Commercial vehicle safety monitoring method and system based on cloud edge collaboration

Abstract

The invention relates to the field of vehicle monitoring, in particular to a commercial vehicle safety monitoring method and system based on cloud edge collaboration. According to the scheme provided by the invention, the analysis of the driver state and the analysis of the driving behavior of the vehicle are realized through the vehicle end, and the weighted decision of the vehicle risk level is completed in combination with the environment context information sent by the cloud end. According to the invention, through multi-modal data fusion, weighted decision making is carried out on the driver state, the driving behavior and the environment information, the defect of false triggering of a single signal is avoided, and the early warning accuracy is greatly improved; the weight and the threshold value of the early warning algorithm can be dynamically adjusted according to information such as real-time weather, road topology and historical risk road spectrum, so that the early warning strategy is more suitable for an actual driving scene, and personalized and predictive early warning is realized.

Description

Technical Field

[0001] This invention relates to the field of vehicle monitoring, and more specifically, to a method and system for commercial vehicle safety monitoring based on cloud-edge collaboration. Background Technology

[0002] In recent years, with the rapid development of vehicle-to-everything (V2X) and cloud computing technologies, the operation and management of commercial vehicles are undergoing profound intelligent transformation. By equipping various sensors and connecting to the network, vehicles can upload operational data to the cloud platform in real time, providing a solid technical foundation for the digital management of fleets. The powerful storage and computing power of cloud computing makes it possible to conduct in-depth analysis and intelligent processing of massive amounts of vehicle data, opening up new paths for improving the safety and operational efficiency of commercial vehicles.

[0003] Commercial buses are characterized by large passenger capacity, long operating hours, and complex road conditions, resulting in extremely high safety monitoring requirements. Existing monitoring methods often employ functionally independent systems; for example, Driver Status Monitoring (DMS) and Advanced Driver Assistance Systems (ADAS) are frequently separated, creating data silos and lacking collaborative analysis of multi-dimensional information on "people, vehicles, roads, and the environment." This leads to high false alarm rates, rigid warning strategies, an inability to dynamically adjust based on actual risks, and reliance on multi-sensor stacking solutions that increase hardware costs and deployment complexity, hindering large-scale application in the aftermarket. Achieving low-cost, high-accuracy panoramic perception and intelligent early warning has become a critical technological bottleneck that the industry urgently needs to overcome. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for monitoring the safety of commercial vehicles with context awareness and adaptive intelligent decision-making capabilities.

[0005] According to a first aspect of the present invention, a cloud-edge collaborative method for commercial vehicle safety monitoring is proposed, applied to the vehicle end, comprising: SA1: Acquire multi-dimensional perception data of this vehicle, including in-cabin video data, vehicle CAN bus data, and vehicle positioning data; SA2: Analyze the driver's state based on in-cabin video data to obtain driver state characteristic values; SA3: Analyzes driving behavior based on vehicle CAN bus data to obtain driving behavior feature values; SA4: Sends an environmental information request to the cloud platform and receives environmental context information returned by the cloud platform. The environmental information request includes vehicle positioning data, and the environmental context information includes at least real-time weather information and road topology information. SA5: Based on preset fusion rules, it makes weighted decisions based on driver state feature values, driving behavior feature values, and environmental context information to determine the current risk level of the vehicle; SA6: Execute corresponding graded early warning operations based on the risk level and send the risk level to the cloud platform.

[0006] According to some embodiments, in the method of the first aspect of the present invention, step SA5 includes: The first and second weighting factors used in the weighted decision-making are dynamically adjusted based on environmental context information. The driver's state characteristic value and driving behavior characteristic value are weighted and summed according to the first weighting factor and the second weighting factor to obtain a comprehensive risk score; According to the fusion rules, rule reasoning is performed based on the comprehensive risk score to determine the risk level.

[0007] According to some embodiments, in the method of the first aspect of the present invention, the process of dynamically adjusting the first weighting factor and the second weighting factor used in the weighted decision includes: The road adhesion coefficient is obtained based on real-time weather information and road topology information; The weighting factor is dynamically adjusted based on the road adhesion coefficient, following these principles: the lower the road adhesion coefficient, the higher the value of the first weighting factor and the lower the value of the second weighting factor.

[0008] According to some embodiments, in the method of the first aspect of the present invention, the fusion rule includes a preset decision threshold, and the process of rule reasoning based on the comprehensive risk score includes: The current risk level is determined by comparing the comprehensive risk score with the decision threshold.

[0009] According to some embodiments, in the method of the first aspect of the present invention, the environmental context information further includes risk road spectrum information, which represents the road segment risk situation in a specific geographical area, and step SA5 further includes: The road risk level corresponding to the vehicle positioning data is determined based on the risk road spectrum information; The decision threshold used for rule-based reasoning is dynamically adjusted based on the road risk level, according to the following principle: the higher the road risk level, the lower the decision threshold required to trigger the same risk level.

[0010] According to a second aspect of the present invention, a cloud-edge collaborative method for commercial vehicle safety monitoring is proposed, applied to a cloud platform, comprising: SB1: Receives environmental information requests from the vehicle, including vehicle location data; SB2: Based on vehicle positioning data, obtain corresponding environmental context information from a third-party information service platform. The environmental context information includes at least real-time weather information and road topology information. SB3: Sends environmental context information to the vehicle.

