A crosswind warning method and system based on machine vision

Abstract

The present invention relates to a method and system for warning crosswinds during driving based on machine vision, wherein the method comprises the following steps: collecting video wind field information through vehicle-mounted camera equipment, collecting wind speed characteristics of road scenes and performing fluid mechanics calculations, and extracting corresponding preset road scene wind field characteristics; performing wind speed analysis of the road wind field based on machine vision to obtain visually predicted joint wind field characteristics; and calculating the crosswind warning level for each vehicle based on the visually predicted joint wind field characteristics combined with the characteristics of the vehicle. Compared with the prior art, the present invention reduces the cost of crosswind warnings in road traffic, enhances the crosswind driving safety of large vehicles, and extracts key characteristics of the road wind field in combination with various factors of the road environment, thereby making up for the difficulty of predicting the road wind field and realizing lane-level real-time estimation of crosswinds.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent traffic warning, and in particular to a method and system for warning crosswinds based on the combined technology of fluid mechanics and machine vision. Background Art

[0002] As a common physical phenomenon, wind has always received widespread attention. Whether it is the quantification of wind energy resources, air pollution monitoring, or weather forecasting, it will rely on the rapid and accurate measurement of local wind conditions. In the field of traffic engineering, wind, especially crosswind, often has a greater impact on driving safety. Especially for trucks and truck convoys, crosswinds will generate aerodynamic forces on the wind pressure center of the car. Due to the difference between the wind pressure center and the center of gravity of the vehicle, the aerodynamic characteristics of the vehicle will change, and the vehicle is prone to deviation, side slip, roll, and rollover accidents. In view of the adverse effects of crosswind on driving safety, it is particularly important to collect wind speed information in advance in order to formulate driving safety measures that should be taken under specific crosswind speeds.

[0003] For example, Chinese patent applications CN115223367A and CN111599199A, in the past, wind information often needed to be obtained with the help of meteorological stations or large-scale installation of anemometers, which was time-consuming and labor-intensive, costly, inefficient, and difficult to maintain. Secondly, the traditional wind speed collection method can only collect the wind speed at a certain point, and cannot reflect the changes in the wind field of the road in real time and continuously. In addition, the side wind information collected in advance cannot be transmitted in advance to the subsequent trucks that are more susceptible to the side wind, making it difficult to achieve safety protection under side winds. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a crosswind warning method and system based on the combined technology of fluid mechanics and machine vision.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] As a first aspect of the present invention, a method for warning crosswind during driving based on machine vision is provided, the method comprising the following steps:

[0007] The video wind field information is collected through the vehicle-mounted camera equipment, the wind speed characteristics of the road scene are collected and the fluid dynamics calculation is performed to extract the corresponding preset road scene wind field characteristics;

[0008] The wind speed analysis of the road wind field based on machine vision can obtain the joint wind field characteristics predicted by vision;

[0009] Each vehicle calculates the crosswind warning level based on visual prediction, combined with wind field characteristics and the characteristics of the vehicle itself.

[0010] Furthermore, the step of collecting wind speed characteristics of the road scene and performing fluid mechanics calculation to extract the characteristics of the corresponding wind field includes: extracting the characteristics of the corresponding wind field by fluid mechanics simulation and extracting the characteristics of the corresponding wind field from the real environment. The specific process is as follows:

[0011] Obtain road environment characteristics for building numerical simulation models, including vegetation, weather, topography, and road width;

[0012] Use fluid dynamics software to establish a fluid dynamics simulation CFD lane scene numerical model for the selected road environment characteristics;

[0013] Based on the CFD lane scene numerical model, fluid dynamics simulation calculations are performed to extract CFD features of point features with special wind speeds and directions;

[0014] Collect real lane scene data, including environmental video data, environmental wind field characteristic data and lane wind field characteristic data;

[0015] The real lane scene data is used to reconstruct the real lane crosswind field, and the points extracted by CFD features are cross-validated to obtain the road scene wind field characteristics;

[0016] The above process is repeated to complete the wind speed feature collection and fluid dynamics calculation of complex road scenes in various situations.

