Agglomerate fog state detection method and equipment based on road region division

Abstract

The invention provides an agglomerate fog state detection method and equipment based on road region division, and belongs to the technical field of intelligent traffic environment detection.The method comprises the steps that a continuous aerial image sequence obtained by an unmanned aerial vehicle through cruising along a road is collected, and a timestamp corresponding to each frame of aerial image and space pose data during shooting of the unmanned aerial vehicle are recorded; dividing each frame of aerial image into a plurality of sub-images along the road extension direction; extracting static visual features and dynamic visual features of each sub-image in parallel; by fusing the static visual features and the dynamic visual features, calculating the posterior probability that the agglomerate fog exists in the sub-image, and judging the agglomerate fog state grade according to the posterior probability; and binding the agglomerate fog state grade with the corresponding timestamp and spatial pose data to generate agglomerate fog early warning information. Based on the method, the invention further provides agglomerate fog state detection equipment based on road region division. According to the method, the sub-images are divided along the perspective direction of the road and are independently analyzed, so that the accurate positioning and grade judgment of the agglomerate fog at the road section level are realized.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation environment detection technology, and specifically relates to a method and device for detecting fog conditions based on road area division. Background Technology

[0002] Fog patches are suspended water droplets formed by the condensation of water vapor in the air. They typically occur under conditions of high humidity and low temperature, especially in the morning or at night. When fog patches appear, visibility is significantly reduced, posing a great threat to traffic safety. Particularly on highways, mountain roads, and other areas prone to fog patches, frequent occurrences not only lead to numerous traffic accidents but can also cause significant loss of life and property damage. Therefore, real-time detection of fog patches and the provision of scientific early warnings have become an urgent need for traffic safety management. Traditional fog patch detection based on high-altitude remote sensing or roadside video primarily employs static spectral analysis methods, which are unsuitable for rapid fog patch identification from the fast-moving perspective of low-altitude drones. With the rapid development of drone technology, combined with image technologies (such as visible light imaging and infrared imaging), drone-based fog patch monitoring methods have become a new and effective solution. Drones can fly in real time, using high-resolution cameras to collect road or aerial images, and combining these with image processing algorithms to accurately detect fog patches.

[0003] Currently, publicly available implementation methods include: one is based on fixed roadside video surveillance, and the other is based on dedicated sensors such as meteorological visibility meters. However, these two methods have the following drawbacks: fixed monitoring equipment has blind spots, making it impossible to conduct flexible and rapid inspections of the entire road area, and it is difficult to cope with the sudden and moving characteristics of fog patches. Traditional video methods mostly rely on dark channel priors, detection of the farthest visible point, or grayscale statistics of specific markers (such as streetlights). These methods experience a sharp decline in stability and a high false alarm rate in aerial photography scenarios with fast-moving drones, changing perspectives, and complex background textures. Moreover, existing methods usually output a global visibility value or a single fog condition judgment, and cannot provide the specific location (which section of the road), spatial range, and intensity level of the fog patches on the road, making it difficult to provide direct evidence for refined traffic guidance measures such as segmented speed limits and variable message sign displays. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a method and device for detecting patchy fog based on road area division. By dividing the image into sub-images along the road perspective direction and analyzing them independently, precise location and severity determination of patchy fog at the road segment level are achieved. The generated warning information includes multi-dimensional labels such as time, location, and severity, which can directly serve traffic guidance for different road segments.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A method for detecting patchy fog based on road area division includes the following steps: Collect a continuous sequence of aerial images taken by a drone while it is cruising along a road, and simultaneously record the timestamp corresponding to each frame of the aerial image and the spatial pose data of the drone when it takes each frame of the aerial image. Based on the principle of road perspective, each frame of aerial image is divided into multiple sub-images along the road extension direction; within each sub-image, static visual features characterizing image texture clarity and dynamic visual features characterizing pixel motion patterns are extracted in parallel. By fusing static and dynamic visual features extracted from the same sub-image, the posterior probability of the presence of fog in the sub-image is calculated, and the fog state level of the sub-image is determined based on the posterior probability. By binding the fog state level of each sub-image with the corresponding timestamp and spatial pose data, a structured segmented fog warning information is generated.

