Road adhesion coefficient determination method and system

Abstract

The invention relates to a road adhesion coefficient determination method and system, and the method is applied to a vehicle, and comprises the steps: collecting a first road image and the motion state data of the vehicle; determining a first road adhesion coefficient according to the motion state data and the first road image; according to the first road adhesion coefficient, a second road adhesion coefficient sent by a roadside unit and a third road adhesion coefficient issued by a cloud platform, a target road adhesion coefficient is determined, the second road adhesion coefficient is determined by the roadside unit according to motion state data uploaded by a vehicle in a detection range of the roadside unit, and the third road adhesion coefficient is determined by the roadside unit according to motion state data uploaded by a vehicle in a detection range of the roadside unit. The third road adhesion coefficient is determined by the cloud platform according to a second road adhesion coefficient and meteorological data uploaded by the road side unit.

Description

Technical Field

[0001] This disclosure relates to the field of vehicle control technology, and more specifically, to a method and system for determining the road surface adhesion coefficient. Background Technology

[0002] The road surface adhesion coefficient is a core parameter for vehicle driving safety, and its accurate estimation directly determines braking and steering performance.

[0003] In existing technical solutions, the road surface adhesion coefficient is estimated using only one of the following: vehicle, roadside unit, or cloud platform. Vehicle-based estimation is easily limited by its own motion and has a limited sensing range. Roadside units have a fixed coverage area and poor adaptability to cross-regional road surface changes. Cloud platform predictions are susceptible to data transmission delays and struggle to respond quickly to sudden changes in local road conditions. This results in low accuracy and insufficient robustness in the estimation of the road surface adhesion coefficient, making it difficult to ensure the safety of vehicle decision-making and control. Summary of the Invention

[0004] The purpose of this disclosure is to provide a method and system for determining the road surface adhesion coefficient, so as to improve the stability and safety of vehicles.

[0005] To achieve the above objectives, in a first aspect, this disclosure provides a method for determining the road surface adhesion coefficient, applied to vehicles, the method comprising: Acquire first road surface images and vehicle motion state data; The first road surface adhesion coefficient is determined based on the motion state data and the first road surface image; The target road surface adhesion coefficient is determined based on the first road surface adhesion coefficient, the second road surface adhesion coefficient sent by the roadside unit, and the third road surface adhesion coefficient sent by the cloud platform. The second road surface adhesion coefficient is determined by the roadside unit based on the motion state data uploaded by vehicles within its own detection range, and the third road surface adhesion coefficient is determined by the cloud platform based on the second road surface adhesion coefficient and meteorological data uploaded by the roadside unit.

[0006] Optionally, determining the first road surface adhesion coefficient based on the motion state data and the first road surface image includes: Based on the first road surface image and the motion state data, a first visual estimate of the road surface adhesion coefficient is determined; Based on the motion state data, a first dynamic estimate of the road surface adhesion coefficient is determined; The first visual estimate and the first dynamic estimate are weighted and fused to obtain the first road surface adhesion coefficient.

[0007] Optionally, the motion state data includes the current vehicle speed and front wheel steering angle, and determining the first visual estimate of the road surface adhesion coefficient based on the first road surface image and the motion state data includes: Based on the current vehicle speed and the front wheel steering angle, a region of interest is determined in the first road surface image, wherein the region of interest is the vehicle's driving area; The road surface type label and confidence score of each pixel in the region of interest are obtained through the pre-trained classification model. The confidence score of the pixel is used to characterize the proportion of pixels with the same road surface type label in the region of interest. By querying a preset mapping table, the target road surface type label is determined to correspond to the target road surface type label. The target road surface type label is the road surface type label with the highest confidence in the region of interest at the current acquisition time. The preset mapping table includes the correspondence between different road surface adhesion coefficients and road surface type labels. If the confidence level of the target road surface type label is greater than or equal to the first threshold, then the target road surface adhesion coefficient is determined as the first visual estimate at the current acquisition time; or If the confidence level of the target road surface type label is less than the first threshold and greater than the second threshold, then the target road surface adhesion coefficient at the current acquisition time is weighted and fused with the first visual estimate at the previous acquisition time to obtain the first visual estimate at the current acquisition time; or If the confidence level of the target road surface type label is less than or equal to the second threshold, then the first visual estimate or preset safety value at the previous acquisition time is determined as the first visual estimate at the current acquisition time.

[0008] Optionally, the motion state data includes initial longitudinal acceleration, wheel speed, and net torque at multiple acquisition times, and determining the first dynamic estimate of the road adhesion coefficient based on the motion state data includes: If the initial acceleration at the current acquisition moment exceeds the third threshold, or the difference between the net torque at the current acquisition moment and the net torque at the previous acquisition moment exceeds the fourth threshold, the reference vehicle speed, initial longitudinal force, and slip ratio at the current acquisition moment are input into a preset tire model to obtain the first dynamic estimate at the current acquisition moment. The reference vehicle speed at the current acquisition moment is obtained by integrating the initial longitudinal accelerations at multiple acquisition moments; the slip ratio at the current acquisition moment is determined based on the wheel speed and reference vehicle speed at the current acquisition moment; and the initial longitudinal force at the current acquisition moment is determined based on the net torque at the current acquisition moment. The preset tire model is used to characterize the correspondence between the vehicle's longitudinal force, slip ratio, and road adhesion coefficient; or If the initial acceleration at the current acquisition time does not exceed the third threshold, or the difference between the net torque at the current acquisition time and the net torque at the previous acquisition time does not exceed the fourth threshold, the first dynamic estimate calculated at the previous acquisition time is determined as the first dynamic estimate at the current acquisition time. The step of inputting the reference vehicle speed, initial longitudinal force, and slip ratio at the current acquisition time into a preset tire model to obtain the first dynamic estimate at the current acquisition time includes: The initial longitudinal force at the current acquisition time is input into the prediction model to obtain the predicted acceleration at the current acquisition time. The prediction model is a model that predicts longitudinal acceleration based on vehicle mass and longitudinal force. Based on the residual between the initial longitudinal acceleration and the predicted longitudinal acceleration at the current acquisition time, adjust the initial longitudinal force at the current acquisition time to determine the target longitudinal force at the current acquisition time. The regression vector is determined based on the slip ratio and target longitudinal force at the current acquisition moment; The Kalman gain at the current acquisition time is determined based on the regression vector, the first dynamic estimate at the previous acquisition time, and the covariance matrix at the previous acquisition time. Based on the Kalman gain and the error at the current acquisition time, a first dynamic estimate for the current acquisition time is obtained, and the covariance matrix at the current acquisition time is updated. The error at the current acquisition time is determined based on the target longitudinal force at the current acquisition time, the regression vector, and the first dynamic estimate from the previous acquisition time.

[0009] Optionally, the step of weightedly fusing the first visual estimate and the first dynamic estimate to obtain the first road surface adhesion coefficient includes: A first confidence level corresponding to the first visual estimate and a second confidence level corresponding to the first dynamic estimate are determined, wherein the first confidence level is determined based on the confidence level of the road surface type label corresponding to the first visual estimate, and the second confidence level is determined based on at least one of the data quality, data convergence degree, and longitudinal excitation degree of the vehicle corresponding to the first dynamic estimate. When both the first confidence level and the second confidence level are within the range of the first confidence level, the first visual estimate and the first dynamic estimate are weighted and fused according to the first confidence level and the second confidence level to determine the first road surface adhesion coefficient; When either the first confidence level or the second confidence level is not within the range of the first confidence level, the estimated value corresponding to the confidence level that is within the range of the first confidence level is determined as the first road surface adhesion coefficient; When neither the first confidence level nor the second confidence level is within the range of the first confidence level, the preset safety value is determined as the first road surface adhesion coefficient.

[0010] Optionally, determining the target road surface adhesion coefficient based on the first road surface adhesion coefficient, the second road surface adhesion coefficient sent by the roadside unit, and the third road surface adhesion coefficient sent by the cloud platform includes: Based on the preset consistency tolerance parameter, the first visual estimate, and the first dynamic estimate, the posterior confidence level is determined, and the confidence levels corresponding to the posterior confidence level, the first visual estimate, and the first dynamic estimate are weighted and averaged to determine the third confidence level corresponding to the first road surface adhesion coefficient. Receive the fourth confidence level corresponding to the second road surface adhesion coefficient sent by the roadside unit and the fifth confidence level corresponding to the third road surface adhesion coefficient sent by the cloud platform; If the third confidence level is greater than or equal to the first preset confidence level, the first road surface adhesion coefficient is determined as the target road surface adhesion coefficient; or If the third confidence level is less than or equal to the second preset confidence level, and both the fourth and fifth confidence levels are greater than or equal to the first preset confidence level, then either the second road surface adhesion coefficient or the third road surface adhesion coefficient shall be used as the target road surface adhesion coefficient; or If the third confidence level, the fourth confidence level, and the fifth confidence level are all less than the first preset confidence level and greater than the second preset confidence level, then the first road surface adhesion coefficient, the second road surface adhesion coefficient, and the third road surface adhesion coefficient are weighted and fused according to the third confidence level, the fourth confidence level, and the fifth confidence level to obtain the target road surface adhesion coefficient; or If the third confidence level, the fourth confidence level, and the fifth confidence level are all less than the second preset confidence level, the preset safety value is determined as the target road surface adhesion coefficient.

