5, β=0.2, and γ=0.3. The highway patrol convoy vehicles generally have excellent hardware, and the equipment weights are increased, with α=0.3, β=0.4, and γ=0.3. The quality of shared electric bicycle equipment varies greatly, and the accumulation of vehicle reputation is slow. Therefore, α=0.4, β=0.3, and γ=0.3 are used. The newly formed fleet does not yet have sufficient vehicle reputation data, so the distance weight is set to the maximum, with α=0.6, β=0.3, and γ=0.1.

[0061] In addition to fixed-scenario assignment, the system also supports two tuning methods: offline training and online adaptive optimization. In the offline stage, massive amounts of real riding data are collected, and the correspondence between vehicle-reported events and real road risks is fitted through logistic regression and lightweight neural networks to solve for the optimal α, β, and γ. During online operation, the system continuously calculates the global false alarm rate. If the false alarm rate is too high, the reputation coefficient γ is automatically increased to give more weight to data from vehicles with reliable historical records. If spatial distance has a greater impact on risk judgment, α is increased. The three initial weight values ​​are fixed in the system configuration file and support firmware over the air (FOTA) remote updates. High-end models can also enable a self-learning mode to dynamically correct the coefficients in real time.

[0062] The entire weighted fusion model differs from the simple majority voting logic. It comprehensively considers three dimensions: vehicle distance, sensor hardware accuracy, and historical alarm reliability. It assigns reasonable weights to the credibility scores of each participating vehicle, and the weighted sum is used to obtain the final confidence score, which effectively filters out false alarms for individual vehicles and greatly improves the accuracy of road risk assessment.

[0063] In one possible implementation, when carrying out the weighting coefficient w i Before fusion calculation, the raw values ​​of the three dimensions—distance, sensor quality, and vehicle reputation—need to be normalized to eliminate calculation biases caused by inconsistencies in the units and ranges of different indicators. The specific processing steps are as follows: The first step is to normalize the distance parameter. To reflect the non-linear characteristic that closer distances have greater weight, the reciprocal of the distance, Inv, is first calculated. Di =1 / d i Set the minimum safe distance D. min (e.g., 1 meter) and maximum effective communication distance D max (For example, 200 meters, which can be adjusted according to actual communication capabilities). At this point, the reciprocal of the distance ranges from [1 / D]. max 1 / D min ]. Through formula d norm=(Inv Di -1 / D max ) / (1 / D min -1 / D max) After normalizing the reciprocal of the distance, a distance score d is obtained, with values ​​ranging from 0 to 1. norm This processing method achieves the effect of the closer the distance is to the risk point (d i The smaller, Inv Di The larger the distance, the closer the normalized distance score is to 1; the farther the distance, the closer the score is to 0. If the calculated result is less than 0, it is taken as 0; if it is greater than 1, it is taken as 1.

[0064] The second step is the normalization of the sensor quality score. This involves normalizing the original sensor quality score q, which ranges from 0 to 100. i Divide by 100 to normalize, and you get q, which has a value between 0 and 1. norm , i.e. q norm =q i / 100.

[0065] The third step is the normalization of vehicle credit scores. This involves normalizing the original vehicle credit scores (r) within the 0-100 range. i Divide by 100 to normalize, and obtain r with a value ranging from 0 to 1. norm That is, r norm =r i / 100.

[0066] The fourth step is the final weight calculation. After normalization, the distance, quality, and reputation dimensions are all within the same dimension range of 0 to 1. The corrected final weight calculation formula is: wi = α × d norm +β×q norm +γ×r norm Where α, β, and γ are adjustable weight coefficients, and satisfy α+β+γ=1.

[0067] After the above unified normalization process, the three indicators—distance, hardware capability, and historical reputation—all fall within the same numerical range (0 to 1), ensuring that these three dimensions are equally comparable during the fusion decision-making process. Meanwhile, due to the distance score d... norm It is based on reciprocal calculation, which retains the nonlinear characteristic of the sharp increase in the weight of vehicles in close proximity, and effectively avoids the weight bias and calculation distortion caused by the difference in the range of the original index values ​​(such as meters and fractions).