[0017] Furthermore, the road wind field wind speed analysis based on machine vision includes the following steps:

[0018] Input the wind field information collected by the vehicle-mounted camera and the wind field characteristics of the corresponding road scene extracted;

[0019] Through semantic segmentation, the boundaries between lanes and the environment in the road driving environment video are segmented, the lane information is recorded, and the environment data is used for optical flow motion modeling;

[0020] Optical flow motion modeling is used to take environmental data as input, extract the position and posture of dynamic objects in time sequence, and estimate the dynamic characteristics of objects that reflect wind speed and direction;

[0021] Through machine learning video frame classification, the dynamic features of the object obtained by optical flow modeling are used as input for classification, and the wind speed and wind direction features are output and mapped to the wind speed classification table.

[0022] Furthermore, in the optical flow motion modeling, the influence of the vehicle motion on the flow direction and speed of objects in the video frame is eliminated based on the position and speed information of the vehicle.

[0023] Furthermore, the specific process of performing the machine learning video frame classification is as follows:

[0024] Input the dynamic features of the object that reflects wind speed and direction extracted based on optical flow motion modeling;

[0025] Use machine learning analysis technology to eliminate irrelevant features and use visual semantic segmentation technology to enhance relevant key features;

[0026] Through Transformer and long short-term memory artificial neural network LSTM, a classification model with the ability to recognize time series dynamic video frames is jointly built to output visually predicted wind speed.

[0027] Furthermore, the vehicle propagates the obtained visual prediction combined wind field features to subsequent vehicles, and the propagation forms of the visual prediction combined wind field features include:

[0028] One vehicle communicates directly with another vehicle; the leading vehicle directly transmits the warning level to the following vehicle with the same configuration. The calculation of the warning level is performed by the leading vehicle, and the following vehicle directly executes the preset plan under the received warning level;

[0029] One vehicle communicates directly with multiple vehicles; when the following vehicle is not configured the same as the preceding vehicle, the calculated wind field characteristics are transmitted to the following vehicle, which then calculates the warning level and warning plan on its own;

[0030] Multiple vehicles communicate based on roadside infrastructure.

[0031] As a second aspect of the present invention, a machine vision-based crosswind warning system is provided, characterized in that the system is used to implement any of the crosswind warning methods described above, including:

[0032] Vehicles equipped with on-board control systems and on-board camera equipment;

[0033] The video processing and calculation module obtains the video based on the vehicle-mounted camera equipment and calculates and obtains the joint wind field characteristics;

[0034] The crosswind warning calculation module calculates the crosswind warning level based on the combined wind field characteristics and the vehicle characteristics;

[0035] The vehicle processes the collected video wind field information based on the video processing calculation module and the crosswind warning calculation module and broadcasts the information to subsequent vehicles.

[0036] Furthermore, the video processing calculation module includes:

[0037] Video data interface, used to access the vehicle-mounted camera device to obtain video data;

[0038] Dynamic optical flow information capture module, which is used to extract key features reflecting wind speed and direction in video data based on preset road scene wind field characteristics and optical flow motion modeling;

[0039] The preset wind field feature module stores the preset road scene wind field features extracted from the real environment in combination with the fluid mechanics simulation;

[0040] The video processor,classifies the results of the dynamic optical flow information capture,module based on machine learning video frame classification, and obtains,the jointly calculated wind field characteristics.

[0041] Furthermore, the crosswind warning calculation module includes:

[0042] A preset scheme module stores preset crosswind levels and crosswind warning schemes;

[0043] The wind field characteristic processing module compares the joint wind field characteristics obtained by the video processing calculation module with the preset scheme in the preset scheme module to obtain the warning level and the warning scheme;

[0044] The in-vehicle internet-connected device connects the vehicle's various driving control subsystems to complete the crosswind warning solution.

[0045] Furthermore, the crosswind warning calculation module is arranged on the roadside so that the roadside infrastructure completes unified processing of warnings or is directly arranged on each vehicle so that each vehicle processes data by itself to issue warnings.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The technical solution provided by the present invention consists of fluid dynamics calculation, machine vision road wind field analysis and V2V-based joint driving crosswind warning. Visual wind speed measurement technology can significantly reduce the cost and time required to measure the wind field. Its great significance lies in:

[0048] 1) Solved the high cost problem of crosswind warning in road traffic: Based on the cheaper and widely popular driving recorder, visual wind speed measurement is adopted to reduce the deployment of anemometers and there is no additional hardware cost.

[0049] 2) Solved the problem of difficulty in estimating the road wind environment: The road environment wind field has its complex nature. Fluid mechanics calculation technology based on real scene data is adopted to extract the key characteristics of the road wind field in combination with various factors of the road environment, which makes up for the problem of difficulty in predicting the road wind field and realizes real-time estimation of lane-level crosswind.