[0011] According to some embodiments, the method of the second aspect of the present invention further includes: It receives driving behavior feature values ​​and their corresponding vehicle location data periodically reported from multiple vehicle terminals, forming a historical dataset; Based on historical datasets, geospatial analysis is performed using an unsupervised clustering algorithm to generate risk road spectrum information; Risk road map information is sent to vehicles in the relevant areas.

[0012] According to some embodiments, the method of the second aspect of the present invention further includes: Risk level transmitted by the receiving vehicle; When the risk level is high-risk and a warning is issued, remote intervention can be carried out on the vehicle through a real-time channel.

[0013] According to some embodiments, in the method of the second aspect of the present invention, obtaining the corresponding environmental context information from a third-party information service platform includes: The system calls third-party map services through a predefined application programming interface to obtain road topology information, and calls third-party meteorological services to obtain real-time weather information.

[0014] According to a third aspect of the present invention, a commercial vehicle safety monitoring system based on cloud-edge collaboration is proposed, comprising: a vehicle terminal, a cloud platform, and a third-party information service platform; The vehicle side includes: The data acquisition module is used to acquire multi-dimensional perception data of this vehicle; The edge computing module is used to calculate driver state feature values ​​and driving behavior feature values ​​based on multi-dimensional perception data; The vehicle-side communication module is used to communicate with the cloud platform; The decision-making and early warning module is used to make weighted decisions based on driver status characteristic values, driving behavior characteristic values, and environmental context information obtained from the cloud platform, generate risk levels, and execute early warnings. The cloud platform includes: The platform communication module is used to communicate with the vehicle and third-party information service platforms; The environment service module is used to obtain environmental context information from a third-party information service platform based on the vehicle's location and provide it to the vehicle. A third-party information service platform is used to provide road topology information and real-time weather information to the cloud platform.

[0015] The solution provided by this invention has the following beneficial effects: 1. Traditional commercial vehicle safety monitoring solutions operate with independent functional modules, preventing data sharing and hindering comprehensive risk assessment from a holistic perspective. This results in incomplete warnings and a high false alarm rate. This invention, through multimodal data fusion, weights driver status, driving behavior, and environmental information for decision-making, avoiding the pitfalls of false triggers from single signals. This significantly improves warning accuracy and enhances driver trust and acceptance. 2. Existing solutions mostly focus only on the vehicle and driver, failing to perceive the external environment and operational status, resulting in rigid early warning strategies that cannot be dynamically adjusted according to actual risks. This invention can dynamically adjust the weights and thresholds of the early warning algorithm based on real-time weather, road topology, and historical risk road spectrum information, making the early warning strategy more aligned with actual driving scenarios and achieving personalized and predictive early warnings; 3. To achieve multi-functionality, traditional solutions require a large number of dedicated sensors, resulting in high hardware costs, complex installation, and difficulty in large-scale promotion in the aftermarket. This invention empowers a limited number of sensors (one in-cabin camera) with AI algorithms, collaborating with a cloud platform in weighted decision-making. The cloud platform uses big data analysis and machine learning to adjust the weights of driver state characteristic values ​​and the decision threshold, greatly simplifying hardware configuration, reducing hardware costs and installation complexity per vehicle, and laying the foundation for large-scale fleet deployment. 4. Most system functions are fixed and cannot learn and optimize themselves using data accumulated during fleet operations. Furthermore, relying entirely on cloud-based data processing leads to warning delays, while relying solely on vehicle-side computing makes it difficult to obtain a global perspective. A balance needs to be struck between real-time performance and intelligent decision-making. This invention's solution, through a cloud-edge collaborative architecture, forms a closed-loop feedback loop of vehicle-side perception, cloud-based analysis, knowledge dissemination, and vehicle-side optimization. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without exceeding the scope of protection claimed by the present invention.

[0017] Figure 1 This is a flowchart illustrating an embodiment 1000 of the present invention, which applies a cloud-edge collaborative commercial vehicle safety monitoring method to a vehicle. Figure 2 for Figure 1 A flowchart illustrating step SA5 in embodiment 1000; Figure 3 for Figure 2 A flowchart illustrating step SA51 in the middle section; Figure 4 for Figure 1 A flowchart illustrating step SA5a, which dynamically adjusts the decision threshold, in step SA5 of embodiment 1000. Figure 5 This is a flowchart illustrating an embodiment 2000 of the present invention, which describes a cloud-edge collaborative method for monitoring the safety of commercial vehicles applied to a cloud platform. Figure 6 for Figure 5 A flowchart illustrating step SB2 in embodiment 2000; Figure 7 This is a flowchart illustrating an embodiment 3000 of the present invention, which describes a cloud-edge collaborative method for commercial vehicle safety monitoring applied to a cloud platform. Figure 8 This is a flowchart illustrating an embodiment 4000 of the present invention, which describes a cloud-edge collaborative method for monitoring the safety of commercial vehicles applied to a cloud platform. Figure 9 This is a schematic diagram of an embodiment 5000 of a cloud-edge collaborative commercial vehicle safety monitoring system according to the present invention. Detailed Implementation

[0018] 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, not all, of the embodiments of the present invention. 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.

[0019] See Figure 1 , Figure 1 This is a flowchart illustrating an embodiment 1000 of the present invention, which applies a cloud-edge collaborative commercial vehicle safety monitoring method to the vehicle.

