A road foggy day dangerous working condition recognition processing method based on multi-sensor fusion

Abstract

The present application relates to a kind of road fog dangerous working condition identification processing method based on multi-sensor fusion, the method includes: step 1, camera, millimeter wave radar and laser radar specificity degradation features under fog environment are extracted in parallel;Step 2, based on the feature real-time evaluation sensor confidence and dynamic weighted fusion, obtain comprehensive visibility estimation V final =η*[W cam *V vis +W rad *φ(C rv )+W lid *ψ(1 / λ)], And dangerous working condition level;Step 3, according to dangerous working condition level, establish the accurate mapping lookup table of grade and vehicle control instruction, according to accurate mapping lookup table, execute the vehicle active safety control strategy corresponding to dangerous working condition level;Step 4, comprehensive visibility estimation and dangerous working condition level are packaged as standard car-road cooperation message, broadcast through vehicle-mounted communication unit, realize the vehicle-vehicle, vehicle-road early warning of over-the-horizon.This technical scheme solves the technical problems such as the present technology fusion level is shallow, information utilization rate is low, output result is rough, lack of adaptive ability etc.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion. Background Technology

[0002] Dense fog, especially sudden and localized patches of fog, is a major cause of catastrophic chain-reaction rear-end collisions on highways in low visibility conditions. Traditional defenses rely on fixed weather station monitoring and manual warnings, which suffer from large monitoring blind spots, high warning delays, and coarse information granularity. With the widespread adoption of intelligent connected vehicles (ICVs), their standard-equipped cameras, millimeter-wave radar, and lidar provide the hardware foundation for real-time, high-resolution road weather perception. Utilizing these mobile sensing sources for hazardous condition identification has become a key technological direction for improving active driving safety. Currently, similar technical solutions in the industry mostly focus on "weather state classification based on decision-level fusion of visual and radar data." These solutions typically use cameras for image classification (such as identifying "foggy weather"), supplemented by radar data (such as changes in the number of targets) for simple verification, ultimately triggering a unified "foggy weather mode." Summary of the Invention

[0003] In view of the above problems, the present invention provides a method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion, in order to solve the technical problems of shallow fusion layer, low information utilization, coarse output results, lack of driving guidance value, serious decoupling between perception and vehicle control, system rigidity, and lack of adaptive capability in the existing technology.

[0004] This invention provides a method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion. The method includes: Step 1, extracting specific degradation features of cameras, millimeter-wave radar, and lidar in foggy environments in parallel; Step 2, evaluating sensor confidence in real time based on the features and performing dynamic weighted fusion to obtain a comprehensive visibility estimate V. final =η*[W cam *V vis +W rad *φ(C rv )+W lid *ψ(1 / λ)], and the hazardous condition level, where φ() and ψ() are mapping functions established based on calibration data, η is the global calibration coefficient, and W cam W rad W lid The dynamic confidence weights for cameras, millimeter-wave radar, and lidar are V, respectively. vis As the initial value for visual visibility estimation, C rvλ is the millimeter-wave radar-visual contradiction index, and λ is the attenuation coefficient; Step 3: Establish a precise mapping lookup table between the hazard level and the vehicle control command based on the hazard level, and execute the vehicle active safety control strategy corresponding to the hazard level according to the precise mapping lookup table; Step 4: Encapsulate the comprehensive visibility estimation and hazard level into a standard vehicle-road cooperative message, and broadcast it through the vehicle communication unit to realize beyond-line-of-sight vehicle-to-vehicle and vehicle-to-road early warning.

[0005] Furthermore, the method also includes: Step 0, using the vehicle's high-precision time reference and high-speed vehicle network to perform hardware-level time synchronization of camera image frames, millimeter-wave radar target list, and lidar point cloud, and unifying all perception data into a dynamic coordinate system with the vehicle's center of mass as the origin through offline calibration parameters and online motion compensation algorithms.

[0006] Furthermore, the high-precision time reference for the vehicle is the PPS signal of GNSS.