[0050] 3) Enhanced crosswind driving safety for large vehicles: Vehicle-to-vehicle communication technology is used. When a car equipped with visual wind speed measurement capability goes first and detects strong crosswinds on the road ahead, the information is transmitted to the following vehicles through ITS intelligent communication of vehicle-to-vehicle communication or vehicle-road collaborative communication to achieve early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A system structure block diagram provided by some aspects of the present disclosure is shown, the system comprising a vehicle-mounted control system, a video processing module, and a crosswind warning module;

[0052] Figure 2 An example flow chart of a fluid dynamics calculation technology for wind speed feature acquisition of a road scene provided by some aspects of the present disclosure is shown;

[0053] Figure 3 An example flow chart of a technology for analyzing wind speed in a road wind field based on machine vision provided by some aspects of the present disclosure is shown;

[0054] Figure 4 An example flow chart of a crosswind warning technology for vehicles based on a joint technology provided by some aspects of the present disclosure is shown. DETAILED DESCRIPTION

[0055] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0056] With the rapid development of machine vision technology, combined with optical flow technology and computational fluid dynamics technology, it is possible to use machine learning to identify lane wind speed. Visual wind speed measurement technology has the potential to significantly reduce the cost and time required to measure wind fields. With the popularization of ITS technologies such as vehicle-road collaboration and vehicle networking, communication and interconnection between vehicles and between vehicles and roads can be achieved. Therefore, when a car equipped with visual wind speed measurement capability goes first and detects that there is a strong crosswind on the road ahead, the information is transmitted to the following vehicle through the intelligent communication of ITS (vehicle networking or vehicle-road collaboration), which can realize a relatively complete early warning system and provide sufficient data reference for subsequent vehicles to prevent crosswind hazards. The traditional method of installing anemometers to monitor a single location is usually costly, and even then only one measurement value can be provided. The only hardware cost of using the visual wind speed measurement method is the camera recording device, and a standard camera phone or dashcam can provide sufficient resolution. Dashcams are equipped on more vehicles, and more cameras are also equipped on some autonomous driving. This provides a basis for the application of this technology. As a result, this approach offers significantly lower costs per measurement point, enables continuous measurement, and eliminates the barriers of time and labor required to install an anemometer at a location.

[0057] The purpose of the present invention is to provide a crosswind warning method and system based on the joint technology of fluid mechanics and machine vision, so as to overcome the problem that the crosswind speed of the road environment cannot be estimated, predicted, and warned. The wind direction and wind speed of the road wind field are measured by the visual wind speed measurement technology through the vehicle's driving recorder, and the degree of influence of the wind field on the vehicle and subsequent vehicles and the corresponding measures to be taken are obtained with the support of V2I and V2V technologies. It can effectively realize the early warning of unfavorable wind speeds, and remind subsequent susceptible vehicles to take reasonable avoidance measures, which can effectively avoid the occurrence of dangerous accidents and improve traffic safety.

[0058] To achieve the above object, the solution provided by the present invention is as follows:

[0059] Fluid mechanics calculations are performed based on the collection of wind speed characteristics of road scenes, and fluid mechanics analysis scenarios are built for multiple working conditions considering factors such as vegetation, topography, and weather in the road environment. A fluid mechanics numerical analysis model is established for lane scenarios that consider the number and width of lanes. Data is collected on wind field characteristics in the road environment. Environmental video information is collected on the road environment, especially the surrounding environment, under the influence of crosswinds. The actual collected data is cross-validated with the constructed numerical analysis scenarios to extract the characteristics of the road scene wind field.

[0060] The machine vision-based analysis of wind speed in a road wind field adopts a driving recorder and an information storage and processing module, wherein at least one driving recorder operates normally, and at least one information storage and processing module can execute the instruction processing of the driving recorder video data, and the instruction output is calculated by combining the preset road scene wind field characteristics; the driving lane wind field characteristics extracted by the above method are combined with the optical flow motion modeling to extract the possible key features in the video data that can effectively reflect the wind speed and wind direction; machine learning analysis technology is adopted to eliminate non-related features, and visual semantic segmentation technology is used to enhance the possible key features; a classification model with the ability to recognize dynamic video frames in time series is jointly built through Transformer and long short-term memory artificial neural network LSTM; and the visually predicted wind speed is output according to the input video of the vehicle driving recorder.