[0020] In step SA1, a cloud-edge collaborative commercial vehicle safety monitoring system (e.g., a vehicle-mounted device containing a processor, hereinafter referred to as the vehicle-mounted device) acquires multi-dimensional perception data of the vehicle, including in-cabin video data, vehicle CAN bus data, and vehicle positioning data.

[0021] Optionally, in step SA1, a camera is deployed in the cockpit, positioned above the center of the windshield or on the top of the dashboard, using a wide-angle or fisheye lens, with a field of view covering the driver's face and steering wheel area, to capture the cabin video stream. Optionally, in practical applications of step SA1, the vehicle-mounted camera has infrared night vision capabilities, supporting the acquisition of cabin video at night. In step SA1, the vehicle-mounted camera performs video encoding of the cabin video stream, obtaining encoded cabin video frames, which constitute the cabin video data.

[0022] Optionally, in step SA1, the vehicle-mounted device reads the vehicle CAN bus data and vehicle positioning data directly through a gateway. Optionally, the vehicle-mounted device obtains the CAN bus data through CAN protocol parsing, specifically including vehicle speed, throttle, brake, steering angle, and turn signal status. Optionally, the vehicle positioning data includes the vehicle's current latitude and longitude data, obtained through the vehicle's built-in GPS / BeiDou positioning module.

[0023] In step SA2, the vehicle-mounted system analyzes the driver's state based on the in-cabin video data to obtain driver state feature values. In some specific embodiments, in step SA2, the vehicle-mounted system performs face detection, tracking, and key point detection, and analyzes the driver's state based on the detection results to obtain driver state feature values. Optionally, the driver state feature values ​​include fatigue level, distraction type, and confidence level. For example, fatigue level includes {0: alert, 1: mild, 2: moderate, 3: severe}, and distraction type includes {looking at mobile phone, looking around, looking down}.

[0024] In some specific embodiments, in step SA2, the vehicle end determines the fatigue level based on the proportion of eyelid closure time, blinking frequency, and yawning monitoring thresholds, and determines the distraction type and confidence level based on the rules of head posture angle estimation and gaze direction estimation.

[0025] In step SA3, the vehicle-side performs driving behavior analysis based on the vehicle's CAN bus data to obtain driving behavior feature values.

[0026] In some specific embodiments, in step SA3, the driving behavior feature values ​​include behavior labels and behavior intensity values. Optionally, the behavior labels include rapid acceleration, rapid deceleration, sudden braking, and sharp turning, and the behavior intensity value includes the rate of change of acceleration. Rapid acceleration and rapid deceleration are calculated by the longitudinal rate of change of acceleration and are triggered when they exceed a set first threshold. Sharp turning is calculated by the lateral acceleration or directly using the steering wheel angular velocity and is triggered when it exceeds a set second threshold. Optionally, in step SA3, the vehicle uses model-assisted detection to detect abnormal driving patterns, for example, by detecting driving behavior feature values ​​through isolated forest or unsupervised clustering.

[0027] In step SA4, the vehicle sends an environmental information request to the cloud platform and receives environmental context information returned by the cloud platform. The environmental information request includes vehicle location data, and the environmental context information includes at least real-time weather information and road topology information. In some specific embodiments, in step SA4, the vehicle sends an environmental information request containing vehicle location data to the cloud platform via an HTTP / HTTPS RESTful API call or the MQTT protocol.

[0028] In step SA5, based on preset fusion rules, the vehicle-side makes a weighted decision based on driver state feature values, driving behavior feature values, and environmental context information to determine the current risk level of the vehicle.

[0029] In some specific embodiments, in step SA5, the vehicle determines the risk level through the following steps: dynamically adjusting the first and second weighting factors used in the weighted decision based on environmental context information; weighting and summing the driver's state feature values ​​and driving behavior feature values ​​according to the first and second weighting factors to obtain a comprehensive risk score; and performing rule reasoning based on the comprehensive risk score according to fusion rules to determine the risk level. Optionally, the risk level determined in step SA5 includes level 1 alert, level 2 warning, and level 3 high risk.

[0030] In step SA6, the vehicle-side performs a corresponding tiered warning operation based on the risk level and sends the risk level to the cloud platform. Optionally, the tiered warning operation on the vehicle-side is executed automatically in real time to ensure minimal latency.

[0031] In some specific embodiments, step SA6, the graded early warning operation specifically includes: The risk level is level 1, indicating a low risk level. The warning operation includes: (1) gentle voice prompt: the vehicle speaker plays the prompt voice; (2) the dashboard displays the icon; The risk level is level 2 warning, which is a high level of risk. The warning operation includes: (1) strong sound and light alarm: triggering a continuous and rapid buzzing sound and flashing red warning on the display; (2) detailed and clear voice warning, forcibly attracting the driver's attention and requiring an immediate response; The risk level is level 3, which is the highest level of risk. The warning operations include: (1) the highest local alarm: the loudest volume beep and red flashing warning; optionally, triggering (2) the primary active intervention process. The primary active intervention triggered may include automatic activation of hazard lights, automatic unlocking of doors, etc.

[0032] In some specific embodiments, in step SA6, the vehicle-mounted device performs the warning operation while simultaneously organizing and reporting the data. Optionally, the vehicle-mounted device's reporting strategy includes: for Level 1 alerts, reporting is usually not performed or is delayed. The vehicle-mounted device stores these alerts locally and uploads them along with other batch data when the network is good or when scheduled for reporting, in order to save bandwidth; for Level 2 warnings and Level 3 high-risk events, immediate real-time reporting is performed as the highest priority network transmission task. Optionally, in step SA6, the reported data packet includes the vehicle ID, event timestamp, event type, risk level, comprehensive risk score, and vehicle location information.