[0011] Secondly, this disclosure also provides a method for determining the road surface adhesion coefficient, applied to roadside units, the method comprising: It receives motion status data from multiple vehicles within its detection range and collects second road surface images and meteorological data; The second road surface adhesion coefficient is determined based on the motion state data of the multiple vehicles and the second road surface image; The meteorological data and the second road surface adhesion coefficient are sent to the cloud platform so that the cloud platform can determine the third road surface adhesion coefficient. The second road surface adhesion coefficient is broadcast outward so that vehicles that receive the second road surface adhesion coefficient can determine the target road surface adhesion coefficient based on their own determined first road surface adhesion coefficient, the second road surface adhesion coefficient, and the third road surface adhesion coefficient issued by the cloud platform.

[0012] Optionally, determining the second road surface adhesion coefficient based on the motion state data of the plurality of vehicles and the second road surface image includes: Based on the second road surface image, a second visual estimate of the road surface adhesion coefficient is determined; Based on the motion state data of the multiple vehicles, a second dynamic estimate of the road surface adhesion coefficient is determined; The second visual estimate and the second dynamic estimate are weighted and fused to obtain the second road surface adhesion coefficient; The step of determining the second visual estimate of the road surface adhesion coefficient based on the second road surface image includes: The road type label for each pixel in the second road image is obtained by using a pre-trained classification model. Based on the preset map link information, select the drivable area of ​​interest in the second road surface image; The road surface type label of the pixel with the highest confidence in the region of interest is determined as the target road surface type label. The confidence of the pixel is used to characterize the proportion of pixels with the same road surface type label in the region of interest. By querying a preset mapping table, the target road surface adhesion coefficient corresponding to the target road surface type label is determined as the second visual estimate.

[0013] Thirdly, this disclosure also provides a method for determining the road surface adhesion coefficient, applied to a cloud platform, the method comprising: Receive meteorological data and second pavement adhesion coefficient uploaded by at least one roadside unit; Based on the meteorological data and the second road surface adhesion coefficient, the third road surface adhesion coefficient is predicted. The third road surface adhesion coefficient is sent to the vehicle so that the vehicle can determine the target adhesion coefficient based on its own determined first road surface adhesion coefficient, the second road surface adhesion coefficient received from the roadside unit, and the third road surface adhesion coefficient. The second road surface adhesion coefficient is determined by the roadside unit based on the motion state data uploaded by the vehicle within its detection range.

[0014] Fourthly, this disclosure also provides a road surface adhesion coefficient determination system, the system comprising at least one of a vehicle, a roadside unit, and a cloud platform; The vehicle is used to perform the method described in the first aspect of this disclosure; The roadside unit is used to perform the method described in the second aspect of this disclosure; The cloud platform is used in the method described in the third aspect of this disclosure.

[0015] The above method enables collaborative determination of the target road surface adhesion coefficient by vehicle, road, and cloud platforms. The vehicle provides a real-time, local first road surface adhesion coefficient based on its own perception; the roadside unit supplements this with a regional second road surface adhesion coefficient through multi-vehicle collaboration; and the cloud platform outputs a forward-looking predicted third road surface adhesion coefficient by combining global meteorological and historical data. The estimation results from these three platforms mutually validate and complement each other, overcoming the limitations of single-platform estimation. This improves the accuracy, real-time performance, and adaptability to all operating conditions of road surface adhesion coefficient estimation, avoids estimation biases caused by single-platform failure, provides a reliable basis for vehicle safety decision-making and control, and ultimately enhances the driving safety of autonomous and assisted driving systems.

[0016] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a method for determining the road surface adhesion coefficient according to an exemplary embodiment.

[0018] Figure 2 This is a flowchart illustrating a method for determining the road surface adhesion coefficient according to an exemplary embodiment.

[0019] Figure 3 This is a schematic diagram illustrating a first road surface image according to an exemplary embodiment.

[0020] Figure 4 This is a flowchart illustrating a method for determining the road surface adhesion coefficient according to an exemplary embodiment.

[0021] Figure 5 This is a flowchart illustrating a method for determining the road surface adhesion coefficient according to an exemplary embodiment.

[0022] Figure 6 This is a block diagram illustrating a road surface adhesion coefficient determination system according to an exemplary embodiment.

[0023] Figure 7 This is a block diagram illustrating a road surface adhesion coefficient determination system according to an exemplary embodiment. Detailed Implementation

[0024] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0025] In the following description, the words "first" and "second" are used only to distinguish the purpose of the description and should not be interpreted as indicating or implying relative importance or order.

[0026] Figure 1 This is a flowchart illustrating a method for determining the road surface adhesion coefficient according to an exemplary embodiment. This method is applied to vehicles. See also... Figure 1 As shown, the method includes the following steps: In step S101, a first road surface image and motion state data of the vehicle are acquired.

[0027] In one embodiment, the vehicle acquires a first road surface image and vehicle motion state data at a first preset frequency, which can be 100Hz, using both an onboard camera and an onboard sensor. The acquired first road surface image is then preprocessed to avoid environmental factors, such as changes in lighting and camera distortion, or differences in data format affecting recognition accuracy. For example, the original acquired first road surface image undergoes the following processing: First, the first road surface image is corrected using the camera's intrinsic parameters to restore the true geometric shape of the road and avoid image distortion caused by the camera's optical characteristics. Second, the distortion-corrected image is uniformly scaled to a fixed size to fit the neural network model input (such as the classification model described later), and the pixel values ​​in the image are normalized to reduce the computational load of the neural network and improve inference speed.

[0028] In step S102, the first road surface adhesion coefficient is determined based on the motion state data and the first road surface image.

[0029] It is worth noting that the first road surface adhesion coefficient is the road surface adhesion coefficient estimated by the vehicle based on its visual function (corresponding to the first road surface image) and dynamic function (corresponding to motion state data), while the second road surface adhesion coefficient is the first road surface adhesion coefficient and vehicle motion state transmitted by the roadside unit from multiple vehicles within its detection range.

[0030] In step S103, the target road surface adhesion coefficient is determined based on the first road surface adhesion coefficient, the second road surface adhesion coefficient sent by the roadside unit, and the third road surface adhesion coefficient sent by the cloud platform.

[0031] The second road surface adhesion coefficient is determined by the roadside unit based on the motion state data uploaded by the vehicles within its detection range, and the third road surface adhesion coefficient is determined by the cloud platform based on the second road surface adhesion coefficient and meteorological data uploaded by the roadside unit.

[0032] In one embodiment, suppose a vehicle is approaching a curve and receives a second road surface adhesion coefficient broadcast by an RSU (Roadside Unit) 1 kilometer away from the curve. Then, it obtains a dynamic digital map of the road surface adhesion coefficient from the cloud platform, which includes a third road surface adhesion coefficient for that road segment, from its navigation system. The vehicle then executes a vehicle-side fusion estimation algorithm to generate a first road surface adhesion coefficient based on vision and dynamics. The vehicle fuses the road surface adhesion coefficients from the three ends to determine the target road surface adhesion coefficient. Based on this target road surface adhesion coefficient, the vehicle can adjust its speed, for example, reducing the planned speed from 100 km / h to 70 km / h, generating a driving trajectory with a gentler curvature and lower speed. It can also send instructions to the chassis control system in advance to shorten the braking response time.

[0033] The above method enables collaborative determination of the target road surface adhesion coefficient by vehicle, road, and cloud platforms. The vehicle provides a real-time, local first road surface adhesion coefficient based on its own perception; the roadside unit supplements this with a regional second road surface adhesion coefficient through multi-vehicle collaboration; and the cloud platform outputs a forward-looking predicted third road surface adhesion coefficient by combining global meteorological and historical data. The estimation results from these three platforms mutually validate and complement each other, overcoming the limitations of single-platform estimation. This improves the accuracy, real-time performance, and adaptability to all operating conditions of road surface adhesion coefficient estimation, avoids estimation biases caused by single-platform failure, provides a reliable basis for vehicle safety decision-making and control, and ultimately enhances the driving safety of autonomous and assisted driving systems.

[0034] In an optional implementation, step S102 above includes the following sub-steps: The first step is to determine the first visual estimate of the road surface adhesion coefficient based on the first road surface image and motion state data.