[0006] This invention also proposes a fog state detection device based on road area division, including at least one processor and a memory. The memory stores a computer program, which, when executed by the at least one processor, implements any one of the fog state detection methods based on road area division.

[0007] The effects described in the invention are merely those of the embodiments, and not all the effects of the invention. One of the above technical solutions has the following advantages or beneficial effects: This invention proposes a method and device for detecting patchy fog based on road area division, belonging to the field of intelligent transportation environment detection technology. The method includes the following steps: acquiring a continuous sequence of aerial images obtained by a drone cruising along a road, simultaneously recording the timestamp corresponding to each frame of the aerial image and the spatial pose data of the drone when capturing each frame; based on the road perspective principle, dividing each frame of the aerial image into multiple sub-images along the road's extension direction; within each sub-image, extracting static visual features characterizing image texture clarity and dynamic visual features characterizing pixel motion patterns in parallel; fusing the static and dynamic visual features extracted within the same sub-image to calculate the posterior probability of patchy fog existing in the sub-image, and determining the patchy fog state level of the sub-image based on the posterior probability; binding the patchy fog state level of each sub-image with the corresponding timestamp and spatial pose data to generate structured, segmented patchy fog warning information. Based on this method, this invention also proposes a patchy fog state detection device based on road area division. This invention achieves precise location and level determination of fog at the road segment level by dividing the image into sub-images along the road perspective direction and analyzing them independently. The generated warning information includes multi-dimensional labels such as time, location, and level, which can directly serve traffic guidance for different road segments.

[0008] This invention innovatively integrates SURF static texture features and optical flow dynamic motion features in parallel. The two types of features are complementary in dealing with challenges such as changes in the drone's perspective and low background texture: when image blur causes texture features to fail, the special optical flow pattern caused by fog can still provide valid evidence, thereby significantly improving the detection reliability in complex scenes.

[0009] The SURF and optical flow algorithms used in this invention are both efficient and classic algorithms. Combined with the local sub-image processing strategy, the overall computational load is controllable, which can meet the real-time requirements of UAV onboard computing units or edge servers and facilitates practical deployment.

[0010] In this invention, the division of road sub-images can be combined with the attitude angle of the UAV for adaptive perspective correction, and the entire process does not require manual specification of the farthest point or reliance on external markers such as streetlights, achieving fully automatic detection, and can work day and night and under various weather conditions. Attached Figure Description

[0011] Figure 1 This is a flowchart of a fog state detection method based on road area division proposed in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the road ROI division proposed in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the extraction results of the SURF algorithm proposed in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the optical flow algorithm extraction results proposed in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of a fog state detection device based on road area division proposed in Embodiment 1 of the present invention. Detailed Implementation

[0012] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure of the invention, components and arrangements of specific examples are described below. Furthermore, reference numerals and / or letters may be repeated in different examples. This repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. Descriptions of well-known components, processing techniques, and processes are omitted in this invention to avoid unnecessarily limiting the invention.

[0013] Example 1 Embodiment 1 of this invention proposes a fog state detection method based on road area division, which is used to overcome the problems of poor adaptability, ambiguous positioning, high false alarm rate and difficulty in real-time processing of existing technologies for road fog detection under UAV aerial photography perspective.

[0014] Figure 1 This is a flowchart of a fog state detection method based on road area division proposed in Embodiment 1 of the present invention; In step S100, a continuous sequence of aerial images acquired by the UAV while cruising along the road is collected, and the timestamp corresponding to each frame of aerial image and the spatial pose data of the UAV when capturing each frame of aerial image are recorded simultaneously. This step is implemented by equipping a high-definition visible light camera and a positioning module on the drone platform to acquire aerial videos and images covering the road area in real time during flight. At the same time, auxiliary information such as timestamps, latitude and longitude coordinates, flight altitude and attitude angles corresponding to the image frames are recorded simultaneously. The camera resolution and frame rate can be flexibly set according to mission requirements to ensure that clear road images can be obtained under different weather conditions. The positioning module, combined with the global satellite navigation system and inertial measurement unit, realizes the accurate measurement of flight attitude, thereby ensuring that each image frame corresponds to the actual geographical location.