[0068] In one possible implementation, the system also includes a human-computer interaction module, which is used to feed back risk warning information to riders in the cyclist group. The human-computer interaction module specifically includes visual interaction devices, auditory interaction devices, and smart handlebars, forming a multimodal risk warning system integrating visual, auditory, and tactile senses.

[0069] In one possible implementation, the visual interaction device may be, but is not limited to, augmented reality (AR) glasses; that is, the auditory interaction device may be, but is not limited to, bone conduction headphones.

[0070] In one possible implementation, when the overall confidence level corresponding to a high-risk event is greater than or equal to a preset first scoring threshold, all three types of terminals simultaneously initiate early warning output: Among them, visual interaction devices can project various early warning information of high-risk events onto the cyclist's real field of vision using augmented reality layers. The displayed content covers the real-time distance of the risk point, the type of risk event, the comprehensive confidence level, and corresponding deceleration and avoidance operation instructions. The cyclist does not need to look down at the onboard screen throughout the entire process, keeping their eyes on the road surface and intuitively perceiving the risks of the road conditions ahead. Figure 3 As shown, Figure 3 This application provides an AR interface diagram to visually demonstrate the integrated risk warning, avoiding visual interference to the driver from multiple repeated alarms. The screen simulates the real-time display effect of AR glasses or a smart helmet, preserving the entire main field of view of the real road with good transparency, ensuring unobstructed visibility for cyclists. The interface uses a semi-yellow highlighted frame to highlight the "continuous potholes" risk area approximately 60 meters ahead. Standardized natural language warning text is displayed centered at the top of the screen: "Continuous potholes exist 60 meters ahead, confirmed by 3 vehicles; it is recommended to reduce speed to below 25 km / h," clearly indicating the risk distance, event type, number of vehicles involved in the verification, and corresponding deceleration suggestions. A small icon in the upper right corner continuously displays the "Collaborative Perception" status, allowing cyclists to intuitively understand that the current risk has been cross-verified by multiple vehicles in the convoy, improving the credibility of the warning. All warning content uses an unobstructed layout, avoiding lane lines, traffic lights, and other key road elements, ensuring the driver's observation of the actual road conditions is not affected.

[0071] At the same time, the bone conduction headphones (i.e., auditory interaction devices) simultaneously broadcast voice messages, including the distance and location of the risk, the risk level, and suggestions for slowing down and avoiding it. Relying on the bone conduction sound generation structure, there is no need to block the ear canal. During the ride, the headphones can still capture environmental sounds such as road horns and pedestrian warnings, without isolating external road conditions and ensuring the ability to perceive the riding environment.

[0072] In addition, the smart handle relies on tactile vibration to distinguish the level of risk credibility. If the overall confidence level is determined to be high, it will output a high-frequency vibration warning; if the overall confidence level is in the medium confidence level, it will output a low-frequency vibration reminder, relying on the tactile sense of the hand to complete the auxiliary warning without visual interference.

[0073] The entire multimodal human-computer interaction solution is tailored to the usage scenarios of two-wheeled vehicle riders, where the line of sight cannot be lowered for extended periods and road conditions are complex and ever-changing. It simultaneously pushes risk alerts from three dimensions: visual, auditory, and tactile, significantly improving the efficiency of warning reception and the safety of riding.

[0074] In one possible implementation, the system distinguishes between two types of high-risk events: visual and non-visual, and configures corresponding computation functions to calculate the credibility score c for each vehicle. i It adapts to the output characteristics of different sensing devices.