[0035] The second step is to determine the first dynamic estimate of the road surface adhesion coefficient based on the motion state data.

[0036] The third step is to perform a weighted fusion of the first visual estimate and the first dynamic estimate to obtain the first road surface adhesion coefficient.

[0037] For example, see Figure 2As shown, the vehicle collects image information and vehicle state information. The image information is the first road surface image, and the vehicle state information is motion state data. After semantic annotation, the image information can be used to train the semantic segmentation model offline. Whenever the vehicle collects this information, it first performs signal preprocessing on this information or data, including distortion correction, standardization, cleaning of abnormal data, data reconstruction, unit conversion, and dynamic filtering. Then, the road adhesion coefficient is estimated based on the preprocessed image information and vehicle state information. On the one hand, based on the image information and vehicle state information, the trained semantic segmentation model is called to complete the selection of drivable areas and the calculation of confidence. Then, the visual estimate of the road adhesion coefficient (first visual estimate) and the confidence are determined by looking up a table. On the other hand, the vehicle slip ratio and longitudinal force are determined based on the vehicle state information (motion state data), thereby constructing a tire model reflecting the longitudinal force, slip ratio, and road adhesion coefficient. Residual correction is performed based on the actual collected information and the model prediction information. After the enable condition is judged, RLS parameter recognition is triggered to obtain the dynamic road adhesion coefficient (first dynamic estimate) and the confidence.

[0038] Finally, the visual and dynamic estimates are weighted and fused to obtain the vehicle-side adhesion coefficient, i.e., the first road surface adhesion coefficient. This value can be further combined with the second adhesion coefficient obtained by multi-vehicle collaboration on the roadside and the third adhesion coefficient obtained by combining meteorological data in the cloud, so as to achieve complementary advantages of the three ends and effectively improve the accuracy and robustness of adhesion coefficient estimation under different working conditions.

[0039] Optionally, the vehicle's motion state data includes the current vehicle speed and front wheel steering angle. In the sub-step of step S102 above, the first step can determine the first visual estimate of the road adhesion coefficient through the following implementation: First, based on the current vehicle speed and the front wheel steering angle, a region of interest is determined in the first road surface image.

[0040] The first road surface image includes driving areas and non-driving areas (e.g., driving areas and non-driving areas). Figure 3 The non-driving area can be any unrelated area in the first road surface image, such as the sky, oncoming vehicles, or road structures. The ROI (Region of Interest) is the driving area of ​​the vehicle, that is, the road surface that the vehicle is about to come into contact with. There may be one or more ROIs, and there is no limitation on this.

[0041] For example, the vehicle's trajectory is determined by the current vehicle speed and the front wheel steering angle. The vehicle speed affects the longitudinal distance of the ROI, that is, the faster the vehicle speed, the farther the road surface needs to be identified in advance. The front wheel steering angle determines the lateral offset of the ROI, that is, the larger the steering angle, the more the ROI shifts towards the turning side. By using the dynamic model and camera imaging principle, the vehicle motion state parameters are mapped to the image coordinate system to determine the ROI.

[0042] See Figure 3 As shown, taking a vehicle moving forward as an example, assuming the vehicle's current speed is v (m / s) and the front wheel steering angle is δ (rad), the distance d between the predicted contact point in front of the left and right wheels and the vehicle can be set to a fixed value, such as d = 3 meters. Based on the transformation relationship between the camera and vehicle coordinate systems, the coordinates (μ) of the left and right ROI centers in the image coordinate system can be obtained. left ,ν left ) and (μ right ,ν right For example, the center of the ROI on the left can be determined by the following formula 1: Formula 1: = , = +

[0043] Among them, f x f y c is the camera's intrinsic parameter. x c y Let w be the coordinates of the camera center point, w be the wheelbase, and h be the camera mounting height. The ROI region on the left can be defined as a region with respect to points (μ). left ,ν left Centered on ), with a size of w A rectangular area of ​​h pixels.

[0044] Then, using the pre-trained classification model, the road surface type label and confidence level of each pixel in the region of interest are obtained.

[0045] In one embodiment, the classification model is a lightweight semantic segmentation neural network model, such as the BiSeNet V2 model. The classification model is trained using road images that include different weather conditions (sunny, rainy, snowy, foggy), lighting conditions (daytime, nighttime, tunnel), road surface types, and pre-labeled road surface type labels. This enables the classification model to map the correspondence between image semantic information such as road surface texture, color, and reflective properties and road surface types. The pre-labeled road surface type labels are shown in the preset mapping table below.

[0046]

[0047] Preset mapping relationship table For example, the first road surface image is input into the classification model to obtain a road surface semantic map, which is represented in matrix form as S(x,y)=L i Where x and y represent the position coordinates of each pixel in the road surface semantic map, and the region of each pixel is the road surface type label L. i For example, S(100, 200) = 0 indicates that the road surface type of this pixel is dry asphalt.

[0048] It is worth noting that since the ROI changes dynamically according to the vehicle's motion state, the ROI of the first road surface image can be determined by first inputting the first road surface image into the classification model to obtain the road surface semantic map, then determining the ROI based on the vehicle's current motion state data, and then determining the road surface type label and confidence level of each pixel in the current road surface semantic map.

[0049] The confidence score of a pixel is used to characterize the proportion of pixels with the same road surface type label in the region of interest. For example, if 80% of the pixels in the ROI correspond to the road surface type label of dry asphalt and 20% of the pixels correspond to the road surface type label of wet asphalt, the confidence score for dry asphalt is 0.8, while the confidence score for wet asphalt is 0.2.

[0050] Finally, the first visual estimate is determined based on the road surface type label and confidence level of each pixel within the region of interest.

[0051] Using the above method, the region of interest is dynamically determined by vehicle speed and front wheel steering angle, accurately focusing on the road surface that the vehicle is about to contact, eliminating interference from non-driving areas, effectively improving the accuracy of visual estimation, and using a pre-trained classification model to obtain pixel-level road surface type labels and confidence levels, which can directly support subsequent multi-source fusion, improve the robustness of the end-to-end adhesion coefficient estimation, and adapt to all vehicle driving conditions.

[0052] Optionally, the vehicle's motion state data includes the current vehicle speed and front wheel steering angle. After determining the road surface type label and confidence level of each pixel within the vehicle's region of interest, a first visual estimate of the road surface adhesion coefficient can be determined using the following implementation method: First, the target road surface adhesion coefficient corresponding to the target road surface type label is determined by querying the preset mapping relationship table.

[0053] The target road surface type label is the road surface type label with the highest confidence level in the region of interest at the current acquisition time. See the preset mapping table shown above, which includes the correspondence between different road surface adhesion coefficients and road surface type labels.

[0054] In one embodiment, the road surface type L with the most pixels in the ROI is selected. maxThe road surface type of the first road surface image is determined, and the confidence level (proportion) of this road surface type (corresponding to the target road surface type label) is determined as the confidence level of the first visual estimate. Furthermore, by querying a preset mapping table, L... max The corresponding road surface adhesion coefficient was determined as the initial visual estimate. For example, if the road surface type with the highest pixel percentage in the ROI is dry asphalt, accounting for 95%, then it is determined that... =0.85, with a confidence level of 0.95.

[0055] Furthermore, the confidence level of the output first visual estimate needs to be within a reasonable range. Therefore, the upper and lower limits of the first visual estimate are set as the first threshold C1 and the second threshold C2, respectively. The first threshold C1 and the second threshold C2 are pre-set confidence thresholds, and C2 < C1.

[0056] Then, based on the confidence level C of the target road surface type label max The relationship between the magnitudes of C1 and C2 determines the first visual estimate. This includes any of the following situations: Case 1: If the confidence level of the target road surface type label is greater than or equal to the first threshold, then the target road surface adhesion coefficient is determined as the first visual estimate at the current acquisition time.

[0057] Case 2: If the confidence level of the target road surface type label is less than the first threshold and greater than the second threshold, the target road surface adhesion coefficient at the current acquisition time is weighted and fused with the first visual estimate at the previous acquisition time to obtain the first visual estimate at the current acquisition time.

[0058] Case 3: If the confidence level of the target road surface type label is less than or equal to the second threshold, then the first visual estimate or the preset safety value at the previous acquisition time is determined as the first visual estimate at the current acquisition time.

[0059] For example, C1=0.7, C2=0.3.

[0060] In C max If C1 or higher, indicating a high confidence level for the target road surface type label, the first visual estimate is directly output. = .