[0015] In this way, the implementation of this application can flexibly cover road areas without relying on fixed ground sensors, forming a sequence of aerial images with spatiotemporal annotations, providing a reliable data foundation for subsequent road area division and fog detection.

[0016] In step S110, based on the principle of road perspective, each frame of aerial image is divided into multiple sub-images along the road extension direction.

[0017] The specific process includes: calculating the vertex pixel coordinates of each trapezoidal sub-image based on the width, height, and number of pre-divisions of the aerial image; wherein the upper base of the trapezoid corresponds to the far end of the road in the image, the lower base corresponds to the near end of the road, and the waistline of the trapezoid follows the same trend as the vanishing perspective line of the road edge.

[0018] Fog patches are characterized by their small spatial scale and localized distribution, while UAVs have a wide field of view during patrols. By defining a key area of ​​interest (ROI) for fog patch detection, it is beneficial to accurately locate the position of the fog patch and reduce computational overhead. Figure 2 This is a schematic diagram of the road ROI division proposed in Embodiment 1 of the present invention.

[0019] The aerial images are divided into There are three trapezoidal sub-images of equal height. Each sub-image represents a different monitoring range, assuming the image resolution is... The formula for the vertex pixel coordinates of each trapezoidal sub-image is: The coordinates of the top left vertex are: ; The coordinates of the top right vertex are: ; The coordinates of the bottom left vertex are: ; The coordinates of the bottom right vertex are: ; in, The number of sub-images; ; This indicates the horizontal pixel position of the top-left corner of the trapezoid. This indicates the horizontal pixel position of the bottom left corner base point of the trapezoid. This indicates the horizontal pixel position of the top right corner base point of the trapezoid; This indicates the horizontal pixel position of the bottom right corner base point of the trapezoid; Indicates the width of the sub-image; Indicates the height of the sub-image.

[0020] When calculating the vertex coordinates of each trapezoidal region, the pitch angle information acquired in real time by the UAV is further combined to adaptively adjust the width of the upper base of the trapezoid to maintain a stable mapping relationship between the key areas of interest and the actual road sections.

[0021] In step S120, static visual features characterizing the image texture sharpness are extracted.

[0022] In this application, the SURF algorithm is used to detect and describe feature points within a sub-image, and the number of detected feature points or surface density is used as static visual features; the specific process is as follows: For any position within the subimage Calculate its integral image The integral image is represented as: , in, Pixel intensity; At multiple scales Below, the box filter response is calculated using the integral image to approximate the second Gaussian derivative of the image. , , Construct the Hessian matrix for each pixel:

[0023] By calculating the determinant of the Hessian matrix Furthermore, non-maximum suppression is performed in both scale space and image space to locate feature points within the sub-image; For each feature point, calculate the Haar wavelet response within a circular neighborhood; calculate the principal direction of the feature point. Specifically: ; and These represent the Haar wavelet transforms in the horizontal and vertical directions, respectively. It is the inverse tangent function in the four quadrants; To construct the SURF descriptor, the square region centered on each keypoint is divided into a 4×4 sub-region grid, and the Haar wavelet response is calculated and summed. and its absolute value This forms a 64-dimensional descriptor vector as feature points.

[0024] The total number of all feature points within a sub-image. and will or its area with sub-image ratio Static visual features characterizing image texture sharpness .

[0025] In step S130, dynamic visual features characterizing pixel motion patterns are extracted.