[0033] According to such Figure 1 As shown in the implementation, the solution proposed in this invention uses multimodal data fusion to weight the driver's state, driving behavior, and environmental information to make decisions, avoiding the defects of false triggering by a single signal and greatly improving the accuracy of the warning. At the same time, by introducing environmental context information as the basis for weighted decision-making through cloud-edge fusion architecture, the warning strategy is more in line with the actual driving scenario, realizing personalized and predictive warnings.

[0034] Figure 2 for Figure 1 A flowchart illustrating step SA5 in embodiment 1000. (See attached diagram.) Figure 2 As shown, step SA5 includes steps SA51-SA53.

[0035] In some specific embodiments, in step SA51, the vehicle dynamically adjusts the first weighting factor and the second weighting factor used in the weighted decision based on the environmental context information. Optionally, in step SA51, the environmental context information includes real-time weather information and road topology information. Optionally, the real-time weather information includes {sunny, light rain, heavy rain, light snow, snow and ice}, and the road topology information includes {highway straight road, urban road, mountain bend, tunnel, school zone}.

[0036] Optionally, in step SA51, the first weighting factor corresponds to the driver's state feature value, expressed as: The second weighting factor corresponds to the driving behavior feature value, expressed as: Optionally, in step SA51, the vehicle end determines the road adhesion coefficient. Dynamic adjustments are made, following these principles: the lower the road adhesion coefficient, the greater the value of the first weighting factor and the less the value of the second weighting factor.

[0037] In some specific embodiments, in step SA52, the vehicle-side performs a weighted summation of the driver's state characteristic value and driving behavior characteristic value based on the first weighting factor and the second weighting factor to obtain a comprehensive risk score.

[0038] Optionally, a specific embodiment of step SA52 is: comprehensive risk scoring. ,in, For driver state characteristic values, These are driving state characteristic values. , The weighting factor is determined in step SA51.

[0039] In some specific embodiments, in step SA53, the vehicle-side performs rule reasoning based on the comprehensive risk score according to the fusion rules to determine the risk level. Optionally, the fusion rules in step SA53 include a preset decision threshold, and the process of rule reasoning based on the comprehensive risk score includes comparing the comprehensive risk score and the decision threshold to determine the current risk level. Optionally, the risk levels in step SA53 include level 1 alert, level 2 warning, and level 3 high risk.

[0040] According to such Figure 2 The embodiment shown in this invention introduces the physical concept of road adhesion coefficient as an intermediate bridge, making decision-making based on vehicle dynamics principles rather than purely mathematical mapping, thus making weight allocation more evidence-based; it solves the problem of multi-factor coupling, naturally and reasonably integrating weather and road type, and quantifying their cross-influence. According to Figure 2 The implementation shown enables the solution to evolve from rule-based intelligence to physical model-based intelligence, thereby improving the solution's professionalism and reliability.

[0041] Figure 3 for Figure 2 A flowchart illustrating step SA51. (See attached diagram.) Figure 2 The step SA51 shown includes steps SA511 and SA512.

[0042] In some specific embodiments, in step SA511, the vehicle obtains the road adhesion coefficient based on real-time weather information and road topology information. The road adhesion coefficient is a key vehicle dynamics parameter that directly determines the vehicle's maximum longitudinal (acceleration and braking) force and lateral turning force. It varies significantly on different road surfaces; the smoother the surface and the lower the resistance, the smaller the value of the road adhesion coefficient.

[0043] Optionally, in step SA511, a preset adhesion coefficient estimation function is set inside the vehicle end. The estimated adhesion coefficient is obtained by inputting real-time weather information and road topology information. For example, the adhesion coefficient estimation function is expressed as: ,in, The baseline adhesion coefficient is set based on a dry road surface in sunny weather; for example, it can be set to 0.9. This is a weather discount factor, with different values ​​for different weather conditions. For example, it is set to 0.75 for light rain and 0.6 for heavy rain. This is the road discount factor, set based on different road surface conditions. For example, it is set to 1.0 for flat straight roads and 0.9 for sharp curves due to the risk of centrifugal force.

[0044] Optionally, in step SA511, a road adhesion estimation table is pre-set inside the vehicle, including combinations of weather information and road types, as well as the corresponding range of estimated adhesion coefficient values. For example, a combination of weather conditions and road types, with an estimated adhesion coefficient of: {heavy rain, high-speed straight road, ...} =0.5-0.6}.

[0045] In step SA512, the vehicle end makes dynamic adjustments based on the road adhesion coefficient, following the principle that the lower the road adhesion coefficient, the higher the value of the first weight factor and the lower the value of the second weight factor.

[0046] In some specific embodiments, in step SA512, the vehicle-side uses a weight allocation function. Determine the values ​​of the first and second weighting factors. Optionally, the weighting function... It is a piecewise function, directly based on the input road adhesion coefficient. The value is used to define the first weight factor in intervals. Second weighting factor Values, for example: At this point, the road surface is dry and the grip is good; the main focus should be on the driver's condition. At this time, the road surface is slippery and grip is reduced, so the main focus should be on vehicle behavior; in other situations... At this point, the road surface was extremely slippery and had very poor grip, so almost all attention was focused on monitoring the vehicle's minute maneuvers.