[0075] If the currently identified high-risk event is a visual event, then function c is used. i =f(Model Score Environmental Factor Complete the credibility calculation, where Model Score The probability value of target existence output by the vehicle-mounted AI target detection model after parsing camera images; Environmental Factor The environmental factor correction coefficient is used to reduce the probability of the original model in adverse scenarios such as insufficient lighting, rain, fog, backlight, and nighttime, which can significantly reduce image recognition accuracy. The operation logic of the function f() is to weight and fuse the original output probability of the model with the environmental correction coefficient, and finally output the standardized confidence score c. i ; If the currently identified high-risk event is a non-visual event, then the piecewise mapping function c is used. i =g(IMU Peak TirePressure Change The calculation is performed at Threshold, where the IMU Peak TirePressure collects and outputs peak vehicle vibration data in real time from the IMU to identify road condition risks such as potholes and severe bumps. Change The instantaneous tire pressure change, collected by the tire pressure sensor, is used to identify vehicle malfunction risks such as tire leaks and abnormal tire pressure. The Threshold system has a pre-configured risk assessment benchmark threshold. The function g() performs a piecewise mapping operation based on the vibration peak value and the deviation range of the tire pressure change from the benchmark threshold. It matches the corresponding score range based on the magnitude of the data deviation from the threshold and outputs the vehicle's confidence score c for the current non-visual risk event. i .

[0076] It should be noted that the environmental factor used in the calculation of visual events is... FactorThe light intensity coefficient is obtained by multiplying the light intensity coefficient, the weather influence coefficient, and the vehicle speed stability coefficient. The light intensity coefficient is 1.0 under normal daylight conditions, 0.8 under nighttime or tunnel conditions, and 0.7 under strong light and backlight conditions. The weather influence coefficient is 1.0 for sunny and dry roads, 0.9 for wet roads in light rain, and 0.7 for roads in heavy rain and snow. The vehicle speed stability coefficient is 1.0 when the vehicle speed is maintained stably between 10-40 km / h without drastic acceleration or deceleration, and 0.8 when the vehicle is in a state of rapid acceleration or deceleration.

[0077] Furthermore, the function f() has two fusion operation methods: one is to directly convert the original output probability Model of the AI ​​model. Score With environmental factors Factor The second method uses a multiplication factor to obtain the corrected reliability, and the third method uses 0.8 × Model. Score +0.2×Environmental Factor The weighted fusion method is used to correct the credibility, for example, when Model Score When the environmental factor E is 0.9 and the operating conditions are during a nighttime rainstorm, the calculated result is 0.56, and the corrected confidence level after multiplication is only 0.504. To standardize the dimensions, this corrected confidence level is multiplied by 100 to convert it into a local confidence level c on a scale of 0 to 100. i That is, c i =50.4 points. This value reduces the reference weight of the single-vehicle perception results, and secondary risk confirmation must be completed through multi-vehicle collaborative verification within the fleet.

[0078] The function g() for non-visual events will first perform calculations based on peak vehicle vibration data from the IMU. Peak Instantaneous tire pressure change (TirePressure) Change Impact duration T dur Complete the validity pre-screening. Threshold includes the vibration threshold. V (e.g., 0.5g) and tire pressure change threshold (Threshold) P (e.g., 0.05 bar). Impact duration T dur It refers to IMU Peak Exceeding the vibration threshold V The continuous duration, measured in milliseconds (ms), represents the complete time window during which road potholes and bumps impact the vehicle body.

[0079] If the impact duration T dur If the time is less than 10ms, it is judged as high-frequency noise, and the confidence level is c. iSet directly to 0, if IMU Peak Less than the vibration threshold. V And TirePressure Change Less than the tire pressure change threshold (Threshold) P If so, it is determined to be a minor vibration interference, with a confidence level of c. i Also set to 0. The data is pre-filtered (i.e., the impact duration T). dur ≥10ms, and IMU Peak TirePressure Change (If the vibration scores are not both below the corresponding thresholds at the same time), the system will calculate the vibration score S separately. v Tire pressure score S p Among them, the vibration score S v The calculation formula is min(100, (IMU)). Peak -0.5) / 2.0×100)), when IIMU Peak When S reaches 2.5g or more v A score of 100 is taken as the maximum; tire pressure score S p The calculation formula is min(100, (TirePressure)). Change -0.05) / 0.2×100)), TirePressure Change When S reaches 0.25 bar or above p A score of 100 is awarded.