[0007] Furthermore, step 1 includes: Step 11, based on the atmospheric scattering physical model, calculating the dark channel intensity and color saturation attenuation of the image as a physical estimate of atmospheric transmittance, then using a lightweight encoder network to extract the high-frequency detail loss coefficients and global texture entropy values ​​of the image, outputting a preliminary visual visibility estimate, and finally forming the visual degradation feature set F of the camera in foggy conditions. v Step 12: After spatiotemporal alignment, perform association matching between radar targets and visually detected targets, and calculate the millimeter-wave radar-visual contradiction index C. rv Then, by monitoring the average signal-to-noise ratio change rate ΔSNR of the millimeter-wave radar receiving link, the penetration characteristic group F of the millimeter-wave radar in foggy conditions is finally formed. r Step 13: Divide the point cloud into distance segments, count the number of effective points in each segment, fit an exponential decay curve of point cloud density with distance, obtain the attenuation coefficient λ that characterizes the extinction properties of fog, and then analyze the variance σ of the reflection intensity distribution of point clouds of known high reflectivity objects. int Finally, the attenuation characteristic group F of the lidar in foggy conditions was obtained. l .

[0008] Furthermore, step 2 includes: step 21, placing F v F r F l The vehicle status is input into a pre-set micro-evaluation network, which infers online the dynamic confidence weight W of each sensor at the current moment. cam W rad W lid Furthermore, the sum of the normalized weights of each dynamic confidence level is 1; Step 22, the final visibility V is calculated using a weighted fusion formula. final =η*[Wcam *V vis +W rad *φ(C rv )+W lid *ψ(1 / λ)], where φ() and ψ() are mapping functions established based on a large amount of calibration data, η is the global calibration coefficient, and W cam W rad W lid The dynamic confidence weights for cameras, millimeter-wave radar, and lidar are V, respectively. vis As the initial value for visual visibility estimation, C rv The millimeter-wave radar-visual contradiction index is given, where λ is the attenuation coefficient; step 23, the fused feature vector [V] is... final C rv ,λ,σ int Input v into the preset classifier to obtain the danger level L0 / L1 / L2 / L3, where L0 is no risk, L1 is a mild warning, L2 is a severe warning, L3 is extremely dangerous, and v is the vehicle speed.

[0009] Furthermore, the WeChat evaluation network is a two-layer fully connected neural network, the vehicle state includes vehicle speed and yaw rate, and the classifier is a gradient boosting tree (GBDT).

[0010] Furthermore, the precise mapping lookup table includes: L1: Projecting fog markers on the head-up display, recommending a vehicle speed ≤80km / h, and increasing the adaptive cruise following distance by 20%; L2: Triggering an audible and visual alarm, automatically limiting the system speed to 60km / h, setting the adaptive cruise to the maximum safe following distance, and forcibly activating the front and rear fog lights and hazard warning lights; L3: Activating the highest level of audiovisual alarm, requesting the driver to take immediate control, and if the vehicle is in autonomous driving mode, executing the "minimum risk strategy," smoothly decelerating and safely stopping within the lane.

[0011] Furthermore, the standard vehicle-road cooperative message also includes a timestamp and the vehicle's latitude and longitude.

[0012] Furthermore, the method also includes: step 5, continuously collecting actual response data after the vehicle executes the active safety control strategy, using it as a feedback signal to perform online incremental learning on a preset micro-evaluation network, so that the algorithm can continuously adapt to changes in individual vehicle characteristics and sensor performance.

[0013] Furthermore, the actual response data includes the actual deceleration and whether the vehicle stability system (ESP) is triggered.

[0014] This invention provides a method for identifying and processing hazardous road fog conditions based on multi-sensor fusion. It primarily addresses the shortcomings of existing technologies, such as: inability to handle the non-uniform characteristics of patchy fog, resulting in limited accuracy and reliability; inability to output quantitative visibility values ​​crucial for autonomous or assisted driving decisions; inability to distinguish between the risk levels of diffuse light fog and sudden, deadly patches of fog; failure to form a closed-loop linkage with the vehicle's longitudinal (Adaptive Cruise Control, Automatic Emergency Braking, AEB) and lateral control systems, hindering real-time risk-based active speed and distance control, leading to low preventative safety benefits; and inability to intelligently adjust based on dynamic contexts such as sensor real-time performance confidence, vehicle speed, and road curvature, resulting in false alarms or missed alarms and poor robustness in complex real-world scenarios. Attached Figure Description