[0047] Optionally, in step SA512, the vehicle-side uses a decreasing function to define a first weighting factor. And then according to This leads to a second weighting factor. For example, it increases the scaling factor of the road adhesion coefficient. and the first weighting factor The linear relationship is used as a decreasing function. .

[0048] Figure 4 for Figure 1 A flowchart illustrating step SA5a, which dynamically adjusts the decision threshold, in step SA5 of embodiment 1000. (See attached diagram.) Figure 4 As shown, step SA5a includes steps Sa1 and Sa2.

[0049] Optionally, the environmental context information obtained in step SA4 also includes risk road spectrum information, representing the road segment risk situation in a specific geographical area. Specifically, risk road spectrum information can be understood as an electronic map layer generated by a cloud platform through analysis of massive amounts of historical data reported by vehicles. The historical data used includes the locations of events such as rapid acceleration, sudden braking, sharp turns, and lane departures. Each road segment or geographical area in the generated risk road spectrum information is assigned a historical risk level. For example, a historical risk level classification includes: level 0 (low risk): no or very few abnormal event clusters; level 1 (medium risk): a considerable number of abnormal events; level 2 (high risk): a large number of abnormal event clusters.

[0050] In some specific embodiments, in step Sa1, the vehicle determines the road risk level corresponding to the vehicle positioning data based on the risk road spectrum information. Optionally, in step Sa1, the vehicle compares its current vehicle positioning data with the risk status of the geographical area in the risk road spectrum information to determine the road risk level of the current location.

[0051] In some specific embodiments, in step Sa2, the vehicle dynamically adjusts the decision threshold used for rule-based reasoning according to the road risk level, based on the principle that the higher the road risk level, the lower the decision threshold required to trigger the same risk level. In step Sa2, the principle of adjusting the decision threshold on the vehicle side means that in high-risk road sections, the monitoring method becomes more sensitive and more likely to trigger alarms, thereby achieving intelligent control based on historical risks.

[0052] In some specific embodiments, the vehicle-side system pre-sets baseline decision thresholds to determine the risk level under normal road conditions. For example, the decision thresholds include a prompt threshold, a warning threshold, and a high-risk threshold, with corresponding comprehensive risk scores of 60, 75, and 90, respectively.

[0053] Optionally, step Sa2 is implemented using a linear discount model, i.e. .in, This is the adjusted actual threshold. As the baseline threshold, The historical risk level is quantified into a specific numerical value; K is the sensitivity coefficient, which determines the degree of influence of historical risk on the decision threshold.

[0054] Optionally, in the linear discount model of step Sa2, the historical risk level The quantification is achieved by the cloud platform through the establishment of a mapping table. Using an equal-interval mapping method, the lowest and highest risk levels are mapped to the values ​​0 and 1, respectively, and n intermediate risk levels are assigned values ​​at equal intervals. Optionally, a non-equal-interval mapping method based on data is used to determine the quantified values ​​of historical risk levels. Specifically, while determining the risk level, the cloud platform calculates the average density of all clusters within a certain risk level based on the results of geographical clustering, and uses this normalized value as its quantified value.

[0055] Optionally, a specific embodiment of the cloud platform determining the quantitative value of historical risk level through non-equal interval mapping method includes: historical risk levels include level1, level2, and level3; level1 cluster: average density of 50 clusters / km² / day; level2 cluster: average density of 200 clusters / km² / day; level3 cluster: average density of 800 clusters / km² / day. The density is calculated using min-max normalization so that all values ​​fall between those mapped to the lowest and highest levels. For example, if the lowest level is mapped to 0 and the highest level to 0.6, the quantization value for level 1 is (50-50) / (800-50)*0.6=0, while the quantization value for level 2 is (200-50) / (800-50)*0.6=0.12, and so on for other levels.

[0056] In step Sa2, the sensitivity coefficient K is a coefficient between 0 and 1. K=0 means no adjustment, and K=1 means the threshold is reduced to 0 (that is, any tiny risk will trigger the highest alarm, which does not exist in real-world scenarios).

[0057] Optionally, the vehicle-side determines the initial value of the sensitivity coefficient based on expert experience and a conservative safety principle. For example, if the historical risk level is level 1, the k value is set to 0.2, and the decision threshold is reduced to 80% of the baseline decision threshold; if the historical risk level is level 2, the k value is set to 0.4, and the decision threshold is reduced to 60% of the baseline decision threshold. Optionally, the sensitivity coefficient is determined based on historical accident data. First, the geographical location data of historical accident records is obtained; then, the accident points are matched with the risk clusters discovered by the system to confirm which clusters are most strongly correlated with accidents. By reverse derivation and analysis of the feature value distribution of driving behavior data (sudden braking, sharp turns, etc.) before the accident at these confirmed high-risk points, the aim is to find a threshold T such that if the system threshold is set to T, an effective warning can be issued in advance when an accident occurs; based on the ideal threshold T and the baseline threshold... By working backward, we can deduce the required K value, that is... .

[0058] According to such Figure 4As shown in the implementation, the solution proposed in this invention uses the risk road spectrum information obtained by the cloud platform through training with a large amount of data to determine the risk level of the road currently being driven. This enables the solution to not only perceive the current physical environment, but also the historical and statistical risks, thus achieving the fusion of spatiotemporal dimensions.