[0061] A crosswind warning method for vehicles based on joint technology includes a vehicle-to-vehicle (road) communication module and a crosswind warning calculation module; the vehicle-to-vehicle communication module broadcasts or unicasts the visual predicted wind speed of the preceding vehicle backward; the warning level calculation module equipped on each vehicle calculates the warning level and gives an avoidance plan based on the characteristics of the vehicle after obtaining the visual predicted wind speed of the preceding vehicle.

[0062] Example 1

[0063] As one implementation of the present invention, this embodiment provides a crosswind warning method and system based on the combined technology of fluid mechanics and machine vision. The warning system structure is as follows:

[0064] refer to Figure 1 The crosswind warning system for ground road lanes disclosed in this embodiment includes a vehicle control system 101, a video processing calculation module 103 and a crosswind warning calculation module 102. The main working process of the system can be described as follows: Figure 1 The car 125 collects the video wind field information, uploads it to the roadside cloud storage 126, and broadcasts 128 the information to subsequent vehicles 129 through information processing 127.

[0065] refer to Figure 1 , which shows a composition diagram of an onboard control system 101. A vehicle dynamics control module 104, a mobile device GPS module 108, and a mobile network 5G interface 109 for a vehicle-to-vehicle communication module 106 and a vehicle-to-road communication module 105 can be implemented in the onboard control system 101. The video processing module 107 reads the lane video data of the data storage module 111 through the processing module 110 interface, transmits it to the data processing module 113 through the video processing 103, and connects the information to the crosswind warning module 112.

[0066] Continue to refer Figure 1 , the video processing module 107 implements the video processing calculation module 103. The crosswind warning calculation module 102 is implemented in the crosswind warning module 109. The video processing calculation module 103 transmits data to the video processor 117 through the processing module interface 110 by accessing the video data interface 114, capturing the information of the dynamic optical flow 115, combining with the preset wind field feature module 116, and then obtains the output 118 of the wind field feature. The wind field feature output 118 transmits the data to the crosswind warning calculation module 102 through the processing module common interface 110 of the crosswind warning module 112. The module 119 for taking the preset crosswind level and warning scheme and the wind field feature processing module 120 are implemented in the crosswind warning calculation module 102. The driver interface subsystem 122, the vehicle sensing subsystem 123 and the vehicle driving subsystem 124 are connected through the in-vehicle networkable device to realize the crosswind warning.

[0067] The driver interface subsystem 122 depends on the output result of the crosswind warning calculation module 102 after the wind field characteristic output 118. The vehicle sensing subsystem 123 and the vehicle driving subsystem 124 depend on the adjustment signal obtained by the driver interface subsystem 122. The driver interface subsystem 122, the vehicle sensing subsystem 123, and the vehicle driving subsystem 124 all control the vehicle dynamics control at the vehicle control system level to prevent crosswinds after the crosswind warning calculation module obtains the wind field warning level and the measures to be taken.

[0068] like Figure 1As shown, the data storage module 111 and the processing and computing module 103 involved are currently available portable vehicle terminals, such as smart phones, smart tablet computing devices, laptop computers and other devices with storage and computing ports. These devices can be used to implement a video data interface 114, capture dynamic optical flow features 115 through a video processor 117, and load a preset stored feature module 116 to achieve the output 118 of the calculated wind field.

[0069] like Figure 1 As shown, the mobile network 5G interface is used to implement the vehicle-to-vehicle communication module 106 and the vehicle-to-road communication module 105. Current communication protocols are readily available, such as point-to-point communication, paging networks, telephone networks, WiFi networks, and other similar networks. The system network of the present application can also be used with other communication devices configured in other vehicle-mounted systems to ensure that other communications are not affected.

[0070] like Figure 1 As shown, the driver interface subsystem 122, the vehicle sensing subsystem 123 and the vehicle driving subsystem 124 can be used together with other vehicle subsystems in the same communication mode. These vehicle subsystems include the control systems of trucks, cars, engineering vehicles, and other large mobile mechanical equipment, but do not include water transportation equipment and systems.

[0071] Example 3

[0072] As another implementation of the present invention, this embodiment provides an implementation example of a crosswind warning method for driving based on fluid mechanics and machine vision, and the main steps of the method are as follows:

[0073] Wind speed feature acquisition and fluid dynamics calculation example for road scenes

[0074] This example provides a joint feature extraction solution for machine vision to more accurately identify road wind field features, relying on fluid mechanics simulation feature extraction 203 and extracting corresponding wind field features from the real environment 204.