[0015] Figure 1 A flowchart of a method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion provided by the present invention; Figure 2 Flowchart of another method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion provided by the present invention; Figure 3 Flowchart of another method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion provided by the present invention; Figure 4 The flowchart of the method for extracting specific degradation features under foggy conditions provided by the present invention is shown below; Figure 5 This is a flowchart of the method for obtaining the hazardous condition level provided by the present invention. Detailed Implementation

[0016] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0017] Example 1: This invention provides a method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion, such as... Figure 1 As shown, the method includes: Step 1: Extract the specific degradation features of cameras, millimeter-wave radar, and lidar in foggy conditions in parallel. Step 2: Based on the aforementioned features, the sensor confidence level is evaluated in real time and dynamically weighted and fused to obtain the comprehensive visibility estimate V. final =η*[Wcam *V vis +W rad *φ(C rv )+W lid *ψ(1 / λ)], and the hazardous condition level, where φ() and ψ() are mapping functions established based on calibration data, η is the global calibration coefficient, and W cam W rad W lid The dynamic confidence weights for cameras, millimeter-wave radar, and lidar are V, respectively. vis As the initial value for visual visibility estimation, C rv λ is the millimeter-wave radar-visual contradiction index, where λ is the attenuation coefficient. Step 3: Establish a precise mapping lookup table between the hazardous condition level and the vehicle control command, and execute the vehicle active safety control strategy corresponding to the hazardous condition level based on the precise mapping lookup table. Step 4: The comprehensive visibility estimate and the hazardous condition level are encapsulated into a standard vehicle-road cooperative message and broadcast through the vehicle communication unit to achieve beyond-line-of-sight vehicle-to-vehicle and vehicle-to-road early warning.

[0018] This invention provides a method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion. This method makes full use of the heterogeneous characteristics exhibited by vehicle-mounted multi-source sensors in foggy weather to achieve accurate, quantitative, and real-time identification of hazardous conditions in dense fog / patch fog. The identification results are seamlessly converted into hierarchical vehicle active safety control commands, thereby constructing a closed-loop safety enhancement system from perception to control.

[0019] Example 2: This invention provides a method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion, such as... Figure 2 As shown, the method includes: Step 0: Using the vehicle's high-precision time reference and high-speed vehicle network, hardware-level time synchronization is performed on camera image frames, millimeter-wave radar target list, and lidar point cloud. Through offline calibration parameters and online motion compensation algorithms, all perception data are unified into a dynamic coordinate system with the vehicle's center of mass as the origin.

[0020] The vehicle's high-precision time reference is the GNSS PPS signal. During hardware-level time synchronization, the timing deviation must be ensured to be less than 10 milliseconds.

[0021] Step 1: Extract the specific degradation features of cameras, millimeter-wave radar, and lidar in foggy conditions in parallel. like Figure 4 As shown, multi-dimensional fog-specific degradation features are extracted using sensors such as cameras, millimeter-wave radar, and lidar. Step 1 includes: Step 11: Based on the atmospheric scattering physical model, calculate the dark channel intensity and color saturation attenuation of the image as a physical estimate of atmospheric transmittance. Then, use a lightweight encoder network to extract the high-frequency detail loss coefficients and global texture entropy values ​​of the image, outputting a preliminary visual visibility estimate. Finally, form the visual degradation feature set F of the camera in foggy conditions. v ; The visual degradation feature set comprises a collection of multiple feature values ​​extracted from camera images that characterize the degree of degradation in foggy conditions, such as dark channel intensity, saturation decay, and texture entropy.