[0059] Figure 5 This is a flowchart illustrating an embodiment 2000 of the present invention, which describes a cloud-edge collaborative commercial vehicle safety monitoring method applied to a cloud platform. Figure 5 As shown, Example 2000 includes steps SB1-SB3.

[0060] In some specific embodiments, in step SB1, the cloud platform receives an environmental information request from the vehicle, which includes vehicle location data. Optionally, the communication API endpoint in the cloud platform receives the environmental information request via MQTT or HTTP protocol, verifies the validity of the request, parses it, and extracts the key data included in the request data packet, namely the vehicle's location.

[0061] In step SB2, based on vehicle positioning data, the cloud platform obtains the corresponding environmental context information from a third-party information service platform. The environmental context information includes at least real-time weather information and road topology information.

[0062] Optionally, in step SB2, the cloud platform obtains corresponding information from third-party map services and third-party weather services respectively. During this process, separate requests need to be constructed, and file parsing needs to be performed according to the specific configuration of the third-party information service platform. Optionally, after obtaining the raw data corresponding to the real-time weather information and road topology information, the cloud platform converts and aggregates them into environmental context information and generates a JSON data packet.

[0063] In step SB3, the cloud platform sends the environmental context information to the vehicle. Optionally, the cloud platform sends the aforementioned JSON data packet back to the requesting vehicle via the original MQTT or HTTP response.

[0064] According to such Figure 5 As shown in the embodiment, the solution proposed in this invention enables the vehicle to obtain environmental context information corresponding to the current driving situation through the bridging role of the cloud platform, providing key support for the vehicle-side safety monitoring combined with context awareness in this invention.

[0065] Figure 6 for Figure 5 A flowchart illustrating step SB2 in embodiment 2000. (See attached diagram.) Figure 6 As shown, step SB2 includes steps SB21-SB22.

[0066] In step SB21, the cloud platform invokes a third-party map service through a predefined application programming interface (API) to obtain road topology information. In some specific embodiments, the cloud platform obtains the current road conditions of the vehicle by executing step SB21.

[0067] In step SB21, the specific interaction process between the cloud platform and the third-party map service includes: the cloud platform constructs a request using the reverse geocoding API, receives the response returned by the third-party map service, and then extracts road topology information from the response. Optionally, the third-party map service includes platforms with road information such as Amap and Baidu Maps.

[0068] In step SB22, the cloud platform calls a third-party meteorological service to obtain real-time weather information. Optionally, in step SB22, the cloud platform uses a real-time weather API to construct a request, and after receiving a response, the cloud platform standardizes the weather level defined in the present invention by parsing the corresponding fields.

[0069] Figure 7 This is a flowchart illustrating an embodiment 3000 of the present invention, which describes a cloud-edge collaborative commercial vehicle safety monitoring method applied to a cloud platform. Figure 7 As shown, Embodiment 3000 includes steps S301-S306. Specifically, steps S301-S303 and... Figure 5 Steps SB1-SB3 in Embodiment 2000 are the same and will not be repeated here.

[0070] In some specific embodiments, in step S304, the cloud platform receives driving behavior feature values ​​and their corresponding vehicle positioning data periodically reported by multiple vehicle terminals, forming a historical dataset. Optionally, in some specific embodiments, the vehicle terminal will package and report its own driving behavior feature values ​​and precise positioning data at regular intervals. The reporting is non-real-time and low-priority to avoid consuming the communication bandwidth of emergency alarms.

[0071] Optionally, in step S304, the cloud platform receives massive data packets from hundreds of thousands or even millions of vehicles and stores them as raw data in a data lake or data warehouse to form a continuously growing historical dataset.

[0072] In some specific embodiments, step S305 involves the cloud platform performing geospatial analysis using an unsupervised clustering algorithm based on historical datasets to generate risk road spectrum information. Optionally, the process of the cloud platform completing step S305 includes: preprocessing data using a big data processing engine, specifically including data cleaning, data transformation, and feature selection; then performing unsupervised clustering analysis using machine learning, the purpose of which is to allow the algorithm to automatically discover geographical areas with dense abnormal events from historical data without pre-labeling; finally, through the cloud platform application, calculating contours to create geofences for each cluster result, and classifying risk levels and risk types based on the density of points in the cluster and the average intensity of events, storing the information in the risk road spectrum database.

[0073] Optionally, the unsupervised clustering analysis algorithm used in step S305 is the DBSCAN algorithm (Density-Based Spatial Clustering of Applications with Noise). This algorithm does not require pre-specifying the number of clusters, is suitable for exploratory analysis, and can discover clusters of arbitrary shapes, making it suitable for discovering irregular and dangerous road sections.

[0074] In some specific embodiments, in step S306, the cloud platform sends the risk roadmap information to the vehicle terminals in the relevant areas.

[0075] Optionally, in step S306, the triggering conditions for the cloud platform to issue the action include: when a vehicle is about to enter a certain range around a risk area, the cloud platform will determine whether the vehicle may pass through the risk point based on the vehicle's driving direction and historical route. For example, when the vehicle is 5 kilometers away from the risk point, the cloud platform issues the risk road map information. Optionally, in step S306, the cloud platform will only issue the risk road map information of areas that the vehicle's future journey may cover.

[0076] According to such Figure 7 The implementation method shown in this invention extracts seemingly meaningless driving behavior data of individual vehicles into a predictive risk road spectrum that represents the driving experience of the group through big data analysis and machine learning in the cloud. This spectrum reflects the safety risks of the road itself and is then fed back to all vehicles. This allows even vehicles driving on the road for the first time to be able to anticipate the risks like experienced drivers and be more vigilant in advance, thereby achieving true proactive prevention and forming a powerful safety closed loop.