[0026] This application calculates the motion vectors of pixels in a sub-image across consecutive frames based on optical flow, and uses the average magnitude and / or orientation consistency index of the motion vectors as dynamic visual features; the specific process is as follows: Based on two consecutive frames of images and Solving the optical flow constraint equations ; Obtain the optical flow vector of each pixel within the sub-image ; The horizontal component representing the optical flow vector; Represents the vertical component of the optical flow vector; is the partial derivative of the subimage in the horizontal direction; This is the partial derivative of the subimage in the vertical direction; For sub-images in time gradient on; Based on the optical flow vector, the dynamic visual feature value of the sub-image is calculated. The dynamic visual feature value includes the average amplitude of the optical flow vector and / or the directional consistency index of the optical flow vector. The formula for calculating the average amplitude is as follows: ; The total number of pixels within the sub-image.

[0027] This application establishes an overdetermined system of equations by using multiple adjacent pixels, which can then be used to solve the equations. The solution to the optical flow constraint equation can be further transformed into a least squares problem.

[0028] ; Referring to previously determined SURF keypoints across consecutive frames, optical flow vectors are calculated to capture the displacement and directionality of characteristic motions. In foggy regions, atmospheric particles typically exhibit subtle and coherent motions, resulting in small optical flow vectors of consistent size and orientation. In contrast, non-foggy regions tend to exhibit irregular and diverse motion characteristics.

[0029] In step S140, static visual features and dynamic visual features extracted from the same sub-image are fused together to calculate the posterior probability of the presence of fog in the sub-image, and the fog state level of the sub-image is determined based on the posterior probability. First, a spatial probability distribution model of the feature points identified by the SURF algorithm is established, specifically as follows: ; in, Represents a spatial probability distribution model; This represents the number of feature points or feature density extracted by the SURF algorithm; Indicates the mean of spatial distribution; It represents the standard deviation of spatial distribution; it is used to measure the centrality and dispersion of the distribution of haze feature points in different image regions.

[0030] Motion vector characteristics obtained by optical flow analysis ; in, Indicates the characteristics of the running vector; This represents the average motion vector or motion intensity extracted from optical flow analysis; This represents the mean of the motion vector; This represents the standard deviation of the motion vector. and Similarly, by statistically analyzing historical videos or a large amount of experimental data, the typical velocity distribution of haze movement, diffusion, or dissipation between different frames is mainly characterized.

[0031] Finally, the posterior probability of the presence of fog in the sub-image is calculated using a Bayesian fusion model: ; in, The prior probability of the occurrence of fog can be given by historical statistics or expert experience; It represents the marginal probability of all possible combinations of features and is usually used as a normalization constant in actual calculations; This represents the initial probability that a patch of fog exists in the sub-image region.

[0032] In this application, the Bayesian probability model is established based on the following prior knowledge: the static visual feature values ​​of the fog region follow a first low-mean Gaussian distribution, and the static visual feature values ​​of the non-fog region follow a first high-mean Gaussian distribution; the optical flow amplitude in the dynamic visual features of the fog region follows a second low-mean Gaussian distribution, and the optical flow amplitude in the non-fog region follows a second high-mean Gaussian distribution.

[0033] In step S150, the fog state level of each sub-image is bound with the corresponding timestamp and spatial pose data to generate structured segmented fog warning information. Specifically, for a sub-image where the fog state level is not empty, the predetermined pixel coordinates in the sub-image are back-projected to the geographic coordinate system based on the UAV position, attitude, and camera parameters in the spatial pose data to obtain the corresponding geographic location; the fog state level, timestamp, and geographic location are associated to form a warning information unit; the set of all warning information units in a frame constitutes structured segmented fog warning information.

[0034] For the same road segment in consecutive time periods Multiple warning levels generated internally The process involves using moving average filtering or median filtering to obtain a stable final fog state determination result.

[0035] In this application, the warning information is sent to the traffic control center for visualization in the geographic information system; the warning information is sent to the variable message signs on the affected road sections to drive them to issue corresponding speed limits or warning information; and the warning information is pushed to the vehicle terminal or navigation application to provide drivers with real-time traffic information.

[0036] Embodiment 1 of this invention proposes a fog state detection method based on road region segmentation, which innovatively integrates SURF static texture features and optical flow dynamic motion features in parallel. The two types of features are complementary in dealing with challenges such as changes in UAV perspective and low background texture: when image blurring causes texture features to fail, the special optical flow pattern caused by fog can still provide valid evidence, thereby significantly improving the detection reliability in complex scenes.