[0080] Final non-visual event credibility c i According to 0.6×S v +0.4×S p Weighted fusion output, for example, IMU data collected when a vehicle travels over severely potholed roads. Peak =2.0g, TirePressure Change =0.15 bar, calculated S v =75、S p =50, final c i =0.6×75+0.4×50=65 points. This average score only represents a suspected risk. It is still necessary to combine the results of multi-vehicle collaborative verification of the fleet to make a comprehensive judgment on whether a formal risk warning has been triggered.

[0081] In one possible implementation, the group-coordinated load reduction module also has the ability to dynamically update the vehicle's historical accuracy r. i The module's function is as follows: when the calculated comprehensive confidence level of a high-risk event is greater than or equal to a preset first scoring threshold, confirming that the risk warning has effective reference value, the module will use a sliding window combined with an exponential smoothing update algorithm to iteratively report the historical accuracy r of the risky vehicle in real time. iThis will form a dynamic vehicle credit evaluation system.

[0082] The iterative update formula for vehicle history accuracy is: r(t) = (1-η)×r(t-1) + η×a t .

[0083] In the formula, r(t) represents the vehicle's historical accuracy updated at time t, and r(t-1) represents the baseline historical accuracy value retained for the vehicle in the previous period; a t This is the accuracy binary identifier corresponding to this risk alert, which only has two values: 0 and 1. t =1 indicates that the risk alarm reported by this vehicle has been verified as true by multiple vehicles in the fleet. t =0 indicates that this alarm is a false alarm; η is the system's preset learning rate, with a value range limited to 0.1~0.2, which smoothly controls the rise and fall of historical accuracy and avoids drastic fluctuations in reputation score caused by a single alarm result.

[0084] The specific update logic is as follows: If multiple vehicles in the fleet that have passed through the risk point complete local perception verification and jointly confirm the existence of the high-risk event, then the `at` value is assigned to 1. After iteration using the exponential smoothing formula, the historical accuracy of that vehicle will steadily increase. If subsequent verification determines that the reported risk is a false event or a false alarm manually marked by the driver, then the `at` value is assigned to 0, and the historical accuracy of that vehicle will decrease after calculation using the formula. This dynamic update mechanism can continuously record the reliability of each vehicle's past risk reports, which will then be used as the weighting coefficient w for subsequent calculations. i The calculation provides dynamic reputation dimension data, filters out vehicle perception information that frequently generates false alarms over a long period of time, and further improves the accuracy of the comprehensive confidence determination after multi-vehicle collaborative fusion.

[0085] For example, in a fleet networking scenario, vehicle A is the current vehicle reporting a risk, while vehicles B, C, D, E, and F in the same fleet are target verification vehicles that meet the distance requirements. Vehicle A, as the current vehicle, broadcasts a road pothole risk report to the fleet. The five target verification vehicles B, C, D, E, and F successively pass through the risk location, each completing the detection using their local sensing modules, all of which corroborate the authenticity and validity of the pothole risk. After each round of valid verification feedback is completed, the identifier 'a' is updated. t Assigning a value of 1, and after multiple rounds of exponential smoothing iterations, the historical accuracy r of the current vehicle A, which initiates the risk, is... i The score is gradually increased to 95 points; conversely, if another vehicle G in the convoy reports a risk event, and none of the other target verification vehicles detect the corresponding risk at the points along their routes, and the driver manually marks the warning as a false alarm, the score for this round of updates is a. t The value is assigned to 0, which serves as the historical accuracy r of the reporting vehicle G. i It then declined.