[0061] In C2≤C max If the value is ≤C1, it indicates that the confidence level of the target road surface type label is at a medium confidence level. The initial estimate (target road surface adhesion coefficient) at this acquisition time and the first visual estimate at the previous acquisition time can be compared. We perform weighted fusion to obtain the first visual estimate at the current acquisition time, for example, through the following calculation formula 2: , 0,1). Alternatively, the adhesion coefficients corresponding to each road surface type in the ROI at the current acquisition time can be weighted and fused to obtain the first visual estimate at the current acquisition time, for example, through the following calculation formula 3: ,in, Let be the road adhesion coefficient corresponding to the i-th road surface type. Let n be the confidence level, and n be the number of road surface types in the ROI.

[0062] In C max If the confidence level of the target road surface type label is ≤C2, it indicates that the confidence level is low, so the first visual estimate from the previous time step is output. If N consecutive frames show low confidence or the sensor malfunctions, a preset safety value will be output. For example, a preset safety value of 0.3 can be set, and a visual perception failure or downgrade can be reported to the vehicle.

[0063] The above scheme improves the accuracy and stability of vehicle-side visual function-based estimation of road adhesion coefficient, providing reliable data support for vehicle safety.

[0064] Optionally, the vehicle's motion state data includes initial longitudinal acceleration, wheel speed, and net torque at multiple acquisition times. In the sub-step of step S102 above, the second step can determine the first dynamic estimate through the following implementation: Step 1: Determine the target longitudinal force at the current acquisition time based on the net torque at the current acquisition time.

[0065] It is worth noting that when the net torque is positive, the corresponding force of the vehicle is the driving force, which is calculated from the engine torque and transmission system parameters. When the net torque is negative, the corresponding force of the vehicle is the braking force, which is calculated from the braking pressure and brake parameters. In other words, the longitudinal force of the vehicle can be calculated based on the driving force or braking force of the vehicle.

[0066] Step 2: Integrate the initial longitudinal acceleration at multiple acquisition times to determine the reference vehicle speed at the current acquisition time.

[0067] For example, the initial longitudinal acceleration a at multiple acquisition times is calculated using the following formula 4. x,i Integrate the data to determine the reference vehicle speed at the current data collection time: Formula 4:

[0068] Among them, v ref (k) represents the reference vehicle speed at the current data collection time, v ref (0) represents the initial vehicle speed at the initial moment. This represents the data collection time interval.

[0069] Step 3: Determine the slip ratio based on the wheel speed and reference vehicle speed at the current collection time.

[0070] For example, the slip ratio s is determined by the following formula 5: Formula 5:

[0071] in, This refers to the wheel linear velocity, which is obtained by converting wheel speed and wheel parameters.

[0072] It is worth noting that the dynamic estimation of the road adhesion coefficient is based on the interaction between the tire and the road surface. Therefore, this estimation process depends on the longitudinal excitation generated by the vehicle's driving or braking. When there is no excitation, the slip ratio is close to 0, and the adhesion coefficient cannot be derived. Therefore, the following step 4 needs to be performed to determine the longitudinal excitation of the vehicle, thereby determining the first dynamic estimate of the vehicle.

[0073] Step 4: Determine the first dynamic estimate for the current acquisition time based on the longitudinal excitation of the vehicle at the current acquisition time.

[0074] The longitudinal excitation of the vehicle is determined based on the initial acceleration and net torque, including at least one of the following: Case 1: If the initial acceleration at the current acquisition time exceeds the third threshold or the difference between the net torque at the current acquisition time and the net torque at the previous acquisition time exceeds the fourth threshold, the reference vehicle speed and slip ratio at the current acquisition time are input into the preset tire model to obtain the first dynamic estimate at the current acquisition time.

[0075] The third threshold is a pre-set acceleration threshold, and the fourth threshold is a pre-set net torque threshold. The preset tire model is used to characterize the relationship between the vehicle's longitudinal force, slip ratio, and road adhesion coefficient. For example, the third threshold can be set to 0.2g, and the fourth threshold can be set to 5g. (Cow rice).

[0076] Case 2: If the initial acceleration at the current acquisition time does not exceed the third threshold or the difference between the net torque at the current acquisition time and the net torque at the previous acquisition time does not exceed the fourth threshold, the first dynamic estimate calculated at the previous acquisition time is determined as the first dynamic estimate at the current acquisition time.

[0077] By setting threshold ranges for initial acceleration and net torque, at any given acquisition time, if at least one of the initial acceleration or net torque at the current acquisition time meets the threshold range (i.e., is greater than the corresponding threshold), it indicates that the longitudinal excitation of the vehicle is sufficient, and the interaction between the tire and the road surface is significant. The road adhesion coefficient can be estimated in real time through the tire model. If neither the initial acceleration nor the net torque at the current acquisition time meets the threshold range (i.e., is less than or equal to the corresponding threshold), it indicates that the longitudinal excitation of the vehicle is insufficient. In this case, the first dynamic estimate determined at the previous acquisition time is used as the estimate at the current time to avoid the divergence of the estimate when there is no excitation and to ensure output stability.

[0078] In one implementation, the preset tire model is based on a simplified magic formula. The core characteristic of this model is that the slip ratio determines the longitudinal force, and the upper limit of the longitudinal force is determined by the road adhesion coefficient. The relationship between the slip ratio s, the longitudinal force, and the road adhesion coefficient is shown in the following calculation formula 6: Formula 6:

[0079] Where s is the slip ratio. F is the road surface adhesion coefficient. x For longitudinal force, F z F represents the vertical load on the wheel, and B, C, and D are tire characteristic parameters. z Determined by vehicle parameters.

[0080] In scenario one above, after inputting the reference vehicle speed and slip ratio at the current acquisition time into the preset tire model, the error between the model's predicted dynamic estimate and the measured longitudinal force can be continuously minimized based on RLS (Recursive Least Squares). The road adhesion coefficient is then iteratively updated to obtain the first dynamic estimate at the current acquisition time output by the preset tire model. This process includes the following steps: First, the regression vector is determined based on the slip ratio and target longitudinal force at the current acquisition time.

[0081] For example, based on a preset tire model, the regression vector H is calculated as shown in Equation 7 below: Formula 7:

[0082] For each acquisition moment, when new s and F are received... x Then, the regression vector H is calculated according to the above calculation formulas 6 and 7. The regression vector is used to characterize the theoretical longitudinal force corresponding to the current slip ratio under the unit adhesion coefficient, thereby converting the nonlinear relationship of the preset tire model into a linear expression.

[0083] Then, based on the regression vector, the first dynamic estimate at the previous acquisition time, and the covariance matrix at the previous acquisition time, the Kalman gain at the current acquisition time is determined.

[0084] Finally, based on the Kalman gain and the error at the current acquisition time, the first dynamic estimate at the current acquisition time is obtained, and the covariance matrix at the current acquisition time is updated. The error at the current acquisition time is determined based on the target longitudinal force at the current acquisition time, the regression vector, and the first dynamic estimate at the previous acquisition time.

[0085] For example, the Kalman gain K at the current time (time k) is calculated based on the following formula 8. k : Formula 8: Where R is the observation noise, P k-1 It is the covariance matrix of the previous time step. If the covariance of the previous time step is large (high estimation uncertainty), then K k If the current observation data has a large correction weight, then K will be higher; if the observation noise R is large (the longitudinal force error of the target is large), then K will be higher. k Smaller values ​​rely more heavily on the previous time-to-time estimate.

[0086] Then, by calculating the error between the target longitudinal force at the current acquisition moment and the longitudinal force predicted based on historical estimates, and combining this with Kalman gain correction of historical estimates, the optimal adhesion coefficient at the current moment is obtained. Simultaneously, the covariance matrix is ​​updated to quantify the uncertainty of the current estimate, providing a basis for calculation at the next moment. Here, the error... (Equation 9), update the adhesion coefficient based on the following equation 10. : (Calculation formula 10), and update the covariance matrix based on the following calculation formula 11: (Calculation formula 11), where Q is the process noise.

[0087] By continuously updating the fit and suppressing measurement noise, the first dynamic estimate at the previous acquisition time is finally determined based on the sum of the product of the Kalman gain and the error and the dynamic estimate from the previous frame. .

[0088] The above methods improve the smoothness of parameter convergence and reduce the noise impact in the process of predicting the road surface adhesion coefficient, thereby improving vehicle stability and safety.

[0089] Alternatively, in step 1 above, the target longitudinal force at the current acquisition moment can be determined using the following implementation method: First, determine the initial longitudinal force at the current acquisition time based on the net torque at the current acquisition time.

[0090] Then, the initial longitudinal force at the current acquisition time is input into the prediction model to obtain the predicted longitudinal acceleration. The prediction model is a model for predicting longitudinal acceleration based on vehicle mass.

[0091] Finally, based on the residual between the initial longitudinal acceleration and the predicted longitudinal acceleration, the initial longitudinal force at the current acquisition moment is adjusted to determine the target longitudinal force at the current acquisition moment.

