A multi-modal pavement water hazard evaluation method and system
By integrating multimodal data and conducting dynamic risk assessment, the problem of insufficient sensitivity and high false detection rate in existing water accumulation monitoring technologies has been solved, enabling efficient and accurate assessment and real-time early warning of road water damage.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA MERCHANTS CHONGQING COMM RES & DESIGN INST
- Filing Date
- 2026-01-28
- Publication Date
- 2026-06-16
AI Technical Summary
Existing technologies for water accumulation monitoring and early warning rely on manual inspections and fixed threshold-based judgment methods, which lack sensitivity, resulting in a high false detection rate and making it difficult to achieve all-weather dynamic monitoring and accurate assessment of road water damage risks.
A multimodal pavement water hazard assessment method is adopted. By combining road construction data, vehicle driving data and multispectral images with the DBSCAN algorithm and water balance equation, a three-dimensional point cloud model is constructed to detect and classify water accumulation areas from multiple angles. LSTM and XGBoost models are used for dynamic risk prediction, and reflectivity difference algorithm is used for accurate identification. Finally, the water dissipation time is constructed to determine the road water hazard risk.
It achieves highly sensitive detection and accurate delineation of waterlogged areas on roads, reduces false detection rate, improves the accuracy and real-time performance of road water damage assessment, and reduces reliance on manual inspection.
Smart Images
Figure CN122222360A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of road engineering technology, and in particular to a multimodal pavement water hazard assessment method and system. Background Technology
[0002] With the rapid increase in highway maintenance mileage and the frequent occurrence of extreme weather events, road surface water accumulation has become a core issue threatening road traffic safety, accelerating pavement damage, and weakening roadbed stability. Especially in high-risk sections such as superelevation curves, bridge approach transitions, and vehicle wheel tracks, water accumulation can easily cause vehicle skidding, roadbed settlement, and drainage system failure, leading to a significant increase in traffic accident rates and road maintenance costs. However, existing technologies have significant limitations in water accumulation monitoring and early warning: traditional detection methods rely on manual inspections or single sensors, making it difficult to achieve all-weather dynamic monitoring; water accumulation identification methods based on fixed thresholds are insufficiently sensitive to differences in pavement materials, pollutant coverage, and changes in rainfall intensity, resulting in a high false detection rate. Summary of the Invention
[0003] To address the shortcomings of existing technologies, this invention provides a multimodal road surface water hazard assessment method, which solves the problems of insufficient sensitivity and low accuracy caused by relying on manual inspections and making judgments based on fixed thresholds in existing technologies.
[0004] According to an embodiment of the present invention, a multimodal pavement water hazard assessment method includes: Obtain road construction data, divide the road surface into multiple first-risk zones based on the road construction data, and calculate the corresponding first-risk values; Acquire vehicle driving data, divide the road surface into multiple secondary risk zones based on the vehicle driving data, and calculate the corresponding secondary risk values; Acquire multispectral images of the road, use a reflectance difference algorithm to divide the road surface into multiple third-risk areas and calculate the corresponding third-risk values; The DBSCAN algorithm is used to spatially aggregate the first risk area, the second risk area, and the third risk area to obtain multiple aggregated areas. The comprehensive risk value of each aggregated area is calculated based on the first risk value, the second risk value, and the third risk value. Then, the aggregated areas with a comprehensive risk value greater than a preset value are identified as water accumulation areas. A 3D point cloud model of each waterlogged area is constructed, and the corresponding water volume is calculated based on the 3D point cloud model. Then, the corresponding water dissipation time is calculated based on the water balance equation, and the road surface water damage risk is determined based on the water dissipation time of all waterlogged areas.
[0005] Preferably, the method for dividing the road surface into multiple first-risk zones and calculating the corresponding first-risk values based on road construction data includes: Based on road construction data, the road surface is divided into multiple first-risk zones, and then the image features of each first-risk zone are extracted. Based on the image features of each first-risk region, key features are extracted using an attention mechanism, and then the corresponding first-risk value is calculated using a trained logit model based on the key features.