[0086] In one possible implementation, the group collaborative load reduction module also possesses the capability for risk message deduplication and integration: the system pre-sets a fixed time window as the judgment period. If, within this pre-set time period (e.g., within 10 seconds), the module continuously receives multiple risk messages uploaded by multiple vehicles in the fleet, all pointing to the same geographical area and with completely identical risk event types, then a geographic location clustering algorithm is activated. Simultaneously, semantic-level deduplication and integration processing is performed on the multiple messages from the same source, combined with high-risk event type tags, eliminating duplicate and redundant reported information and generating a unique standardized merged message. Subsequently, when calculating the comprehensive confidence level corresponding to the risk of this road segment, the system no longer parses the original multiple duplicate messages one by one, but instead uniformly relies on the integrated merged message to complete the weighted calculation of multi-vehicle confidence.

[0087] This mechanism addresses scenarios where multiple vehicles repeatedly report risks on the same road surface. It significantly reduces the computational overhead of redundant data, avoids repeatedly parsing similar risk information and consuming vehicle computing power, and prevents multiple identical warnings from being pushed simultaneously, thus preventing driver overload. It truly achieves streamlined convergence of multi-source risk information and reduces information interference during riding.

[0088] In one possible implementation, once the overall confidence level is met and the road segment risk is confirmed to be covered by multi-vehicle collaboration within the fleet, the group collaborative load reduction module reduces the operating frequency of the local sensing modules. This adjustment is not a uniform frequency reduction, but rather a differentiated resource scheduling strategy for various sensing devices within the module. Specific scheduling details are as follows: the original real-time acquisition frame rate of the cameras is reduced to a preset low frame rate, exemplarily 10fps; the AI ​​target detection model associated with the cameras is simultaneously disabled, retaining only basic image acquisition capabilities and ceasing real-time image recognition inference calculations; for tire pressure sensors, the original first sampling cycle of once per second is extended to a second sampling cycle of 5 seconds, reducing the frequency of tire pressure data acquisition; for IMU units, on the one hand, the overall hardware original sampling rate is reduced, and on the other hand, the high-frequency vibration analysis background calculation logic is directly shut down, retaining only basic vibration data acquisition. Through differentiated load reduction scheduling for each device, while relying on fleet collaborative information to ensure risk identification capabilities, the computing power and power consumption caused by image AI inference and high-frequency sensor sampling are significantly reduced, achieving energy-saving operation of the two-wheeled vehicle's onboard equipment and extending the vehicle's range.

[0089] In addition, this application embodiment also provides a vehicle, which includes the risk warning and coordination system for two-wheeled vehicles as described above.

[0090] This application constructs a fleet collaborative perception system through V2V dynamic networking. First, local perception modules identify high-risk road events and generate standardized risk messages, which are then broadcast to other vehicles in the fleet. Next, target vehicles whose distance to the risk point meets a threshold are selected to obtain multi-source verification data. A group collaborative load reduction module then integrates the credibility scores of each vehicle with the multi-vehicle verification information to calculate a comprehensive confidence level. This effectively overcomes the shortcomings of existing solutions that rely solely on single-vehicle sensing, leading to insufficient reliability in risk assessment. Furthermore, once the comprehensive confidence level is met and the risk is confirmed to be genuine and reliable, the operating frequency of the local perception modules is proactively reduced. This addresses the problem of high onboard computing power and hardware power consumption burden caused by traditional sensing devices collecting massive amounts of data at high frequencies over long periods, balancing the requirements for risk identification accuracy with the lightweight operation of onboard equipment.

[0091] The foregoing has provided a detailed description of a risk warning and collaborative system for two-wheeled vehicles provided in this application. The various embodiments in the specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0092] It should also 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, system, 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 a process, system, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, system, article, or apparatus that includes said element.