[0022] Step 12: After spatiotemporal alignment, perform association matching between radar targets and visually detected targets, and calculate the millimeter-wave radar-visual contradiction index C. rv Then, by monitoring the average signal-to-noise ratio change rate ΔSNR of the millimeter-wave radar receiving link, the penetration characteristic group F of the millimeter-wave radar in foggy conditions is finally formed. r ; The millimeter-wave radar-visual contradiction index is the percentage of targets that are stably tracked by the radar but completely undetectable by the visual system. This index is a direct and sensitive indicator of visual obstruction caused by fog, and is a feature group F. r The core feature of this method quantifies the proportion of targets that radar can "see" but vision cannot, making it one of the most direct indicators for diagnosing fog concentration. Monitoring the average signal-to-noise ratio (SNR) change rate ΔSNR of the millimeter-wave radar receiver link serves as supplementary evidence of millimeter-wave fog penetration attenuation. The millimeter-wave radar penetration feature set comprises a set of features extracted from millimeter-wave radar data that characterize its fog-penetrating ability and contradict visual perception.

[0023] Step 13: Divide the point cloud into distance segments, count the number of effective points in each segment, fit an exponential decay curve of point cloud density with distance, obtain the attenuation coefficient λ that characterizes the extinction properties of fog, and then analyze the variance σ of the reflection intensity distribution of point clouds of known high reflectivity objects. int Finally, the attenuation characteristic group F of the lidar in foggy conditions was obtained. l .

[0024] The point cloud is segmented by distance, the number of effective points in each segment is counted, and an exponential decay curve of the point cloud density with distance is fitted. The core parameter—the lidar point cloud density decay coefficient λ—is a feature set F. l The core features of fog are obtained by fitting the attenuation curve of point cloud with distance, directly characterizing the extinction ability and properties of fog. The variance σ of the reflection intensity distribution of point clouds of known high-reflectivity objects (such as lane line reflective paint) is analyzed. int The variance of the reflection intensity distribution is the characteristic set F lThe variance in the feature set represents the dispersion of the reflection intensity of a fixed object point cloud; fog droplet scattering increases this variance. The lidar attenuation feature set contains a set of features extracted from the lidar point cloud that characterize the degree of signal attenuation in fog.

[0025] Step 2: Based on the aforementioned features, the sensor confidence level is evaluated in real time and dynamically weighted and fused to obtain the comprehensive visibility estimate V. final =η*[W cam *V vis +W rad *φ(C rv )+W lid *ψ(1 / λ)], and the hazardous condition level, where φ() and ψ() are mapping functions established based on calibration data, η is the global calibration coefficient, and W cam W rad W lid The dynamic confidence weights for cameras, millimeter-wave radar, and lidar are V, respectively. vis As the initial value for visual visibility estimation, C rv λ is the millimeter-wave radar-visual contradiction index, where λ is the attenuation coefficient. This step utilizes a miniature network to synthesize the features of various current sensors (F... v F r F l Based on the vehicle state (v), the system dynamically calculates which sensor should be trusted and what weight (W) it should have. cam W rad W lid Then, these weights are compared with the most representative visibility cue (V) from each sensor. vis ,φ(C rv The visibility (V) is obtained by weighting and fusing ψ(1 / λ) and then passing it through a calibration coefficient (η). final Finally, combining visibility and key features, a classifier determines the specific danger level (L0-L3). The algorithm follows a progressive logic of "evaluation (weights) -> fusion (formula) -> decision (classification)". Figure 5 As shown, step 2 includes: Step 21, F v F r F l The vehicle status is input into a pre-set micro-evaluation network, which infers online the dynamic confidence weight W of each sensor at the current moment. cam W rad W lid Furthermore, the sum of the normalized weights of each dynamic confidence level is 1; Pre-design a miniature evaluation network (such as a two-layer fully connected neural network) to evaluate the real-time features F. vF r F l Taking vehicle status (vehicle speed, yaw rate) as input, the system infers the dynamic confidence weight W of each sensor at the current moment online. cam W rad W lid (Normalized to a sum of 1, i.e., W) cam +W rad +W lid =1). Vehicle speed is important contextual information; visibility requirements are even higher at high speeds, affecting risk assessment. Vehicle yaw rate characterizes the vehicle's steering state and is used to help determine whether sensor data is affected by its own motion.