[0075] refer to Figure 2 In order to establish a numerical simulation model, road environment features 201 are first considered, which include vegetation, weather, topography and road width. Fluid mechanics software is used to perform a CFD (fluid mechanics simulation) lane scene numerical model establishment 202 on the selected road environment features 201. Software that can be used for fluid mechanics calculations include but are not limited to XFlow and Fluent, which are all mature fluid mechanics calculation software.

[0076] After the CFD lane scene numerical model is established 202, fluid mechanics simulation calculations are performed, and CFD feature extraction is performed on the point features where special wind speeds and wind directions appear 203. For example, in some cases, there may be sudden changes in wind speed and turbulence in the road environment. In this case, the wind speed collected on the roadside will not truly reflect the impact of wind speed on the lane. Therefore, it is necessary to extract the features of special points in the CFD simulation stage to guide the real lane scene data collection 204.

[0077] Continue to refer Figure 2 , the real lane scene data collection 204 includes environmental video data collection, environmental wind field characteristic data collection, and lane wind field characteristic data collection. Exemplarily, environmental video data collection is completed by a camera device, including but not limited to a Gopro-type sports camera, a smart phone, and a driving recorder, as long as a suitable resolution can be provided. Environmental wind field characteristic data collection is collected by an anemometer in the surrounding environment of the side road, and it is necessary to collect wind speed and wind direction data and record the collection location. Lane wind field characteristic data collection is collected by an anemometer on the lane, and it is necessary to collect wind speed and wind direction data and record the collection location.

[0078] Continue to refer Figure 2 , through real lane scene data collection 204, use these data to reconstruct the real lane crosswind wind field, and extract the real features 205 at the points of CFD feature extraction 203. Then generate preset road scene wind field features 206 to provide preset data of the joint technology for the information storage and processing module 302 of the machine vision processing module 306. The above is an example of fluid mechanics technology for wind speed feature collection of a road scene. Repeating the above process can complete wind speed feature collection and fluid mechanics calculation of complex road scenes in various situations, and store them in the preset feature database 305.

[0079] An example of wind speed analysis in a road wind field based on machine vision

[0080] Figure 3 The example flow chart of the machine vision-based road wind speed analysis technology provided by various aspects of the present disclosure is shown. The technology consists of a driving recorder module 301, an information storage and processing module 302 and a joint calculation of wind field characteristics 303.

[0081] like Figure 3As shown, the dashcam module 301 captures the video data of the driving environment and stores the data in the video and wind field database 304 of the information storage and processing module 302. The preset feature database 305 combines the video and wind field database 304 to transmit the data to the machine vision processing module 306. Through the semantic segmentation 307 of the scene in the road environment, the environmental features around the roadway such as the roadside trees are selected; the optical flow motion modeling 308 is used to extract the changes over time, such as the changes in the movement posture of the leaves to estimate the posture and reflect the flow of the wind; the results of the optical flow modeling are classified by the machine learning video frame classification 309 technology, mapped to the wind speed classification table, and the jointly calculated wind field features 303 are obtained. The video and wind field database 304 will also store the jointly calculated wind field features 303 to further improve the processing accuracy of the machine vision. Since it is necessary to realize vehicle-to-vehicle communication to realize the early warning of the crosswind, the preset feature database 305 is an online access database, which is convenient for all vehicles participating in the communication to read the constantly updated preset features and improve the accuracy of the visual wind speed prediction.

[0082] refer to Figure 3 , the machine vision processing module 306 includes three major categories of processing technologies, semantic segmentation 307, optical flow motion modeling 308 and machine learning video frame classification 309 technology. These methods define the scope of the processing technology used, but do not limit the types of processing technologies. There are multiple processing methods under each processing technology, and different methods can be formed by modifying parameters. As long as the appropriate input and output are met, they can be implemented by this solution.