[0006] Preferably, the method for dividing the road surface into multiple second risk zones based on vehicle driving data and calculating the corresponding second risk values includes: Based on vehicle driving data, an LSTM model is used to predict the dynamic trend of vehicle driving on the road, and the road surface is divided into multiple secondary risk zones according to the driving dynamic trend. Vehicle driving data and future driving data are concatenated, and then the XGBoost model is used to calculate the second risk value for each risk change region based on the concatenated vehicle driving data and future driving data.
[0007] Preferably, the method for dividing the road surface into multiple third risk zones and calculating the corresponding third risk values using a reflectivity difference algorithm includes: The brightness difference curve of the road is calculated based on the multispectral image, and the road is divided according to the brightness difference curve to obtain multiple third risk areas. Then, the radiometric calibration coefficient is calculated based on the brightness difference curve. The reflectance difference value of each third risk region is calculated based on the multispectral image and radiometric calibration coefficient of each third risk region. The third risk value for each third risk zone is determined based on the magnitude between the reflectance difference value and the preset threshold.
[0008] Preferably, after obtaining the reflectance difference value, the road surface material is obtained based on the road construction data, the pollution sensitivity coefficient of the third risk area is determined based on the multispectral image, and then the reflectance difference value is corrected based on the road surface material and the pollution sensitivity coefficient.
[0009] Preferably, the correction formula for the reflectance difference value is as follows: in, This is the corrected reflectance difference value. The difference in reflectance before correction; λ is the road surface material weighting factor; λ is the pollution sensitivity coefficient; C is the pollutant coverage rate (0-1).
[0010] Preferably, if the reflectance difference value is greater than or equal to a preset value, then the third risk value of the corresponding third risk area is 1; If the reflectance difference is less than the preset value, the third risk value is calculated using linear interpolation.
[0011] On the other hand, according to embodiments of the present invention, a multimodal pavement water hazard assessment system is also provided. This system uses the above-described multimodal pavement water hazard assessment method, including: The data acquisition module is used to acquire road construction data, vehicle driving data and multispectral images of the road, and to construct a three-dimensional point cloud model of each water accumulation area, and to calculate the corresponding water accumulation volume based on the three-dimensional point cloud model. The area division module is used to divide the road surface into multiple first risk areas based on road construction data and calculate the corresponding first risk value, divide the road surface into multiple second risk areas based on vehicle driving data and calculate the corresponding second risk value, and divide the road surface into multiple third risk areas using a reflectivity difference algorithm and calculate the corresponding third risk value. The aggregation judgment module is used to spatially aggregate the first risk area, the second risk area, and the third risk area using the DBSCAN algorithm to obtain multiple aggregated areas. It calculates the comprehensive risk value of each aggregated area based on the first risk value, the second risk value, and the third risk value, and then identifies the aggregated areas with a comprehensive risk value greater than a preset value as water accumulation areas. The water damage assessment module is used to calculate the corresponding water dissipation time based on the water balance equation, and to determine the road surface water damage risk based on the water dissipation time corresponding to all waterlogged areas.
[0012] Compared with the prior art, the present invention has the following beneficial effects: This invention uses three different methods—road construction data, vehicle driving data, and multispectral images of roads—to classify areas where water accumulation may occur on roads. By using multimodal data and corresponding classification methods, water accumulation areas are detected and classified from multiple perspectives, improving the sensitivity to various road data. This eliminates the need for manual inspection. The DBSCAN algorithm is then used to spatially aggregate the areas classified by the three methods, resulting in multiple aggregated regions. The corresponding comprehensive risk is calculated to accurately detect water accumulation areas. Finally, the degree of road flooding is accurately determined based on the dissipation time of the water in the water accumulation areas. Attached Figure Description
[0013] Figure 1 This is a diagram illustrating the multimodal pavement water hazard assessment method according to an embodiment of the present invention. Detailed Implementation
[0014] The technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments.