[0026] Step 22: Calculate the final visibility V using a weighted fusion formula. final =η*[W cam *V vis +W rad *φ(C rv )+W lid *ψ(1 / λ)], where φ() and ψ() are mapping functions established based on calibration data, η is the global calibration coefficient, and W cam W rad W lid The dynamic confidence weights for cameras, millimeter-wave radar, and lidar are V, respectively. vis As the initial value for visual visibility estimation, C rv λ is the millimeter-wave radar-visual contradiction index, where λ is the attenuation coefficient. This model adaptively fuses visibility cues from various sensors based on real-time confidence levels. The initial visual visibility estimate is a core output of the feature set, representing a preliminary estimation of the visibility distance based solely on visual information. The radar inconsistency index mapping function φ() is a function that maps the radar-visual inconsistency index C... rv The function mapped to an equivalent visibility reference value, i.e., φ(C) rv The function is calibrated based on a large amount of historical data. The lidar attenuation coefficient mapping function ψ() is a function that maps the reciprocal of the attenuation coefficient λ, 1 / λ, to an equivalent visibility reference value, i.e., ψ(1 / λ), which is also established based on calibration data. The global calibration coefficient η is an adjustable scaling factor used to perform overall calibration on the final fusion result to eliminate system bias.

[0027] Step 23, fuse the feature vector [V] final C rv ,λ,σ int Input v into the preset classifier to obtain the danger level L0 / L1 / L2 / L3, where L0 is no risk, L1 is a mild warning, L2 is a severe warning, L3 is extremely dangerous, and v is the vehicle speed.

[0028] Construct a classifier (such as a gradient boosting tree GBDT) and input the fused feature vector [V] final C rv ,λ,σ int [,v], outputs four levels of hazardous conditions: L0 (no risk), L1 (mild warning), L2 (serious warning), and L3 (extreme danger / fog).

[0029] Step 3: Establish a precise mapping lookup table between the hazardous condition level and the vehicle control command, and execute the vehicle active safety control strategy corresponding to the hazardous condition level based on the precise mapping lookup table. Step 4: The comprehensive visibility estimate and the hazardous condition level are encapsulated into a standard vehicle-road cooperative message and broadcast through the vehicle communication unit to achieve beyond-line-of-sight vehicle-to-vehicle and vehicle-to-road early warning.

[0030] This invention provides a method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion. This method makes full use of the heterogeneous characteristics exhibited by vehicle-mounted multi-source sensors in foggy weather to achieve accurate, quantitative, and real-time identification of hazardous conditions in dense fog / patch fog. The identification results are seamlessly converted into hierarchical vehicle active safety control commands, thereby constructing a closed-loop safety enhancement system from perception to control.

[0031] Example 3: This invention provides a method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion, such as... Figure 3 As shown, the method includes: Step 1: Extract the specific degradation features of cameras, millimeter-wave radar, and lidar in foggy conditions in parallel. Step 2: Based on the aforementioned features, the sensor confidence level is evaluated in real time and dynamically weighted and fused to obtain the comprehensive visibility estimate V. final =η*[W cam *V vis +W rad *φ(C rv )+W lid *ψ(1 / λ)], and the hazardous condition level, where φ() and ψ() are mapping functions established based on calibration data, η is the global calibration coefficient, and W cam W rad W lid The dynamic confidence weights for cameras, millimeter-wave radar, and lidar are V, respectively. vis As the initial value for visual visibility estimation, C rv λ is the millimeter-wave radar-visual contradiction index, where λ is the attenuation coefficient. Step 3: Establish a precise mapping lookup table between the hazardous condition level and the vehicle control command, and execute the vehicle active safety control strategy corresponding to the hazardous condition level based on the precise mapping lookup table. The exact mapping lookup table includes: L1: Head-up display projects fog indicators; recommended speed ≤ 80km / h; adaptive cruise control increases following distance by 20%. L2: Triggering the audible and visual alarm, the system automatically limits the speed to 60km / h, sets the adaptive cruise control to the maximum safe distance, and forcibly turns on the front and rear fog lights and hazard warning lights; L3: Activate the highest level audio-visual alarm and request the driver to take immediate control. If the vehicle is in autonomous driving mode, execute the "minimum risk strategy" to smoothly decelerate and safely stop in the lane.