[0083] To illustrate the purpose and input and output of different methods, for example, the semantic segmentation 307 technology needs to segment the boundaries between lanes and the environment in the road driving environment video, record lane information (number of lanes, number of vehicles and arrangement) and transmit the data of the environment part to the next process, optical flow motion modeling 308; optical flow motion modeling 308 takes the environment data as input, extracts the position and posture of dynamic objects according to the time sequence of video frames, and estimates the direction and speed of object flow (such as leaves, branches and grass). Since it is the video data of a moving car mobile phone, the influence of car movement on the flow direction and speed of objects in the video frame needs to be eliminated in the optical flow motion modeling 308. This requires the dashcam module 301 to collect the position and speed information of the vehicle. The machine learning video frame classification 309 technology takes the dynamic features processed in the previous step as input and outputs wind speed and wind direction features.

[0084] Continue to refer Figure 3In this technology, the jointly calculated wind field features 303 that need to be obtained include wind speed features, lane features, wind direction features, geographic location, time series features, and vehicle distribution on the lane. The wind speed features, wind direction features, and time series features are output by machine learning video frame classification 309, and the lane features and vehicle distribution are output by semantic segmentation 307.

[0085] Example of crosswind warning based on joint technology

[0086] Figure 4 FIG. 2 shows an example flow chart of a lane warning technology based on a combined technology disclosed in this embodiment. Figure 4 As shown, in the early warning stage, the technology mainly consists of a vehicle-to-vehicle (road) communication module 401 and a crosswind early warning calculation module 402. The vehicle-to-road communication module 105 and the vehicle-to-vehicle communication module 106 in the vehicle-mounted control system 101 are the same modules as the vehicle-to-road communication module 406 and the vehicle-to-vehicle communication module 407 in the vehicle-to-vehicle (road) communication module 401 in this example. The communication method is described in this example.

[0087] refer to Figure 4 , when the joint calculation wind field feature 403 (the feature comes from Figure 3 After the wind field characteristics are jointly calculated 303), the characteristics are compared with the crosswind warning scheme in the preset scheme database 404 to obtain the warning level 405. The information is broadcast to other vehicles under the vehicle (road) communication module 401 or the communication method is completed through the infrastructure 408. Then the dynamic control of the subsequent vehicle is performed 409.

[0088] Continue to refer Figure 4 , car 1, car 2 and car 3 collected visual wind speed information in the previous driving, and completed the one-to-one backward information unicast (such as the black line connecting the cars) through the car-to-car communication in the communication mode 408. Or the dotted line communication mode from car 1 to car 4, car 5 and car 6 completes the broadcast mechanism. When the communication involves a large range, when car 1, car 2 and car 3 cannot communicate effectively with car 4, car 5 and car 6 within the car-to-car communication range, the communication can be completed with the help of roadside infrastructure 410.

[0089] The crosswind warning calculation module 402 can exist on the roadside, and the roadside infrastructure completes unified processing, or exists directly on each vehicle, and each vehicle processes the data by itself. When the crosswind warning calculation module is installed on the roadside, the communication method combined with the infrastructure in the communication method 408 is used at this time. When the crosswind warning calculation module is installed on the vehicle, the communication is completed in accordance with the vehicle-to-vehicle communication method of the communication method 408. At this time, the information transmission can be a joint calculation of the wind field characteristics 303 or the crosswind warning level 405. When the information is transmitted according to the joint calculation of the wind field characteristics 303, the warning level 405 is calculated locally by each vehicle. When the warning level 405 is directly transmitted by vehicle-to-vehicle communication, the warning level 405 is calculated by the front vehicle and then directly transmitted to the rear vehicle to realize the warning. The warning level 405 depends on the wind speed level designed by the preset scheme database 404. Therefore, the preset scheme database 404 has a networking function to facilitate all vehicles to access the preset warning scheme data.