[0037] To fully demonstrate the effectiveness of the fog detection method based on road region division proposed in Embodiment 1 of this invention, multiple sets of UAV aerial photography experiments were conducted. The number of SURF feature points and optical flow amplitude were statistically analyzed in different ROI regions and compared with the fusion discrimination results. This invention divides the fog detection into four equally sized trapezoidal ROI maps. Figure 3This is a schematic diagram of the SURF algorithm extraction results proposed in Embodiment 1 of the present invention. Under fog-free conditions, SURF feature points are evenly distributed and sufficient in each ROI, while in foggy scenes, the number of feature points is significantly reduced. Table 1 below shows the SURF distribution for different images and ROIs.

[0038] Table 1: SURF distribution of different images and ROIs.

[0039]

[0040] according to Figure 3 As shown in Table 2, the distribution of SURF feature points follows a pattern: the density of distant ROIs is lower, while the density of nearby ROIs is higher. This is because the fog concentration is higher in distant areas, allowing SURF to capture the static features of the fog. In contrast, visibility is relatively better in nearby areas, enabling the generation of more feature points for ground targets. Therefore, SURF features can effectively describe the state of fog in real-world scenarios.

[0041] Figure 4 The diagram below shows the optical flow extraction results proposed in Embodiment 1 of the present invention; Table 2 below shows the optical flow amplitude distribution for different images and ROIs.

[0042] Table 2: Optical flow amplitude distribution for different images and ROIs

[0043] refer to Figure 4 As shown in Table 2, the optical flow amplitude exhibits a distribution pattern: smaller values ​​in distant ROIs and larger values ​​in nearby ROIs. This is because the high fog concentration in distant ROIs results in smaller pixel variations between consecutive frames captured by the moving drone. In contrast, the better visibility at closer distances leads to significant pixel variations due to the dynamic changes in the drone's viewpoint.

[0044] In summary, Embodiment 1 of this invention proposes a fog state detection method based on road area division, which intuitively reflects the dual impact of fog on static and dynamic states in the image detection results. It has the advantages of strong localization, good interpretability and high real-time performance.

[0045] Embodiment 1 of this invention proposes a fog state detection method based on road area division. Within multiple ROIs consistent with the road, it fuses two types of evidence: SURF static features and optical flow dynamic features. It can output the fog level and boundary located by road segment. Compared with methods that only use full frame contrast or single motion, it significantly reduces false alarms and improves robustness to UAV perspective changes. The algorithm is lightweight and can run in real time at the edge, and the threshold can be adapted with height and pitch. The results are bound to time and latitude and longitude, which is convenient for traffic guidance and emergency linkage.

[0046] Example 2 The present invention also proposes a device, Figure 5 This is a schematic diagram of a fog state detection device based on road area division proposed in Embodiment 1 of the present invention, including: Memory, used to store computer programs; When a processor executes the computer program, the method steps are as follows: In step S100, a continuous sequence of aerial images acquired by the UAV while cruising along the road is collected, and the timestamp corresponding to each frame of aerial image and the spatial pose data of the UAV when capturing each frame of aerial image are recorded simultaneously. In step S110, based on the road perspective principle, each frame of aerial image is divided into multiple sub-images along the road extension direction; In step S120, static visual features characterizing the image texture clarity are extracted. In this application, the SURF algorithm is used to detect and describe feature points in the sub-image, and the number of detected feature points or surface density is used as static visual features. In step S130, dynamic visual features characterizing pixel motion patterns are extracted. This application calculates the motion vectors of pixels in a sub-image in consecutive frame images based on optical flow, and uses the average amplitude and / or direction consistency index of the motion vectors as dynamic visual features. In step S140, static visual features and dynamic visual features extracted from the same sub-image are fused together to calculate the posterior probability of the presence of fog in the sub-image, and the fog state level of the sub-image is determined based on the posterior probability. In step S150, the fog state level of each sub-image is bound with the corresponding timestamp and spatial pose data to generate structured segmented fog warning information.