[0015] like Figure 1 As shown in the figure, this invention proposes a multimodal pavement water hazard assessment method, including: Obtain road construction data, divide the road surface into multiple first-risk zones based on the road construction data, and calculate the corresponding first-risk values; Road construction data includes existing engineering construction documents, as-built drawings, construction drawings, and maintenance data. From this data, the primary risk area for potential water accumulation is identified, and the image features of this primary risk area are extracted. As shown in Table 1.
[0016] Table 1: Characteristics of the First Risk Zone Value table Then the logit model is used. The system learns the impact of "water accumulation characteristics on single-scenario risk" and outputs a risk score through an attention mechanism. By studying the "importance of multiple scenarios involving water accumulation," a weighted summation is finally performed to obtain the first risk value for each first risk area: in, The dynamic attention weight (dimensionless) for the i-th first risk region. The vector parameters (dimensionless) of the attention mechanism. This is the weight matrix (unitless) for the attention mechanism. The risk score (dimensionless) for the i-th scenario output by the teacher model. For the image features of the i-th first risk region, This is the weight matrix (dimensionless) for the logit model. This is the bias term (dimensionless) of the logit model.
[0017] Acquire vehicle driving data, and based on the vehicle driving data, determine the road surface to be divided into multiple secondary risk zones and calculate the corresponding secondary risk values; Vehicle driving data includes dynamic parameters such as braking frequency and lane departure rate collected in real time when vehicles are driving on the road. After normalization, these parameters are optimized by the collaboration of LSTM network and XGBoost classifier to achieve dynamic, accurate and multi-dimensional prediction of water accumulation risk.
[0018] (1) Dynamic feature extraction Where X is the feature vector of vehicle driving data, which contains all vehicle driving data. This feature vector is a time feature sequence, from "t - n" to "t - 1", and n is the window size.
[0019] (2) Prediction of water accumulation risk trend The risk of flooding is continuous over time. Therefore, LSTM learns the "feature change patterns of the past n time steps" to capture the causal relationship in the time dimension and predict future risk trends.
[0020] in, To represent the dynamic trend of vehicle travel at time t in the LSTM model, such as "risk increase", "risk decrease", or "risk stability", the trend is numerically represented as: risk increase = 1, risk decrease = -1, and risk stability = 0. Simultaneously, the road is divided into multiple secondary risk regions based on the driving dynamic trend. For example, if the first half of the driving dynamic trend is a continuous "risk increase" and the second half is a continuous "risk stability", then the road segments corresponding to the continuous "risk increase" are classified as one secondary risk region, and the road segments corresponding to the continuous "risk stability" are classified as another secondary risk region.
[0021] Then the LSTM output As a new feature dimension, it is merged with the feature vector X of the vehicle driving data: This results in richer input information.
[0022] (3) Real-time risk level determination XGBoost can learn instantaneous features and historical trends simultaneously, constructing a mapping rule from "feature to risk level" to improve the accuracy of judging "progressive risks" (such as continuous rainfall leading to worsening waterlogging) and real-time classification.
[0023] in, The second risk value (dimensionless) for a certain risk variation region output by XGBoost.
[0024] Acquire multispectral images of the road, use a reflectance difference algorithm to divide the road surface into multiple third risk regions and calculate the corresponding third risk values; The multi-temporal reflectance difference model constructed based on the reflectance difference algorithm achieves accurate identification and dynamic prediction of waterlogged road sections through multi-stage data acquisition and dynamic modeling, combined with the physical characteristics of multispectral images of roads and the meteorological-hydrological coupling mechanism.
[0025] Based on the multispectral image, the model will output two metrics: brightness difference. and the difference in reflectance The former reflects the visual difference between waterlogged and dry road surfaces in multispectral images, while the latter reflects the difference in physical reflectance characteristics between waterlogged and dry road surfaces. The two are correlated through multispectral inversion and benchmark calibration to jointly support the accuracy of waterlogging detection.