[0093] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A risk warning and collaborative system for two-wheeled vehicles, characterized in that, Each two-wheeled vehicle connects to the target fleet through a dynamic networking mechanism. The system is mounted on each two-wheeled vehicle in the target fleet and includes a local perception module, a V2V communication module, and a group collaborative unloading module. The local perception module is used to collect the first surrounding environment data of the current vehicle and identify whether a first high-risk event has occurred during the current vehicle's driving process based on the first surrounding environment data. The first high-risk event is a road safety risk event; The group collaborative load reduction module is used to calculate the credibility score of the first high-risk event; If the credibility score is greater than or equal to the credibility threshold, a first risk message is generated based on the first risk data output by the local perception module and the credibility score of the first high-risk event. The first risk message carries the risk location and the credibility score of the first high-risk event; The first risk data includes the first risk event type and the coordinates of the risk point; The V2V communication module is used to broadcast the first risk message to the remaining vehicles in the target convoy and receive the first verification data returned by the first target vehicle in the target convoy; the remaining vehicles are all vehicles in the target convoy except the current vehicle; the first target vehicle is the vehicle among the remaining vehicles whose distance from the risk point is less than or equal to a distance threshold. The group collaborative load reduction module is also used to calculate the first comprehensive confidence level of the first high-risk event by combining the current vehicle's confidence score for the first high-risk event and the first verification data of the first target vehicle. If the first comprehensive confidence level is greater than or equal to the first scoring threshold, a risk warning is issued, and the operating frequency of the local perception module in the current vehicle is reduced.

2. The system according to claim 1, characterized in that, The V2V communication module is also used to receive a second risk message broadcast by any remaining vehicle; the second risk message is a risk message generated and broadcast by any remaining vehicle after identifying a second high-risk event. When the V2V communication module receives the second risk message, the group cooperative load reduction module is also used to calculate the distance between the current vehicle and the risk point carried in the second risk message to obtain the risk distance; if the risk distance is greater than the distance threshold, it is determined that the current vehicle is not within the effective perception range, and the second risk message is discarded; If the risk distance is less than or equal to the distance threshold, then it is determined that the current vehicle is within the effective perception range, and its own local perception module is activated to collect the second surrounding environment data of the current vehicle. Based on the second surrounding environment data, it is identified whether the second high-risk event is a real event. If the second high-risk event is determined to be a real event, the group collaborative load reduction module is also used to calculate the credibility score of the second high-risk event; The V2V communication module is also used to receive second verification data returned by a second target vehicle in the target convoy; the second target vehicle is a vehicle among the remaining vehicles that, excluding those that broadcast the second risk message, is less than or equal to the distance threshold from the risk point. The group collaborative load reduction module is also used to calculate the second comprehensive confidence level of the second high-risk event by combining the current vehicle's confidence score for the second high-risk event, the second risk message, and the second verification data of the second target vehicle; If the second comprehensive confidence level is greater than or equal to the first scoring threshold, then the operating frequency of the local perception module in the current vehicle is reduced.

3. The system according to claim 1 or 2, characterized in that, The formula for calculating the overall confidence level is as follows: FinalConfidence=Σ(in i ×c i ); Where FinalConfidence is the overall confidence level; c i The credibility score output by the i-th vehicle for high-risk events; w i w is the weighting factor for the rating of the i-th vehicle. i =α×(1 / d i )+β×q i +γ×r i ;d i Let q be the distance between the i-th vehicle and the risk point; i For the sensor quality score of vehicle i, r i Let be the historical accuracy of vehicle i; α be the relative distance weighting coefficient of vehicles; β be the sensor quality score weighting coefficient; γ be the vehicle reputation score weighting coefficient; i iterates through all vehicles participating in the confidence calculation of this high-risk event, and each vehicle i is the current vehicle for identifying the high-risk event or the target vehicle that meets the distance condition.

4. The system according to claim 1 or 2, characterized in that, The system also includes a human-computer interaction module, which includes a visual interaction device, an auditory interaction device, and a smart handle. When the overall confidence level is greater than or equal to the first scoring threshold, the visual interaction device is used to overlay and display the relevant early warning information of high-risk events in the driver's field of vision in an augmented reality form; When the overall confidence level is greater than or equal to the first scoring threshold, the auditory interaction device is used to deliver high-risk event warning information to the driver in the form of voice broadcast; When the overall confidence level is greater than or equal to the first scoring threshold, the smart handlebar is used to transmit a high-risk event warning signal to the driver through vibration feedback.