[0077] Figure 8 This is a flowchart illustrating an embodiment 4000 of the present invention, which describes a cloud-edge collaborative commercial vehicle safety monitoring method applied to a cloud platform. Figure 8As shown, Embodiment 4000 includes steps S401-S405. Specifically, steps S401-S403 and... Figure 5 Steps SB1-SB3 in Embodiment 2000 are the same and will not be repeated here.

[0078] In step S404, the cloud platform receives the risk level sent by the vehicle. In step S405, if the risk level is a high-risk warning, the cloud platform remotely intervenes in the vehicle through a real-time channel.

[0079] In some specific embodiments, in step S405, after receiving the risk level, the cloud platform issues a real-time alarm. Specifically, for Level 2 warnings and Level 3 high-risk events, the cloud platform displays the corresponding information in real time at the monitoring center and issues an alarm sound at the monitoring center, while simultaneously pushing the information to the mobile device of the security management personnel.

[0080] Optionally, in step S405, after receiving the alarm information, the safety manager sends real-time voice messages to the vehicle via the cloud platform, triggering real-time intercom between the vehicle's microphone and speaker to issue warnings or instructions to the driver. If there is no voice response, the system automatically dials the driver's number and coordinates shift changes or rescue operations. Optionally, in step S405, the cloud platform can also save alarm information, operation records, and other data for subsequent accident liability determination, driver training, and algorithm model optimization.

[0081] Figure 9 This is a schematic diagram of an embodiment 5000 of a cloud-edge collaborative commercial vehicle safety monitoring system according to the present invention. Figure 9 As shown, embodiment 5000 includes vehicle terminal 501, cloud platform 502, and third-party information service platform 503.

[0082] In some specific embodiments, the vehicle-side 501 includes a data acquisition module 5011, an edge computing module 5012, a vehicle-side communication module 5013, and a decision-making and early warning module 5014.

[0083] In some specific embodiments, the data acquisition module 5011 is used to acquire multi-dimensional perception data of the vehicle.

[0084] The multi-dimensional sensing data includes in-cabin video data, vehicle CAN bus data, and vehicle positioning data. Optionally, the data acquisition module 5011 includes a camera deployed inside the cockpit, positioned above the center of the windshield or on the top of the dashboard, using a wide-angle or fisheye lens, with a field of view covering the driver's face and steering wheel area to capture in-cabin video streams. Optionally, in practical applications, the camera has infrared night vision capabilities, supporting in-cabin video acquisition at night. Optionally, vehicle CAN bus data and vehicle positioning data are read via direct access through a gateway. Optionally, the data acquisition module 5011 obtains CAN bus data through CAN protocol parsing, specifically including vehicle speed, throttle, brake, steering angle, and turn signal status. Optionally, vehicle positioning data includes the vehicle's current latitude and longitude data, obtained through a built-in GPS / BeiDou positioning module.

[0085] In some specific embodiments, the edge computing module 5012 is used to calculate driver state feature values ​​and driving behavior feature values ​​based on multi-dimensional perception data.

[0086] Optionally, the edge computing module 5012 performs driver status analysis based on in-cabin video data to obtain driver status feature values, specifically including: performing face detection, tracking, and key point detection, and performing driver status analysis based on the detection results.

[0087] Optionally, the edge computing module 5012 performs driving behavior analysis based on vehicle CAN bus data to obtain driving behavior feature values, which include behavior labels and behavior intensity values. Optionally, the behavior labels include rapid acceleration, rapid deceleration, sudden braking, and sharp turning, and the behavior intensity value includes the rate of change of acceleration. Rapid acceleration and rapid deceleration are calculated by the longitudinal rate of change of acceleration and are triggered when they exceed a set first threshold. Sharp turning is calculated by the lateral acceleration or directly using the steering wheel angular velocity and is triggered when it exceeds a set second threshold.

[0088] In some specific embodiments, the vehicle-side communication module 5013 is used to communicate with the cloud platform 502, specifically including a communication module that supports HTTP / HTTPS RESTful API calls or the MQTT protocol. Optionally, the vehicle-side communication module 5013 is used to send an environmental information request containing vehicle positioning data to the cloud platform 502 and receive environmental context information sent by the cloud platform 502.

[0089] In some specific embodiments, the decision warning module 5014 is used to make weighted decisions based on driver state feature values, driving behavior feature values ​​and environmental context information obtained from the cloud platform 502, generate risk levels and execute warnings.

[0090] In some specific embodiments, the decision warning module 5014 determines the risk level through the following process: dynamically adjusting the first and second weighting factors used in the weighted decision based on environmental context information; weighting and summing the driver's state characteristic value and driving behavior characteristic value according to the first and second weighting factors to obtain a comprehensive risk score; and performing rule reasoning based on the comprehensive risk score according to fusion rules to determine the risk level. Optionally, the risk levels determined by the decision warning module 5014 include level 1 alert, level 2 warning, and level 3 high risk.