[0026] Brightness difference calculation: in, Let be the brightness value of the road at time t (dimensionless). This is the road surface roughness correction factor (0.8-1.2). The angle of incidence of the sun (°). The inherent single-band reflection characteristics of the water surface (0.05-0.25). The sensor's depression angle (°) The image attenuation coefficient (1 / s) is caused by atmospheric turbulence.
[0027] Based on the brightness difference calculated by the above formula, all brightness differences are fitted to obtain the brightness difference curve between brightness difference and time. Based on the brightness difference curve, the upper and lower limits of the brightness difference of the road in the whole time domain can be obtained. Then, this interval is divided into multiple smaller intervals. The brightness difference curve is divided according to the interval range between the smaller intervals to obtain multiple line segments. For each line segment, the road is divided according to the road area corresponding to the corresponding time period to obtain multiple third risk areas.
[0028] Furthermore, the brightness difference reflects the visual difference between water accumulation and dry road surfaces in multispectral images. Therefore, the third risk area can be initially delineated based on the brightness difference. If the brightness difference is small, it indicates that there is no water accumulation in the third risk area; if the brightness difference is large, it indicates that there is water accumulation in the third risk area. For the difference in reflectance First, the echo intensity of the onboard LiDAR from the vehicle's driving data needs to be used to perform radiometric calibration on the multispectral image: Where γ is the radiation calibration coefficient; The echo intensity of the vehicle-mounted lidar; The reflectance of the multispectral image.
[0029] For each third risk region, calculate its corresponding reflectance difference value: in, The inversion reflectance (dimensionless) of the dry road surface area in the third risk region is the i-th band of the multispectral image. denoted as the dimensionless inversion reflectance of the waterlogged area in the i-th band of the multispectral image within the third risk region, where n is the number of bands involved in the calculation.
[0030] Then, material weighting factors were introduced. And the pollution sensitivity coefficient λ, corrected for the difference in reflectance: in, This is the corrected reflectance difference value. The difference in reflectance before correction; λ is the road surface material weighting factor; λ is the pollution sensitivity coefficient; C is the pollutant coverage rate (0-1).
[0031] The preset threshold is dynamically adjusted based on road surface material (asphalt, concrete, etc.) and meteorological data. : Only when the reflectivity difference value ≥ When the probability is 20%, the probability of determining the third risk area as a valid area is 100%. =1, if 15%≤ If the risk level is less than 20%, then the third risk area may be a valid area. The probability of the third risk value is calculated using linear interpolation. 1.
[0032] By using multimodal data and corresponding segmentation methods, waterlogged areas can be detected and divided from multiple perspectives, improving the sensitivity to various road data, while eliminating the need for manual inspection.
[0033] The DBSCAN algorithm is used to spatially aggregate the first risk area, the second risk area, and the third risk value to obtain multiple aggregated areas. The comprehensive risk value of each aggregated area is calculated based on the first risk value, the second risk value, and the third risk value. Then, the aggregated areas with a comprehensive risk value greater than a preset value are identified as water accumulation areas. The DBSCAN algorithm is used to spatially aggregate the overlapping areas of the first risk region, the second risk region, and the third risk value, resulting in multiple aggregated regions. The comprehensive risk value of each overlapping region is then calculated. Then, the aggregated areas with a comprehensive risk value greater than 0.6 were designated as water accumulation areas.
[0034] A 3D point cloud model of each waterlogged area is constructed, and the corresponding water volume is calculated based on the 3D point cloud model. Then, the corresponding water dissipation time is calculated based on the water balance equation, and the road surface water damage risk is determined based on the water dissipation time of all waterlogged areas.
[0035] After identifying the waterlogged area, a drone equipped with a 3D scanning device is used to scan the entire road and extract the 3D point cloud of the waterlogged area to construct a 3D point cloud model of the waterlogged area. Based on the 3D point cloud model, the volume and surface area of the waterlogged area are calculated.