[0092] For example, the vehicle obtains the net torque at the current acquisition time via the CAN bus. Then, based on the transmission system parameters, wheel parameters, and initial longitudinal force in the vehicle parameters, the initial longitudinal force at the current acquisition time is determined. Based on the prediction model, the theoretically predicted longitudinal acceleration is inferred from the initial longitudinal force. The difference between the predicted longitudinal acceleration and the acquired initial acceleration is determined as the residual. This residual reflects the unmodeled dynamic effects such as load transfer, air resistance, and slope resistance.

[0093] Finally, the target longitudinal force F is determined based on the following calculation formula 12. target : Formula 12: , of which F initial Let m be the initial longitudinal force, and m be the vehicle mass. The residual is the acceleration.

[0094] The above scheme improves the accuracy of vehicle-based dynamic estimation of road adhesion coefficient.

[0095] In one implementation, the third step described above can be based on the following implementation: weighted fusion of the first visual estimate and the first dynamic estimate to obtain the first road surface adhesion coefficient. Step 1: Determine the first confidence level corresponding to the first visual estimate and the second confidence level corresponding to the first dynamic estimate.

[0096] Wherein, the first confidence level C v It is determined based on the confidence level of the road surface type label corresponding to the first visual estimate, and the second confidence level C d It is determined based on at least one of the data quality, data convergence degree, and longitudinal excitation degree of the vehicle corresponding to the first dynamic estimate.

[0097] In one embodiment, the second confidence level is obtained by weighted fusion of scores from data quality, data convergence, and the longitudinal excitation degree of the vehicle, wherein the weights for the fusion are pre-set, for example, C. d =0.4G+0.3N+0.3M, where G is the data quality score, N is the data convergence score, and M is the longitudinal incentive score.

[0098] The data quality score is determined based on data integrity, stability (fluctuation amplitude), and noise level; the longitudinal excitation score is determined based on the rate of change of acceleration and net torque, with the rate of change categorized as strong, weak, or no excitation according to a set range, and different scores corresponding to different ranges of the rate of change; the data convergence score is based on the convergence range of the diagonal key element of the updated covariance matrix, for example, a score of 1 is given if the diagonal key element is less than 0.001, a score of 0 is given if it is greater than 0.01, and a score of 0.5 is given if it is in between; or the score is determined by calculating the fluctuation amplitude between the maximum and minimum estimates of the first dynamic estimate in the last N frames and determining the score based on the range of the amplitude.

[0099] Step 2: Based on the magnitudes of the first and second confidence levels, determine the first road surface adhesion coefficient, specifically including at least one of the following cases: Scenario 1: When both the first confidence level and the second confidence level are within the range of the first confidence level, the first visual estimate and the first dynamic estimate are weighted and fused according to the first confidence level and the second confidence level to determine the first road surface adhesion coefficient.

[0100] Scenario 2: When either the first confidence level or the second confidence level is not within the range of the first confidence level, the estimated value corresponding to the confidence level that is within the range of the first confidence level shall be determined as the first road surface adhesion coefficient.

[0101] Scenario 3: When neither the first confidence level nor the second confidence level is within the range of the first confidence level, the preset safety value is determined as the first road surface adhesion coefficient.

[0102] In some embodiments, functional states corresponding to confidence levels within a first confidence level range are considered normal and degraded, while functional states not corresponding to the first confidence level range are considered failed. The first confidence level range is set to between 0.3 and 0.7. v and C d A value greater than 0.7 indicates normal function; a value between 0.3 and 0.7 indicates degraded function; and a value below 0.3 or a detected sensor malfunction indicates functional failure. The vehicle's visual and dynamic functions are assessed in real-time based on a first confidence level range. If both visual and dynamic functions are normal or degraded, a first road adhesion coefficient is obtained using a confidence-based weighted fusion. The calculation is as follows: Formula 13: ,in, This is the first visual estimate. This is the first dynamic estimate.

[0103] If visual function fails, but dynamic function is normal or degraded, then the output will be... If the dynamic function fails, but the visual function is normal or degraded, then the output will be... If both the vision and dynamics functions fail, the safety fault-tolerance function is triggered, limp mode is activated, a preset safety value is output, and a warning is issued through the vehicle's central control display or instrument panel. At the same time, a function degradation indicator is transmitted to the upper-level control system.

[0104] Optionally, step S103 above can be implemented by fusing the road surface adhesion coefficients estimated by the vehicle, roadside unit, and cloud platform to obtain the target road surface adhesion coefficient: S103.1, Based on the preset consistency tolerance parameter, the first visual estimate and the first dynamic estimate, determine the posterior confidence level, and then perform a weighted average of the confidence levels corresponding to the posterior confidence level, the first visual estimate and the first dynamic estimate to determine the third confidence level corresponding to the first road surface adhesion coefficient.

[0105] The weights for the weighted average of the posterior confidence, the first visual estimate, and the first dynamic estimate can be constant or calibrated based on the magnitude of the visual confidence and the dynamic confidence.

[0106] For example, after calculating the fusion estimate (first road adhesion coefficient) based on vision and dynamics in the vehicle depot, a posterior confidence level C is determined. cst The calculation is performed to verify whether the results from the two sources support each other, as detailed in Equation 14 below: Formula 14: ,in, The preset consistency tolerance parameter is obtained through calibration.

[0107] Then, the confidence scores of the first visual estimate, the first dynamic estimate, and the posterior confidence score are combined, and a weighted geometric mean is used to obtain the final fusion confidence score, as detailed in Equation 15 below: Formula 15: Where w1, w2, and w3 are weighting coefficients, which are calibrated according to the actual situation.

[0108] S103.2, receive the fourth confidence level corresponding to the second road surface adhesion coefficient sent by the roadside unit and the fifth confidence level corresponding to the third road surface adhesion coefficient sent by the cloud platform.

[0109] S103.3, determine the target road surface adhesion coefficient based on the third, fourth, and fifth confidence levels.

[0110] In one optional implementation, the target road surface adhesion coefficient can be determined based on the relationship between the third, fourth, and fifth confidence levels and their respective corresponding confidence thresholds, specifically including the following cases: First: When the third confidence level is greater than or equal to the first preset confidence level, the first road surface adhesion coefficient is determined as the target road surface adhesion coefficient.

[0111] The above situation indicates that the vehicle is in a highly excited state. Since the vehicle's own perception is more accurate, the road adhesion coefficient obtained by vehicle-side fusion is trusted first, that is, the first road adhesion coefficient.

[0112] Second: If the third confidence level is less than or equal to the second preset confidence level, and the fourth and fifth confidence levels are both greater than or equal to the first preset confidence level, then either the second road surface adhesion coefficient or the third road surface adhesion coefficient shall be used as the target road surface adhesion coefficient.

[0113] The above situation illustrates that the vehicle's sensors may fail or the perception conditions may be poor (e.g., bad weather conditions). Since the data of the roadside unit and the cloud platform are consistent, either the road surface adhesion coefficient estimated by the roadside unit (second road surface adhesion coefficient) or the road surface adhesion coefficient estimated by the cloud platform (third road surface adhesion coefficient) can be arbitrarily selected as the target road surface adhesion coefficient. When there is a conflict between the roadside unit and the cloud platform, the second road surface adhesion coefficient corresponding to the roadside unit is preferred as the target road surface adhesion coefficient.

[0114] Third: When the third, fourth, and fifth confidence levels are all less than the first preset confidence level and greater than the second preset confidence level, the first road surface adhesion coefficient, the second road surface adhesion coefficient, and the third road surface adhesion coefficient are weighted and fused according to the third, fourth, and fifth confidence levels to obtain the target road surface adhesion coefficient.

[0115] Fourth: If the third, fourth, and fifth confidence levels are all less than the second preset confidence level, the preset safety value will be determined as the target road surface adhesion coefficient.

[0116] It is worth noting that when all three ends fail or the confidence level is extremely low, a degradation strategy is triggered, and a preset safety value set is output as the target road surface adhesion coefficient.

[0117] This solution integrates the first visual estimate and the first dynamic estimate from the vehicle, calculates the posterior confidence level using preset parameters, and then calculates a third confidence level using a weighted average, thereby enhancing the reliability of the vehicle-side fusion results. Furthermore, it makes scenario-specific decisions based on the relationship between the confidence levels from the vehicle, roadside, and cloud perspectives and thresholds, adapting to multiple operating conditions and ensuring the accuracy of the target adhesion coefficient and robustness across all scenarios.

[0118] This disclosure also provides a method for determining the road surface adhesion coefficient, applied to roadside units, see [link to relevant documentation]. Figure 4 As shown, the method includes the following steps: In step S201, motion status data of multiple vehicles within its detection range are received, and second road surface images and meteorological data are acquired.

[0119] It is worth noting that there may be multiple roadside units, which are set up near the road. There may be multiple vehicles within the detection range of each roadside unit, and each roadside unit is connected to the cloud platform.