[0047] Embodiment 2 of this invention proposes a fog state detection device based on road region segmentation, which innovatively integrates SURF static texture features and optical flow dynamic motion features in parallel. The two types of features are complementary in dealing with challenges such as changes in UAV perspective and low background texture: when image blurring causes texture features to fail, the special optical flow pattern caused by fog can still provide valid evidence, thereby significantly improving the detection reliability in complex scenes.

[0048] It should be noted that the present invention also provides an electronic device, including: a communication interface capable of interacting with other devices such as network devices; and a processor connected to the communication interface to enable information interaction with other devices, used to execute a fog state detection method based on road area division provided by one or more of the above technical solutions when running a computer program, wherein the computer program is stored in a memory. In practical applications, the various components of the electronic device are coupled together through a bus system. It is understood that the bus system is used to realize the connection and communication between these components. In addition to a data bus, the bus system also includes a power bus, a control bus, and a status signal bus. The memory in the embodiments of this application is used to store various types of data to support the operation of the electronic device. Examples of this data include any computer program used to operate on the electronic device. It is understood that the memory can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic random access memory, flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache.By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM). The memories described in the embodiments of this application are intended to include, but are not limited to, these and any other suitable types of memory. The methods disclosed in the embodiments of this application can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by integrated logic circuits in the processor hardware or by instructions in software form. The processor can be a general-purpose processor, a DSP (Digital Signal Processing, i.e., a chip capable of implementing digital signal processing technology), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. A general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software modules can be located in a storage medium, which is located in memory. The processor reads the program from the memory and, in conjunction with its hardware, completes the steps of the aforementioned method. When the processor executes the program, it implements the corresponding processes in the various methods of the embodiments of this application; for simplicity, these will not be elaborated further here.

[0049] The description of the relevant parts of the fog state detection device based on road area division provided in Embodiment 2 of this application can be found in the detailed description of the corresponding parts of the fog state detection method based on road area division provided in Embodiment 1 of this application, and will not be repeated here.

[0050] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that the elements inherent in a process, method, article, or apparatus that includes a list of elements are included. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Additionally, portions of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0051] While specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art can make other modifications or variations based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for detecting patchy fog state based on road area division, characterized in that, Includes the following steps: Collect a continuous sequence of aerial images taken by a drone while it is cruising along a road, and simultaneously record the timestamp corresponding to each frame of the aerial image and the spatial pose data of the drone when it takes each frame of the aerial image. Based on the principle of road perspective, each frame of aerial image is divided into multiple sub-images along the road extension direction; within each sub-image, static visual features characterizing image texture clarity and dynamic visual features characterizing pixel motion patterns are extracted in parallel. By fusing static and dynamic visual features extracted from the same sub-image, the posterior probability of the presence of fog in the sub-image is calculated, and the fog state level of the sub-image is determined based on the posterior probability. By binding the fog state level of each sub-image with the corresponding timestamp and spatial pose data, a structured segmented fog warning information is generated.

2. The method according to claim 1, characterized in that, The spatial pose data includes at least: longitude, latitude, altitude, and attitude angle.

3. The method according to claim 1, characterized in that, Based on the principle of road perspective, each frame of aerial imagery is divided into multiple sub-images along the road's extension direction, specifically: Based on the width, height, and number of pre-divisions of the aerial image, the vertex pixel coordinates of each trapezoidal sub-image are calculated; where the upper base of the trapezoid corresponds to the far end of the road in the image, the lower base corresponds to the near end of the road, and the waistline of the trapezoid follows the same trend as the vanishing perspective line of the road edge.