[0032] Step 4: The comprehensive visibility estimate and the hazardous condition level are encapsulated into a standard vehicle-road cooperative message and broadcast through the vehicle communication unit to achieve beyond-line-of-sight vehicle-to-vehicle and vehicle-to-road early warning.

[0033] Use {timestamp, latitude and longitude, V} final The hazard level L is encapsulated as a standard vehicle-road cooperative message (such as CAM / DENM) and broadcast through the on-board unit (OBU) to achieve beyond-line-of-sight vehicle-to-vehicle and vehicle-to-road early warning.

[0034] Step 5: Continuously collect actual response data after the vehicle executes the active safety control strategy, and use it as a feedback signal to perform online incremental learning on the preset micro evaluation network, so that the algorithm can continuously adapt to changes in individual vehicle characteristics and sensor performance.

[0035] The actual response data includes the actual deceleration and whether the Electronic Stability Program (ESP) is triggered.

[0036] This invention provides a method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion. This method makes full use of the heterogeneous characteristics exhibited by vehicle-mounted multi-source sensors in foggy weather to achieve accurate, quantitative, and real-time identification of hazardous conditions in dense fog / patch fog. The identification results are seamlessly converted into hierarchical vehicle active safety control commands, thereby constructing a closed-loop safety enhancement system from perception to control.

[0037] In summary, this invention provides a method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion, achieving a paradigm shift from "qualitative classification" to "quantitative safety assessment." This invention outputs precise real-time visibility values ​​(in meters) and differentiated risk levels, providing autonomous driving decision-making systems with quantitative inputs that can be directly used for planning and control, fundamentally surpassing the vague qualitative labels of existing technologies. It constructs a highly robust "sensor health diagnosis" perception paradigm: creatively utilizing sensor performance degradation itself as a feature, the system can still reliably judge based on "diagnostic reports" from sensors such as radar even when some sensors, such as cameras, are "blind," greatly enhancing the system's resilience and safety in extreme weather conditions. It forms a full-link active safety closed loop of "perception-decision-control-coordination": the identification results are directly and automatically converted into hierarchical vehicle control commands, combined with network-connected collaborative early warning, achieving collaborative protection from single-vehicle intelligence to collective intelligence, effectively preventing high-speed chain-reaction rear-end collisions. It endows the system with continuous self-optimization and self-adaptation capabilities: by introducing an online learning mechanism based on real driving feedback, the algorithm can evolve autonomously and adapt to different vehicle platforms, sensor configurations and regional climate conditions, significantly improving the practical value and life cycle of the technology.

[0038] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion, characterized in that, The method includes: Step 1: Extract the specific degradation features of cameras, millimeter-wave radar, and lidar in foggy conditions in parallel. Step 2: Based on the aforementioned features, the sensor confidence level is evaluated in real time and dynamically weighted and fused to obtain the comprehensive visibility estimate V. final =η*[W cam *V vis +W rad *φ(C rv )+W lid *ψ(1 / λ)], and the hazardous condition level, where φ() and ψ() are mapping functions established based on calibration data, η is the global calibration coefficient, and W cam W rad W lid The dynamic confidence weights for cameras, millimeter-wave radar, and lidar are V, respectively. vis As the initial value for visual visibility estimation, C rv λ is the millimeter-wave radar-visual contradiction index, where λ is the attenuation coefficient. Step 3: Establish a precise mapping lookup table between the hazardous condition level and the vehicle control command, and execute the vehicle active safety control strategy corresponding to the hazardous condition level based on the precise mapping lookup table. Step 4: The comprehensive visibility estimate and the hazardous condition level are encapsulated into a standard vehicle-road cooperative message and broadcast through the vehicle communication unit to achieve beyond-line-of-sight vehicle-to-vehicle and vehicle-to-road early warning.

2. The method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion according to claim 1, characterized in that, The method further includes: Step 0: Using the vehicle's high-precision time reference and high-speed vehicle network, hardware-level time synchronization is performed on camera image frames, millimeter-wave radar target list, and lidar point cloud. Through offline calibration parameters and online motion compensation algorithms, all perception data are unified into a dynamic coordinate system with the vehicle's center of mass as the origin.