[0120] For example, a roadside unit receives motion status data broadcast by connected vehicles within its coverage area via V2X (Vehicle-to-Everything) communication. This motion status data includes at least one of the following: vehicle ID, latitude and longitude coordinates, speed, acceleration, yaw rate, and timestamp. The roadside unit performs data preprocessing operations on the multi-vehicle data, such as spatiotemporal alignment and validity checks. For example, time alignment: based on the timestamps and GNSS (Global Navigation Satellite System) time in the motion status data of each vehicle, all motion status data are aligned to a unified time reference, and outlier data points that clearly exceed the physical limits of vehicle dynamics are removed. Another example is spatial alignment: based on the vehicle coordinate information in the motion status data, the motion status data of vehicles located on the target road segment is selected.

[0121] Secondly, in order to improve the estimation accuracy, MEC prioritizes the selection of vehicle data that are in a state of high longitudinal or lateral excitation as valid input. Specifically, longitudinal acceleration thresholds and lateral acceleration thresholds are set to select vehicles whose absolute values ​​of current longitudinal or lateral acceleration exceed the set thresholds.

[0122] In step S202, the second road surface adhesion coefficient is determined based on the motion state data of the plurality of vehicles and the second road surface image.

[0123] The size range of the road surface image collected by the roadside unit may be larger than the range of the road surface image collected by any vehicle. Therefore, the second road surface image may be an image at the same location as the first road surface image collected by the corresponding vehicle, or it may be a complete image collected by the roadside unit. By arranging the positions of the roadside units in a balanced manner, multiple roadside units can collect road surface images of any road segment, thereby establishing and updating map link information.

[0124] In step S203, the meteorological data and the second road surface adhesion coefficient are sent to the cloud platform so that the cloud platform can determine the third road surface adhesion coefficient.

[0125] The third road surface adhesion coefficient is a road surface adhesion coefficient for a future period of time, predicted by the cloud platform based on the second road surface image uploaded by the roadside unit and meteorological information. It is worth noting that the second road surface adhesion coefficient and the third road surface adhesion coefficient can be applied to any road segment, including the road segment corresponding to the first road surface image. Therefore, this disclosure does not impose specific restrictions on them.

[0126] In step S204, the second road surface adhesion coefficient is broadcast outward so that the vehicle receiving the second road surface adhesion coefficient can determine the target road surface adhesion coefficient based on its own determined first road surface adhesion coefficient, the second road surface adhesion coefficient, and the third road surface adhesion coefficient issued by the cloud platform.

[0127] The above technical solution achieves the determination of the target road surface adhesion coefficient through the collaboration of vehicle, road, and cloud. The vehicle provides a real-time first road surface adhesion coefficient based on its own perception, the roadside unit supplements the regional second road surface adhesion coefficient through multi-vehicle collaboration, and the cloud platform outputs a predicted third road surface adhesion coefficient by combining global meteorological and historical data. The data from the three ends mutually verify and complement each other, effectively overcoming the limitations of single-end estimation, improving the estimation accuracy, real-time performance, and adaptability to all operating conditions of the road surface adhesion coefficient, avoiding estimation bias caused by the failure of a single end, providing a reliable basis for vehicle safety decision-making and control, and enhancing the driving safety of autonomous driving or assisted driving systems.

[0128] In one embodiment, step S203 above includes the following implementation sub-steps: S203.1: Determine a second visual estimate of the road surface adhesion coefficient based on the second road surface image.

[0129] S203.2: Based on the motion state data of the plurality of vehicles, determine a second dynamic estimate of the road surface adhesion coefficient.

[0130] S203.3: The second visual estimate and the second dynamic estimate are weighted and fused to obtain the second road surface adhesion coefficient.

[0131] For example, the roadside unit employs simplified vehicle dynamics models such as the dual-track model and a tire model based on the simplified magic formula to construct a state-space equation with the road surface adhesion coefficient as the core state variable. Using unscented Kalman filtering or extended Kalman filtering algorithms, the longitudinal motion state data of multiple highly excited vehicles are used as observation inputs to jointly constrain and update the estimated value of the road surface adhesion coefficient. This process involves continuous iteration of prediction and updating. In the prediction step, based on the prior value of the adhesion coefficient, the mapping relationship between the adhesion coefficient and tire force is quantified through the tire model, and then the predicted motion state of multiple vehicles is calculated using the dual-track model. In the update step, the motion state data actually collected by multiple vehicles are integrated, the error between the predicted value and the observed value is compared, and the adhesion coefficient state variable is continuously corrected until the convergence threshold is met. Finally, a second dynamic estimate based on dynamics is output. Simultaneously, the roadside unit acquires roadside sensor data, such as a second road surface image captured by a roadside camera, and calls a built-in lightweight image recognition model to identify the road surface state, obtaining a second visual estimate based on vision. This value, through collaborative verification of multi-vehicle data, can avoid the bias of local data from a single vehicle and more accurately reflect the overall adhesion characteristics of the target road segment.

[0132] Finally, MEC fuses the estimation results from the two sources and performs a confidence assessment. The fusion rule is similar to that of vehicle-side vision and dynamics fusion, and will not be elaborated here. After fusion, MEC encapsulates the estimated second road surface adhesion coefficient and confidence level, corresponding road segment location information, and effective time range into a standardized roadside message, and continuously broadcasts it to vehicles within the road segment range at a specific frequency through the roadside unit.

[0133] The above methods improve the accuracy of road surface adhesion coefficient estimation by roadside units, providing a data basis for vehicle stability control.

[0134] Optionally, in step S203.1 above, the roadside unit estimates the second visual estimate based on its visual function, which includes the following implementation: First, the road type label for each pixel in the second road image is obtained through a pre-trained classification model.

[0135] For example, the classification model can be a lightweight classification model, such as BiSeNet V2 and Fast-SCNN. This classification model has been trained offline with massive amounts of road images and has the ability to classify road images pixel by pixel. It can accurately identify the road type corresponding to each pixel and label each pixel with the road type.

[0136] Then, based on the preset map link information, the drivable area of ​​interest is selected in the second road surface image, and the road surface type label of the pixel with the highest confidence in the area of ​​interest is determined as the target road surface type label.

[0137] For example, the roadside unit reads the preset map link information (including the road link ID, number of lanes, and lane boundary coordinates covered by the roadside unit), maps the drivable area coordinates in the preset map link information to the pixel coordinate system of the second road surface image, selects a rectangular or polygonal ROI that includes only the drivable lanes of the target road segment, removes all pixels corresponding to non-driving areas, counts the number of pixels of each road surface type label within the ROI, calculates the pixel proportion of each road surface type as the confidence level, and selects the road surface type label with the highest confidence level as the target road surface type label to ensure that the decision result reflects the overall road surface condition of the road segment.

[0138] Among them, the confidence score of a pixel is used to characterize the proportion of pixels with the same road surface type label in the region of interest.

[0139] Finally, by querying the preset mapping relationship table, the target road surface adhesion coefficient corresponding to the target road surface type label is determined as the second visual estimate.

[0140] By relying on a pre-trained classification model to achieve pixel-level road surface recognition, the dominance of road surface type is quantified by pixel proportion, which can accurately identify the overall road surface condition of the road segment, avoid misjudgment caused by local abnormal areas (such as small-scale water accumulation), and ensure that the estimated value is close to the actual road surface on which the vehicle is driving.

[0141] This disclosure also provides a method for determining the road surface adhesion coefficient, applied to a cloud platform, see [link to relevant documentation]. Figure 5 As shown, the method includes the following steps: In step S301, meteorological data and the second road surface adhesion coefficient uploaded by at least one roadside unit are received.

[0142] In step S302, a third road surface adhesion coefficient is predicted based on the meteorological data and the second road surface adhesion coefficient.

[0143] For example, the cloud platform uses LSTM (Long Short-Term Memory) or Transformer models to periodically perform training tasks. The goal is to train a prediction model. Inputting this model with weather data for a future period, current time, and road conditions, it can output predicted road surface adhesion coefficient values ​​for that period. For instance, the model can learn the pattern that "on a night when the temperature suddenly drops to -2°C and there is light rain, the adhesion coefficient value of a curved section of a highway will rapidly decrease from 0.7 to 0.3 within one hour." The training data consists of historical adhesion coefficient data, weather data, and time data from the past year.