[0036] According to the following water balance equation, the water dissipation time corresponds to the water accumulation area in each water accumulation zone: in, Water volume (mm) 3 / min), K is the permeability coefficient (mm / min), determined according to the degree of pavement crack development. Cracked pavement sections: Where e is the crack width, The kinematic viscosity coefficient of the water flow; road surface without cracks: 10−7~10−5cm / s, where S is the surface area of the accumulated water (m²) 2 ), Rainfall intensity (mm / min) Evaporation rate (mm / min) τ represents the time interval between rainfall stages (min), and τ is the environmental response time constant (min).
[0037] Then determine the road surface water damage risk value for each waterlogged area: Among them, α, β, γ, All are weighting coefficients. This is the vectorized value of wind speed / wind direction (dimensionless).
[0038] Then, the CNN-GRU model is used to identify the water type in each waterlogged area. For each waterlogged area, multiple water types and corresponding confidence levels are obtained. In this invention, the water type with the highest confidence level is selected as the water type of the waterlogged area, and the road water damage level of the waterlogged area is judged comprehensively based on the road water damage risk value of each waterlogged area and the confidence level of the corresponding water type.
[0039] Due to the complexity and constant changes in the road environment, the threshold for classifying road surface water damage levels will change dynamically. Therefore, it is necessary to calculate it based on relevant road environment data at the current moment. The calculation formula for the classification threshold is as follows: in, The base threshold (s) is the static initial reference value of the traffic control system determined based on historical road data. γ and δ are the adjustment coefficients (dimensionless), and V is the real-time traffic flow (veh / h) at the current moment. This represents the maximum capacity of road traffic (veh / h). The rainfall intensity (mm / h) is then normalized to the range (0, 0.3) using the Sigmoid function.
[0040] Then, the road surface water damage risk value and the confidence level of water accumulation type for each waterlogged area are weighted to obtain the comprehensive risk value. The risk level of road surface water damage in waterlogged areas is classified according to the comprehensive risk value and the classification threshold: Finally, maintenance priorities were ranked according to the road surface water damage risk level of each waterlogged area, and maintenance was carried out on each part of the road in that order.
[0041] On the other hand, embodiments of the present invention also provide a multimodal pavement water hazard assessment system, which uses the above-mentioned multimodal pavement water hazard assessment method, including: The data acquisition module is used to acquire road construction data, vehicle driving data and multispectral images of the road, and to construct a three-dimensional point cloud model of each water accumulation area, and to calculate the corresponding water accumulation volume based on the three-dimensional point cloud model. The area division module is used to divide the road surface into multiple first risk areas based on road construction data and calculate the corresponding first risk value; to divide the road surface into multiple second risk areas based on vehicle driving data and calculate the corresponding second risk value; and to divide the road surface into multiple third risk areas using a reflectivity difference algorithm and calculate the corresponding third risk value. The aggregation judgment module is used to spatially aggregate the first risk area, the second risk area, and the third risk value using the DBSCAN algorithm to obtain multiple aggregated areas, and calculate the comprehensive risk value of each aggregated area based on the first risk value, the second risk value, and the third risk value. Then, the aggregated areas with a comprehensive risk value greater than a preset value are identified as water accumulation areas. The water damage assessment module is used to calculate the corresponding water dissipation time based on the water balance equation, and to determine the road surface water damage risk based on the water dissipation time corresponding to all waterlogged areas.
[0042] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A multimodal pavement water hazard assessment method, characterized in that: include: Obtain road construction data, divide the road surface into multiple first-risk zones based on the road construction data, and calculate the corresponding first-risk values; Acquire vehicle driving data, divide the road surface into multiple secondary risk zones based on the vehicle driving data, and calculate the corresponding secondary risk values; Acquire multispectral images of the road, use a reflectance difference algorithm to divide the road surface into multiple third-risk areas and calculate the corresponding third-risk values; The DBSCAN algorithm is used to spatially aggregate the first risk area, the second risk area, and the third risk area to obtain multiple aggregated areas. The comprehensive risk value of each aggregated area is calculated based on the first risk value, the second risk value, and the third risk value. Then, the aggregated areas with a comprehensive risk value greater than a preset value are identified as water accumulation areas. A 3D point cloud model of each waterlogged area is constructed, and the corresponding water volume is calculated based on the 3D point cloud model. Then, the corresponding water dissipation time is calculated based on the water balance equation, and the road surface water damage risk is determined based on the water dissipation time of all waterlogged areas.