3. The method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion according to claim 2, characterized in that, The vehicle's high-precision time reference is the GNSS PPS signal.

4. The method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion according to claim 2, characterized in that, Step 1 includes: Step 11: Based on the atmospheric scattering physical model, calculate the dark channel intensity and color saturation attenuation of the image as a physical estimate of atmospheric transmittance. Then, use a lightweight encoder network to extract the high-frequency detail loss coefficients and global texture entropy values ​​of the image, outputting a preliminary visual visibility estimate. Finally, form the visual degradation feature set F of the camera in foggy conditions. v ; Step 12: After spatiotemporal alignment, perform association matching between radar targets and visually detected targets, and calculate the millimeter-wave radar-visual contradiction index C. rv Then, by monitoring the average signal-to-noise ratio change rate ΔSNR of the millimeter-wave radar receiving link, the penetration characteristic group F of the millimeter-wave radar in foggy conditions is finally formed. r ; Step 13: Divide the point cloud into distance segments, count the number of effective points in each segment, fit an exponential decay curve of point cloud density with distance, obtain the attenuation coefficient λ that characterizes the extinction properties of fog, and then analyze the variance σ of the reflection intensity distribution of point clouds of known high reflectivity objects. int Finally, the attenuation characteristic group F of the lidar in foggy conditions was obtained. l .

5. The method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion according to claim 4, characterized in that, Step 2 includes: Step 21, F v F r F l The vehicle status is input into a pre-set micro-evaluation network, which infers online the dynamic confidence weight W of each sensor at the current moment. cam W rad W lid Furthermore, the sum of the normalized weights of each dynamic confidence level is 1; Step 22: Calculate the final visibility V using a weighted fusion formula. final =η*[W cam *V vis +W rad *φ(C rv )+W lid *ψ(1 / λ)], where φ() and ψ() are mapping functions established based on calibration data, η is the global calibration coefficient, and W cam W rad W lid The dynamic confidence weights for cameras, millimeter-wave radar, and lidar are V, respectively. vis As the initial value for visual visibility estimation, C rv λ is the millimeter-wave radar-visual contradiction index, where λ is the attenuation coefficient. Step 23, fuse the feature vector [V] final C rv ,λ,σ int Input v into the preset classifier to obtain the danger level L0 / L1 / L2 / L3, where L0 is no risk, L1 is a mild warning, L2 is a severe warning, L3 is extremely dangerous, and v is the vehicle speed.

6. The method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion according to claim 5, characterized in that, The WeChat evaluation network is a two-layer fully connected neural network, the vehicle state includes vehicle speed and yaw rate, and the classifier is a gradient boosting tree (GBDT).

7. The method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion according to claim 1, characterized in that, The exact mapping lookup table includes: L1: Head-up display projects fog indicators; recommended speed ≤ 80km / h; adaptive cruise control increases following distance by 20%. L2: Triggering the audible and visual alarm, the system automatically limits the speed to 60km / h, sets the adaptive cruise control to the maximum safe distance, and forcibly turns on the front and rear fog lights and hazard warning lights; L3: Activate the highest level audio-visual alarm, requesting the driver to take immediate control. If the vehicle is in autonomous driving mode, execute the minimum risk strategy, smoothly decelerate, and safely stop within the lane.

8. The method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion according to claim 1, characterized in that, The standard vehicle-road cooperative message also includes a timestamp and the vehicle's latitude and longitude.

9. The method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion according to claim 1, characterized in that, The method further includes: Step 5: Continuously collect actual response data after the vehicle executes the active safety control strategy, and use it as a feedback signal to perform online incremental learning on the preset micro evaluation network, so that the algorithm can continuously adapt to changes in individual vehicle characteristics and sensor performance.

10. The method for identifying and processing hazardous road conditions in foggy weather based on multi-sensor fusion according to claim 9, characterized in that, The actual response data includes the actual deceleration and whether the Electronic Stability Program (ESP) is triggered.

Images