[0144] The cloud platform aggregates multi-dimensional input data, including the second pavement adhesion coefficient uploaded by roadside units (including second visual / dynamic estimates and confidence levels), real-time meteorological data (temperature, precipitation, humidity, etc.) and future weather forecasts (1-24 hours) for the corresponding road segment, historical adhesion coefficient time-series data for the target road segment (in the same meteorological / seasonal scenarios), and road segment attributes (pavement material, drainage design) from high-precision maps. This data is then spatiotemporally aligned, anomaly cleaned, and feature-engineered. Specifically, the data uploaded by roadside units is bound to the road segment ID, the time granularity is standardized, and features such as the temporal change rate of pavement adhesion coefficient and meteorological-derived risk factors are extracted. Finally, by using a pre-trained prediction model (such as LSTM / Transformer), the dynamic correlation between road surface adhesion coefficient, meteorological data, and road segment attributes is mined. Specifically, based on the second road surface adhesion coefficient uploaded in real time by the roadside unit, combined with meteorological trends and historical similar scene data, the forward adhesion coefficient of the target road segment, namely the third road surface adhesion coefficient, is inferred. This value not only integrates the regional real-time nature of roadside data, but also enhances scene adaptability through meteorological and historical data, and can also link upstream and downstream road segment data to achieve global coordination, providing vehicles with an adhesion coefficient reference beyond line of sight.

[0145] In step S303, the third road surface adhesion coefficient is sent to the vehicle so that the vehicle can determine the target adhesion coefficient based on its own determined first road surface adhesion coefficient, the second road surface adhesion coefficient received from the roadside unit, and the third road surface adhesion coefficient.

[0146] In one embodiment, the cloud platform's data access server receives data packets (including second road surface adhesion coefficients and meteorological data) uploaded from tens of thousands of roadside units and stores them in a distributed database or data lake. The cloud platform's map engine matches and fuses this data with the road chain information of the high-precision map, and generates a dynamic digital map of the entire road network with multiple layers through spatial interpolation and spatiotemporal smoothing algorithms. This map service can be queried externally through an API (Application Programming Interface), returning the historical adhesion coefficient curve, current adhesion coefficient value, and short-term future forecast value for a certain road.

[0147] The cloud platform will send the trained new prediction model or optimized model parameters to the relevant roadside units and V2X vehicles via OTA (Over-the-Air) over a secure link, thus completing the algorithm update and performance improvement of the entire system.

[0148] The second road surface adhesion coefficient is determined by the roadside unit based on the motion state data uploaded by the vehicles within its detection range.

[0149] This disclosure also provides a road surface adhesion coefficient determination system, which includes at least one of a vehicle, a roadside unit, and a cloud platform.

[0150] The vehicle is used to perform the road adhesion coefficient determination method for vehicles provided in this disclosure.

[0151] The roadside unit is used to perform the road surface adhesion coefficient determination method applied to the roadside unit provided in this disclosure.

[0152] The cloud platform is used to execute the road surface adhesion coefficient determination method applied to the cloud platform provided in this disclosure.

[0153] In one embodiment, see Figure 6 As shown, the road surface adhesion coefficient determination system includes a vehicle-side perception and processing module, a roadside unit roadside perception and fusion module, and a cloud platform cloud big data and early warning module.

[0154] The vehicle-side perception and processing module includes: The data acquisition submodule, which is a sensor group mounted on the vehicle, uses visual sensors (such as a forward-facing camera) to capture visual information such as road surface texture, color, and road conditions.

[0155] Dynamic sensors, including IMU (Vehicle-to-Infrastructure), wheel speed sensors, torque sensors, etc., are used to estimate the adhesion coefficient based on the vehicle's dynamic response in real time; GNSS positioning sensors are used to obtain the vehicle's precise latitude, longitude, heading, speed information, etc.

[0156] The vehicle-side estimation submodule runs a lightweight semantic segmentation neural network and a dynamics-based estimation model to achieve a preliminary estimate of the road adhesion coefficient at the single-vehicle level.

[0157] The vehicle-to-everything (V2X) communication module is responsible for packaging and sending the raw data and estimation results from the vehicle to the roadside or cloud.

[0158] The roadside perception and fusion module includes: The roadside perception submodule is equipped with a rich array of sensors, such as cameras, which provide wide-angle, high-definition images of the road surface to overcome the obstruction of a single vehicle's view; and meteorological sensors, which monitor road surface temperature, humidity, precipitation, icing temperature points, etc., and can directly determine risks such as black ice and slippery conditions.

[0159] The roadside unit and edge computing submodule receive estimated data from multiple vehicles within the coverage area, fuse roadside sensor data, and use algorithms such as Kalman filtering or machine learning to generate a road-level adhesion coefficient profile map, which characterizes the adhesion coefficient at different locations on the road segment. Then, it performs preliminary analysis and generates early warning information.

[0160] The roadside communication module integrates the processed road segment-level adhesion coefficient information and early warnings, and broadcasts them in real time to all vehicles in the area via low-latency V2I (Vehicle-to-Infrastructure) communication.

[0161] In a specific embodiment, the cloud-based big data and early warning module includes: The big data platform and collaborative estimation receive and store data uploaded from the roadside, use historical data to train and optimize the adhesion coefficient estimation model, further analyze macroscopic patterns, combine real-time data and historical big data to generate a more accurate and forward-looking heat map of the adhesion coefficient of the entire road network, and then establish and continuously update a dynamic digital map, and integrate the adhesion coefficient as a dynamic attribute layer into the map.

[0162] The early warning service and model delivery system generates tiered early warning information based on the analysis results and sends warnings to vehicles about to enter the risk area via cellular networks or RSUs. Finally, the cloud-trained and optimized model is updated to the vehicle and roadside units via OTA, enabling the entire system to evolve on its own.

[0163] See Figure 7 As shown, the vehicle-side equipment (vehicle), roadside equipment (roadside unit), and cloud platform in the road surface adhesion system determine the bidirectional data interaction through their respective communication modules, and work together to complete the estimation and early warning functions of the road surface adhesion coefficient.

[0164] The vehicle-mounted equipment is integrated into the vehicle and includes a sensing unit, a positioning unit, an on-board computing unit, and a first communication unit.

[0165] The system comprises the following components: a perception unit for acquiring vehicle status and environmental information, including a forward-facing camera, an IMU (Integrated Virtual Machine), and wheel speed sensors; a positioning unit for providing high-precision vehicle location information, including a high-precision GNSS receiver; an onboard computing unit for running vehicle-to-everything (V2X) estimation algorithms, with a built-in AI processing unit; and a first communication unit for data interaction with external systems, including a V2X onboard unit and a cellular network communication module.

[0166] Roadside equipment is deployed at key road nodes and includes a roadside sensing unit, a roadside computing unit, and a second communication unit.

[0167] The roadside sensing unit monitors the road surface and weather conditions of specific road sections and includes cameras and weather sensors. The roadside computing unit performs regional data fusion and computation, and consists of a roadside edge computing server (MEC). The second communication unit is responsible for core communication and computation, communicating with vehicles within the area and the cloud platform; it includes roadside units, V2X communication modules, and device interfaces.

[0168] The cloud platform is deployed in a data center and includes a cloud server cluster and a third communication unit.

[0169] The cloud server cluster provides massive data storage and large-scale computing capabilities; the third communication unit is used to establish connections with vehicle-mounted devices and roadside equipment via a wide area network.

[0170] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0171] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0172] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

Claims

1. A method for determining the road surface adhesion coefficient, characterized in that, Applied to vehicles, the method includes: Acquire first road surface images and vehicle motion state data; The first road surface adhesion coefficient is determined based on the motion state data and the first road surface image; The target road surface adhesion coefficient is determined based on the first road surface adhesion coefficient, the second road surface adhesion coefficient sent by the roadside unit, and the third road surface adhesion coefficient sent by the cloud platform. The second road surface adhesion coefficient is determined by the roadside unit based on the motion state data uploaded by vehicles within its own detection range, and the third road surface adhesion coefficient is determined by the cloud platform based on the second road surface adhesion coefficient and meteorological data uploaded by the roadside unit.

2. The method for determining the road surface adhesion coefficient according to claim 1, characterized in that, Determining the first road surface adhesion coefficient based on the motion state data and the first road surface image includes: Based on the first road surface image and the motion state data, a first visual estimate of the road surface adhesion coefficient is determined; Based on the motion state data, a first dynamic estimate of the road surface adhesion coefficient is determined; The first visual estimate and the first dynamic estimate are weighted and fused to obtain the first road surface adhesion coefficient.