2. The multimodal pavement water hazard assessment method as described in claim 1, characterized in that: Methods for dividing road surfaces into multiple primary risk zones and calculating corresponding primary risk values based on road construction data include: Based on road construction data, the road surface is divided into multiple first-risk zones, and then the image features of each first-risk zone are extracted. Based on the image features of each first-risk region, key features are extracted using an attention mechanism, and then the corresponding first-risk value is calculated using a trained logit model based on the key features.
3. The multimodal pavement water hazard assessment method as described in claim 1, characterized in that: Methods for dividing road surfaces into multiple secondary risk zones and calculating corresponding secondary risk values based on vehicle driving data include: Based on vehicle driving data, an LSTM model is used to predict the dynamic trend of vehicle driving on the road, and the road surface is divided into multiple secondary risk zones according to the driving dynamic trend. Vehicle driving data and future driving data are concatenated, and then the XGBoost model is used to calculate the second risk value for each risk change region based on the concatenated vehicle driving data and future driving data.
4. The multimodal pavement water hazard assessment method as described in claim 1, characterized in that: Methods for dividing road surfaces into multiple third-risk zones and calculating corresponding third-risk values using a reflectivity difference algorithm include: The brightness difference curve of the road is calculated based on the multispectral image, and the road is divided according to the brightness difference curve to obtain multiple third risk areas. Then, the radiometric calibration coefficient is calculated based on the brightness difference curve. The reflectance difference value of each third risk region is calculated based on the multispectral image and radiometric calibration coefficient of each third risk region. The third risk value for each third risk zone is determined based on the magnitude between the reflectance difference value and the preset threshold.
5. The multimodal pavement water hazard assessment method as described in claim 1, characterized in that: After obtaining the reflectance difference value, the road surface material is obtained based on the road construction data, the pollution sensitivity coefficient of the third risk area is determined based on the multispectral image, and then the reflectance difference value is corrected based on the road surface material and the pollution sensitivity coefficient.
6. The multimodal pavement water hazard assessment method as described in claim 1, characterized in that: The correction formula for the reflectance difference value is as follows: in, This is the corrected reflectance difference value. The difference in reflectance before correction; λ is the road surface material weighting factor; λ is the pollution sensitivity coefficient; C is the pollutant coverage rate (0-1).
7. The multimodal pavement water hazard assessment method as described in claim 1, characterized in that: If the reflectance difference value is greater than or equal to the preset value, the third risk value of the corresponding third risk area is 1; If the reflectance difference is less than the preset value, the third risk value is calculated using linear interpolation.
8. A multimodal pavement water hazard assessment system, characterized in that: The system uses a multimodal pavement water hazard assessment method as described in any one of claims 1-7, comprising: The data acquisition module is used to acquire road construction data, vehicle driving data and multispectral images of the road, and to construct a three-dimensional point cloud model of each water accumulation area, and to calculate the corresponding water accumulation volume based on the three-dimensional point cloud model. The area division module is used to divide the road surface into multiple first risk areas based on road construction data and calculate the corresponding first risk value, divide the road surface into multiple second risk areas based on vehicle driving data and calculate the corresponding second risk value, and divide the road surface into multiple third risk areas using a reflectivity difference algorithm and calculate the corresponding third risk value. The aggregation judgment module is used to spatially aggregate the first risk area, the second risk area, and the third risk area using the DBSCAN algorithm to obtain multiple aggregated areas. It calculates the comprehensive risk value of each aggregated area based on the first risk value, the second risk value, and the third risk value, and then identifies the aggregated areas with a comprehensive risk value greater than a preset value as water accumulation areas. The water damage assessment module is used to calculate the corresponding water dissipation time based on the water balance equation, and to determine the road surface water damage risk based on the water dissipation time corresponding to all waterlogged areas.