[0090] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A machine vision-based crosswind warning method, characterized in that: The method steps include: The video wind field information is collected through the vehicle-mounted camera equipment, the wind speed characteristics of the road scene are collected and the fluid dynamics calculation is performed to extract the corresponding preset road scene wind field characteristics. The specific process is as follows: Obtain road environment characteristics for building numerical simulation models, including vegetation, weather, topography, and road width; Use fluid dynamics software to establish a CFD fluid dynamics simulation lane scene numerical model for the selected road environment characteristics; Based on the CFD lane scene numerical model, fluid dynamics simulation calculations are performed to extract CFD features of point features with special wind speeds and directions; Collect real lane scene data, including environmental video data, environmental wind field characteristic data and lane wind field characteristic data; The real lane scene data is used to reconstruct the real lane crosswind field, and the points extracted by CFD features are cross-validated to obtain the road scene wind field characteristics; Repeat the above process to complete the wind speed feature collection and fluid dynamics calculation of complex road scenes in various situations; Based on the analysis of the wind speed in the road wind field by machine vision, the joint wind field characteristics predicted by vision are obtained. The specific process is as follows: Input the wind field information collected by the vehicle-mounted camera and the wind field characteristics of the corresponding road scene extracted; Through semantic segmentation, the boundaries between lanes and the environment in the road driving environment video are segmented, the lane information is recorded, and the environment data is used for optical flow motion modeling; Optical flow motion modeling is used to take environmental data as input, extract the position and posture of dynamic objects in time sequence, and estimate the dynamic characteristics of objects that reflect wind speed and direction; Through machine learning video frame classification, the object dynamic features obtained by optical flow modeling are used as input for classification, and wind speed and wind direction features are output and mapped to the wind speed classification table. The specific process of machine learning video frame classification is as follows: Input the dynamic features of the object that reflects wind speed and direction extracted based on optical flow motion modeling; Use machine learning analysis technology to eliminate irrelevant features and use visual semantic segmentation technology to enhance relevant key features; Through Transformer and long short-term memory artificial neural network LSTM, a classification model capable of identifying time series dynamic video frames is jointly built to output visually predicted wind speed; Each vehicle calculates the crosswind warning level based on visual prediction, combined with wind field characteristics and the characteristics of the vehicle itself.

2. The machine vision-based crosswind warning method for vehicles according to claim 1 is characterized in that: In the optical flow motion modeling, the influence of the vehicle motion on the flow direction and speed of objects in the video frame is eliminated based on the position and speed information of the vehicle.

3. The machine vision-based crosswind warning method for vehicles according to claim 1 is characterized in that: The vehicle propagates the obtained visual prediction combined wind field features to subsequent vehicles, and the propagation forms of the visual prediction combined wind field features include: One vehicle communicates directly with another vehicle; the leading vehicle directly transmits the warning level to the following vehicle with the same configuration. The calculation of the warning level is performed by the leading vehicle, and the following vehicle directly executes the preset plan under the received warning level; One vehicle communicates directly with multiple vehicles; when the following vehicle is not configured the same as the preceding vehicle, the calculated wind field characteristics are transmitted to the following vehicle, which then calculates the warning level and warning plan on its own; Multiple vehicles communicate based on roadside infrastructure.

4. A crosswind warning system based on machine vision, characterized in that: The system is used to implement the crosswind warning method according to any one of claims 1 to 3, including: A vehicle equipped with an on-board control system (101) and an on-board camera device; A video processing and calculation module (103) is used to obtain the video based on the vehicle-mounted camera device and calculate and obtain the joint wind field characteristics; A crosswind warning calculation module (102) calculates a crosswind warning level based on the combined wind field characteristics and the vehicle characteristics; The vehicle processes the collected video wind field information based on the video processing calculation module (103) and the crosswind warning calculation module (102), and broadcasts the information to subsequent vehicles.

5. The machine vision-based crosswind warning system for vehicles according to claim 4 is characterized in that: The video processing calculation module (103) comprises: A video data interface (114), used to access and obtain video data from a vehicle-mounted camera device; A dynamic optical flow information capture module (115) is used to extract key features reflecting wind speed and wind direction in video data based on preset road scene wind field features and optical flow motion modeling; A preset wind field feature module (116) stores corresponding preset road scene wind field features extracted from a real environment in combination with fluid mechanics simulation; The video processor (117) classifies the results of the dynamic optical flow information capture module (115) based on machine learning video frame classification, and obtains jointly calculated wind field features.

6. The machine vision-based crosswind warning system for vehicles according to claim 4, characterized in that: The crosswind warning calculation module (102) comprises: A preset scheme module (119) stores preset crosswind levels and crosswind warning schemes; A wind field characteristic processing module (120) compares the combined wind field characteristics obtained by the video processing calculation module (103) with the preset scheme in the preset scheme module (119) to obtain a warning level and a warning scheme; The in-vehicle networkable device (121) is connected to various driving control subsystems of the vehicle to complete the crosswind warning solution.

7. The machine vision-based crosswind warning system for vehicles according to claim 6, characterized in that: The crosswind warning calculation module (102) is arranged on the roadside so that the roadside infrastructure can complete unified processing and warning, or is directly arranged on each vehicle so that each vehicle can process data and issue a warning on its own.

Images