4. The method according to claim 3, characterized in that, The formula for calculating the vertex pixel coordinates of each trapezoidal sub-image, based on the image's width, height, and the number of pre-divisions, is as follows: The coordinates of the top left vertex are: ; The coordinates of the top right vertex are: ; The coordinates of the bottom left vertex are: ; The coordinates of the bottom right vertex are: ; in, The number of sub-images; ; This indicates the horizontal pixel position of the top-left corner of the trapezoid. This indicates the horizontal pixel position of the bottom left corner base point of the trapezoid. This indicates the horizontal pixel position of the top right corner base point of the trapezoid; This indicates the horizontal pixel position of the bottom right corner base point of the trapezoid; Indicates the width of the sub-image; Indicates the height of the sub-image.

5. The method according to claim 1, characterized in that, Within each sub-image, static visual features characterizing image texture sharpness and dynamic visual features characterizing pixel motion patterns are extracted in parallel, specifically: The SURF algorithm is used to detect and describe feature points within a sub-image, and the number of detected feature points or surface density is used as static visual features. The motion vectors of pixels in a sub-image in consecutive frames are calculated based on the optical flow method, and the average magnitude and / or direction consistency index of the motion vectors are used as dynamic visual features.

6. The method according to claim 5, characterized in that, The SURF algorithm is used to detect and describe feature points within a sub-image, and the number of detected feature points or surface density is used as static visual features. Specifically: For any position within the subimage Calculate its integral image The integral image is represented as: ， in, Pixel intensity; At multiple scales Below, the box filter response is calculated using the integral image to approximate the second Gaussian derivative of the image. , , Construct the Hessian matrix for each pixel: By calculating the determinant of the Hessian matrix Furthermore, non-maximum suppression is performed in both scale space and image space to locate feature points within the sub-image; For each feature point, calculate the Haar wavelet response within a circular neighborhood; calculate the principal direction of the feature point. Specifically: ; and These represent the Haar wavelet transforms in the horizontal and vertical directions, respectively. It is the inverse tangent function in the four quadrants; Construct a 64-dimensional descriptor vector for each feature point based on the main direction; count the total number of all feature points within the sub-image. and will or its area with sub-image ratio Static visual features characterizing image texture sharpness .

7. The method according to claim 6, characterized in that, The motion vectors of pixels in a sub-image in consecutive frames are calculated based on optical flow, and the average magnitude and / or direction consistency index of the motion vectors are used as dynamic visual features. Specifically, based on two consecutive frames of images and Solving the optical flow constraint equations ; Obtain the optical flow vector of each pixel within the sub-image ; The horizontal component representing the optical flow vector; Represents the vertical component of the optical flow vector; is the partial derivative of the subimage in the horizontal direction; This is the partial derivative of the subimage in the vertical direction; For sub-images in time gradient on; Based on the optical flow vector, the dynamic visual feature value of the sub-image is calculated, and the dynamic visual feature value includes the average amplitude of the optical flow vector and / or the directional consistency index of the optical flow vector; The formula for calculating the average amplitude is as follows: ; The total number of pixels within the sub-image.

8. The method according to claim 7, characterized in that, By fusing static and dynamic visual features extracted from the same sub-image, the posterior probability of the presence of fog in that sub-image is calculated; specifically: The posterior probability of the presence of fog in this sub-image is calculated using a Bayesian fusion model: ； ； ； in, This represents the prior probability of the occurrence of fog patches; Represents the marginal probability of all possible combinations of features; This represents the initial probability that a patch of fog exists in a sub-image region; Represents a spatial probability distribution model; This represents the number of feature points or feature density extracted by the SURF algorithm; Indicates the mean of spatial distribution; Indicates the standard deviation of the spatial distribution; Indicates the characteristics of the running vector; This represents the average motion vector or motion intensity extracted from optical flow analysis; This represents the mean of the motion vector; This represents the standard deviation of the motion vector.

9. The method according to claim 1, characterized in that, After generating structured, segmented fog warning information, the method further includes: analyzing the same road segment over consecutive time periods. Multiple warning levels generated internally The process involves using moving average filtering or median filtering to obtain a stable final fog state determination result.

10. A fog state detection device based on road area division, comprising at least one processor and a memory, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the at least one processor, it implements a fog state detection method based on road area division as described in any one of claims 1 to 9.

Images