5. The system according to claim 1 or 2, characterized in that, When the high-risk event is a visual event, the credibility score c i The calculation formula is as follows: c i =f(Model Score Environmental Factor Model Score The probability value output by the AI ​​object detection model, Environmental Factor Here, f() represents the environmental factor correction coefficient, and f() denotes the confidence score c obtained by weighted fusion of the model output probability value and the environmental correction coefficient. i ; When the high-risk event is a non-visual event, the credibility score c i The calculation formula is as follows: c i =g(IMU Peak TirePressure Change Threshold), IMU Peak TirePressure provides peak vibration data output from the inertial measurement unit (IMU). Change The instantaneous tire pressure change collected by tire pressure monitoring is represented by Threshold, which is a preset risk assessment benchmark threshold. g() represents the confidence score c output by performing a piecewise mapping calculation combining the vibration peak value, tire pressure change, and benchmark threshold. i .

6. The system according to claim 1 or 2, characterized in that, The group-coordinated load reduction module is also used for: If the overall confidence level is greater than or equal to the first scoring threshold, a sliding window combined with an exponential smoothing update algorithm is used to update the vehicle's historical accuracy r in real time. i ; Among them, the vehicle history accuracy rate r i The update formula is: r(t)=(1-η)×r(t-1)+η×a t r(t) represents the vehicle's historical accuracy at time t, and r(t-1) represents the vehicle's historical accuracy at the previous time step; a t This is the accuracy indicator for this risk alert; η is the preset learning rate; when multiple vehicles in the fleet verify and confirm the existence of a high-risk event, a t Assigning a value of 1 corresponds to a gradual increase in the vehicle's historical accuracy; when a risk event is determined to be a false alarm, a... t Assigning a value of 0 will decrease the accuracy of the vehicle history for the corresponding vehicle.

7. The system according to claim 1 or 2, characterized in that, The verification data includes the confidence score of the target vehicle's output for the high-risk event, the sensor quality score of the target vehicle, the vehicle's historical accuracy, and the distance between the target vehicle and the risk point.

8. The system according to claim 1, characterized in that, The group-coordinated load reduction module is also used for: When the overall confidence level is less than the first scoring threshold but greater than or equal to the second scoring threshold, the high-risk event is determined to be a medium-confidence risk event. The event is marked as pending observation, and no formal warning is pushed out. We continue to wait for the perception data of other vehicles in the fleet to complete the secondary verification. When the overall confidence level is less than the second scoring threshold, the high-risk event is determined to be a low-confidence suspected risk, and the risk message is directly discarded, or only a record is kept in the local background.

9. The system according to claim 1, characterized in that, The group-coordinated load reduction module is also used for: If multiple vehicles report risk messages targeting the same area and with the same risk event type within a preset time period, the multiple risk messages are semantically deduplicated and integrated based on a geographic location clustering algorithm and high-risk event type to obtain a merged message. Subsequent comprehensive confidence calculation is then performed based on the merged message.

10. The system according to claim 1 or 2, characterized in that, Reducing the operating frequency of the local sensing module specifically includes: performing differentiated scheduling for various sensing devices in the local sensing module; The differentiated scheduling includes reducing the real-time frame rate of the camera to a preset frame rate, disabling the AI ​​detection model and retaining only the basic image acquisition function of the camera, extending the tire pressure sampling cycle from the first sampling cycle to the second sampling cycle, reducing the sampling rate of the IMU, and turning off the high-frequency vibration analysis function of the IMU.

11. A vehicle, characterized in that, The vehicle includes a risk warning and collaborative system based on a two-wheeled vehicle as described in any one of claims 1-10.

Images