3. The method for determining the road surface adhesion coefficient according to claim 2, characterized in that, The motion state data includes the current vehicle speed and front wheel steering angle. Determining a first visual estimate of the road surface adhesion coefficient based on the first road surface image and the motion state data includes: Based on the current vehicle speed and the front wheel steering angle, a region of interest is determined in the first road surface image, wherein the region of interest is the vehicle's driving area; The road surface type label and confidence score of each pixel in the region of interest are obtained through the pre-trained classification model. The confidence score of the pixel is used to characterize the proportion of pixels with the same road surface type label in the region of interest. By querying a preset mapping table, the target road surface type label is determined to correspond to the target road surface type label. The target road surface type label is the road surface type label with the highest confidence in the region of interest at the current acquisition time. The preset mapping table includes the correspondence between different road surface adhesion coefficients and road surface type labels. If the confidence level of the target road surface type label is greater than or equal to the first threshold, then the target road surface adhesion coefficient is determined as the first visual estimate at the current acquisition time; or If the confidence level of the target road surface type label is less than the first threshold and greater than the second threshold, then the target road surface adhesion coefficient at the current acquisition time is weighted and fused with the first visual estimate at the previous acquisition time to obtain the first visual estimate at the current acquisition time; or If the confidence level of the target road surface type label is less than or equal to the second threshold, then the first visual estimate or preset safety value at the previous acquisition time is determined as the first visual estimate at the current acquisition time.

4. The method for determining the road surface adhesion coefficient according to claim 2, characterized in that, The motion state data includes initial longitudinal acceleration, wheel speed, and net torque at multiple acquisition times. Determining the first dynamic estimate of the road surface adhesion coefficient based on the motion state data includes: If the initial acceleration at the current acquisition moment exceeds the third threshold, or the difference between the net torque at the current acquisition moment and the net torque at the previous acquisition moment exceeds the fourth threshold, the reference vehicle speed, initial longitudinal force, and slip ratio at the current acquisition moment are input into a preset tire model to obtain the first dynamic estimate at the current acquisition moment. The reference vehicle speed at the current acquisition moment is obtained by integrating the initial longitudinal accelerations at multiple acquisition moments; the slip ratio at the current acquisition moment is determined based on the wheel speed and reference vehicle speed at the current acquisition moment; and the initial longitudinal force at the current acquisition moment is determined based on the net torque at the current acquisition moment. The preset tire model is used to characterize the correspondence between the vehicle's longitudinal force, slip ratio, and road adhesion coefficient; or If the initial acceleration at the current acquisition time does not exceed the third threshold, or the difference between the net torque at the current acquisition time and the net torque at the previous acquisition time does not exceed the fourth threshold, the first dynamic estimate calculated at the previous acquisition time is determined as the first dynamic estimate at the current acquisition time. The step of inputting the reference vehicle speed, initial longitudinal force, and slip ratio at the current acquisition time into a preset tire model to obtain the first dynamic estimate at the current acquisition time includes: The initial longitudinal force at the current acquisition time is input into the prediction model to obtain the predicted acceleration at the current acquisition time. The prediction model is a model that predicts longitudinal acceleration based on vehicle mass and longitudinal force. Based on the residual between the initial longitudinal acceleration and the predicted longitudinal acceleration at the current acquisition time, adjust the initial longitudinal force at the current acquisition time to determine the target longitudinal force at the current acquisition time. The regression vector is determined based on the slip ratio and target longitudinal force at the current acquisition moment; The Kalman gain at the current acquisition time is determined based on the regression vector, the first dynamic estimate at the previous acquisition time, and the covariance matrix at the previous acquisition time. Based on the Kalman gain and the error at the current acquisition time, a first dynamic estimate for the current acquisition time is obtained, and the covariance matrix at the current acquisition time is updated. The error at the current acquisition time is determined based on the target longitudinal force at the current acquisition time, the regression vector, and the first dynamic estimate from the previous acquisition time.

5. The method for determining the road surface adhesion coefficient according to claim 2, characterized in that, The step of weightedly fusing the first visual estimate and the first dynamic estimate to obtain the first road surface adhesion coefficient includes: A first confidence level corresponding to the first visual estimate and a second confidence level corresponding to the first dynamic estimate are determined, wherein the first confidence level is determined based on the confidence level of the road surface type label corresponding to the first visual estimate, and the second confidence level is determined based on at least one of the data quality, data convergence degree, and longitudinal excitation degree of the vehicle corresponding to the first dynamic estimate. When both the first confidence level and the second confidence level are within the range of the first confidence level, the first visual estimate and the first dynamic estimate are weighted and fused according to the first confidence level and the second confidence level to determine the first road surface adhesion coefficient; When either the first confidence level or the second confidence level is not within the range of the first confidence level, the estimated value corresponding to the confidence level that is within the range of the first confidence level is determined as the first road surface adhesion coefficient; When neither the first confidence level nor the second confidence level is within the range of the first confidence level, the preset safety value is determined as the first road surface adhesion coefficient.

6. The method for determining the road surface adhesion coefficient according to claim 2, characterized in that, The step of determining the target road surface adhesion coefficient based on the first road surface adhesion coefficient, the second road surface adhesion coefficient sent by the roadside unit, and the third road surface adhesion coefficient sent by the cloud platform includes: Based on the preset consistency tolerance parameter, the first visual estimate, and the first dynamic estimate, the posterior confidence level is determined, and the confidence levels corresponding to the posterior confidence level, the first visual estimate, and the first dynamic estimate are weighted and averaged to determine the third confidence level corresponding to the first road surface adhesion coefficient. Receive the fourth confidence level corresponding to the second road surface adhesion coefficient sent by the roadside unit and the fifth confidence level corresponding to the third road surface adhesion coefficient sent by the cloud platform; If the third confidence level is greater than or equal to the first preset confidence level, the first road surface adhesion coefficient is determined as the target road surface adhesion coefficient; or If the third confidence level is less than or equal to the second preset confidence level, and both the fourth and fifth confidence levels are greater than or equal to the first preset confidence level, then either the second road surface adhesion coefficient or the third road surface adhesion coefficient shall be used as the target road surface adhesion coefficient; or If the third confidence level, the fourth confidence level, and the fifth confidence level are all less than the first preset confidence level and greater than the second preset confidence level, then the first road surface adhesion coefficient, the second road surface adhesion coefficient, and the third road surface adhesion coefficient are weighted and fused according to the third confidence level, the fourth confidence level, and the fifth confidence level to obtain the target road surface adhesion coefficient; or If the third confidence level, the fourth confidence level, and the fifth confidence level are all less than the second preset confidence level, the preset safety value is determined as the target road surface adhesion coefficient.

7. A method for determining the road surface adhesion coefficient, characterized in that, Applied to roadside units, the method includes: It receives motion status data from multiple vehicles within its detection range and collects second road surface images and meteorological data; The second road surface adhesion coefficient is determined based on the motion state data of the multiple vehicles and the second road surface image; The meteorological data and the second road surface adhesion coefficient are sent to the cloud platform so that the cloud platform can determine the third road surface adhesion coefficient. The second road surface adhesion coefficient is broadcast outward so that vehicles that receive the second road surface adhesion coefficient can determine the target road surface adhesion coefficient based on their own determined first road surface adhesion coefficient, the second road surface adhesion coefficient, and the third road surface adhesion coefficient issued by the cloud platform.

8. The method for determining the road surface adhesion coefficient according to claim 7, characterized in that, The step of determining the second road surface adhesion coefficient based on the motion state data of the multiple vehicles and the second road surface image includes: Based on the second road surface image, a second visual estimate of the road surface adhesion coefficient is determined; Based on the motion state data of the multiple vehicles, a second dynamic estimate of the road surface adhesion coefficient is determined; The second visual estimate and the second dynamic estimate are weighted and fused to obtain the second road surface adhesion coefficient; The step of determining the second visual estimate of the road surface adhesion coefficient based on the second road surface image includes: The road type label for each pixel in the second road image is obtained by using a pre-trained classification model. Based on the preset map link information, select the drivable area of ​​interest in the second road surface image; The road surface type label of the pixel with the highest confidence in the region of interest is determined as the target road surface type label. The confidence of the pixel is used to characterize the proportion of pixels with the same road surface type label in the region of interest. By querying a preset mapping table, the target road surface adhesion coefficient corresponding to the target road surface type label is determined as the second visual estimate.

9. A method for determining the road surface adhesion coefficient, characterized in that, Applied to a cloud platform, the method includes: Receive meteorological data and second pavement adhesion coefficient uploaded by at least one roadside unit; Based on the meteorological data and the second road surface adhesion coefficient, the third road surface adhesion coefficient is predicted. The third road surface adhesion coefficient is sent to the vehicle so that the vehicle can determine the target adhesion coefficient based on its own determined first road surface adhesion coefficient, the second road surface adhesion coefficient received from the roadside unit, and the third road surface adhesion coefficient. The second road surface adhesion coefficient is determined by the roadside unit based on the motion state data uploaded by the vehicle within its detection range.

10. A system for determining the road surface adhesion coefficient, characterized in that, The system includes at least one of a vehicle, a roadside unit, and a cloud platform; The vehicle is used to perform the method according to any one of claims 1-6; The roadside unit is used to perform the method according to any one of claims 7-8; The cloud platform is used to perform the method of claim 9.

Images