[0091] In some specific embodiments, the cloud platform 502 includes a platform communication module 5021 and an environment service module 5022. The platform communication module 5021 is used to communicate with the vehicle. The environment service module 5022 is used to obtain environmental context information from a third-party information service platform 503 based on the vehicle's location and provide it to the vehicle. Optionally, the environmental context information includes at least real-time weather information and road topology information. The cloud platform 502 obtains the corresponding information from the road service platform 5031 and the weather service platform 5032, respectively. During the acquisition process, requests need to be constructed separately, and file parsing needs to be performed according to the specific configuration of the third-party information service platform 503.

[0092] In some specific embodiments, the third-party information service platform 503 includes a road service platform 5031 and a weather service platform 5032. The road service platform 5031 provides road topology information to the cloud platform 502; the weather service platform 5032 provides real-time weather information to the cloud platform 502. Optionally, the road service platform 5031 may include platforms with road information such as Gaode Maps and Baidu Maps, and the weather service platform 5032 may include platforms providing real-time weather information such as Hefeng Weather.

[0093] The embodiments of the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, any changes or modifications made by those skilled in the art based on the ideas of the present invention, its specific implementation methods, and its application scope, are all within the scope of protection of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A cloud-edge collaboration-based commercial vehicle safety monitoring method applied to a vehicle end, characterized in that, The method comprises: SA1: obtaining multi-dimensional perception data of the vehicle, the multi-dimensional perception data comprising cabin video data, vehicle CAN bus data and vehicle positioning data; SA2: performing driver state analysis based on the cabin video data to obtain a driver state feature value; SA3: performing driving behavior analysis based on the vehicle CAN bus data to obtain a driving behavior feature value; SA4: sending an environment information request to a cloud platform, the environment information request comprising the vehicle positioning data, and receiving environment context information returned by the cloud platform, the environment context information at least comprising real-time weather information and road topology information; SA5: based on a preset fusion rule, performing weighted decision according to the driver state feature value, the driving behavior feature value and the environment context information to determine a risk level to which the vehicle currently belongs; SA6: performing a corresponding hierarchical warning operation according to the risk level, and sending the risk level to the cloud platform.

2. The method of claim 1, wherein, The step SA5 comprises: dynamically adjusting a first weight factor and a second weight factor used in the weighted decision according to the environment context information; performing weighted summation on the driver state feature value and the driving behavior feature value according to the first weight factor and the second weight factor to obtain a comprehensive risk score; based on the comprehensive risk score, performing rule inference according to the fusion rule to determine the risk level.

3. The method of claim 2, wherein, The process of dynamically adjusting the first weight factor and the second weight factor used in the weighted decision comprises: obtaining a road adhesion coefficient according to the real-time weather information and the road topology information; dynamically adjusting according to the road adhesion coefficient, following the principle that the lower the road adhesion coefficient, the greater the value of the first weight factor and the smaller the value of the second weight factor.

4. The method of claim 2, wherein, The fusion rule comprises a preset decision threshold, and the process of performing rule inference based on the comprehensive risk score comprises: comparing the comprehensive risk score with the decision threshold to determine the current risk level.

5. The method of claim 4, wherein, The environment context information further comprises risk road spectrum information, which represents the road risk situation of a specific geographic area, and the step SA5 further comprises: determining a road risk level corresponding to the vehicle positioning data according to the risk road spectrum information; dynamically adjusting the decision threshold used for rule inference according to the road risk level, based on the principle that the higher the road risk level, the lower the decision threshold required to trigger the same risk level.

6. A cloud-edge collaboration based commercial vehicle safety monitoring method, characterized in that, Applied to a cloud platform, the method comprises: SB1: receiving an environment information request from a vehicle end, the environment information request comprising vehicle positioning data; SB2: based on the vehicle positioning data, obtaining corresponding environment context information from a third-party information service platform, the environment context information at least comprising real-time weather information and road topology information; SB3: sending the environment context information to the vehicle end.

7. The method of claim 6, wherein, The method further comprises: Receiving driving behavior characteristic values and corresponding vehicle positioning data periodically reported by a plurality of vehicle terminals, and forming a historical data set; Based on the historical data set, performing geographical spatial analysis through an unsupervised clustering algorithm to generate risk road spectrum information; Distributing the risk road spectrum information to vehicle terminals in a relevant area.

8. The method of claim 6, wherein, The method further comprises: Receiving a risk level sent by the vehicle terminal; In the case of high-risk warning, remotely intervening in the vehicle through a real-time channel.

9. The method of claim 6, wherein, The corresponding environmental context information obtained from the third-party information service platform comprises: Calling a third-party map service through a predetermined application program interface to obtain road topology information, and calling a third-party weather service to obtain real-time weather information.

10. A cloud-edge collaboration based commercial vehicle safety monitoring system, characterized in that, Comprise: A vehicle terminal, a cloud platform, and a third-party information service platform; The vehicle terminal comprises: A data acquisition module for acquiring multi-dimensional perception data of the vehicle; An edge computing module for calculating driver state characteristic values and driving behavior characteristic values based on the multi-dimensional perception data; A vehicle terminal communication module for communicating with the cloud platform; A decision warning module for performing weighted decision based on the driver state characteristic values, the driving behavior characteristic values, and the environmental context information obtained from the cloud platform, generating a risk level, and performing a warning; The cloud platform comprises: A platform communication module for communicating with the vehicle terminal and the third-party information service platform; An environmental service module for obtaining environmental context information from the third-party information service platform according to the location of the vehicle terminal, and providing it to the vehicle terminal; The third-party information service platform is configured to provide road topology information and real-time weather information to the cloud platform.

Images