Cloud Height Measurement Method and System Based on Distributed Cameras

Abstract

This invention discloses a cloud height measurement method and system based on distributed cameras. The method includes: capturing images of a target cloud region using each image acquisition device in a distributed camera system to obtain images of each target cloud; performing a two-stage filtering process based on the saturation features, neighborhood texture features, and grayscale difference features of pixels in each target cloud image to obtain a classification result for each pixel in each target cloud image and performing differential processing on all pixels to obtain optimized images of each cloud; performing feature point matching on the optimized cloud images to obtain first position data for each cloud target; and calculating the height data of each cloud target based on the difference features between each first position data and the second position data of all image acquisition devices. The cloud height measurement method and system based on distributed cameras provided by this invention improves the accuracy of cloud height measurement.

Description

Technical Field

[0001] This invention relates to the field of cloud height measurement technology, and in particular to a cloud height measurement method and system based on distributed cameras. Background Technology

[0002] The formation and evolution of clouds is one of the concrete manifestations of the intricate physical processes in the atmosphere. Cloud height has a direct impact on predicting future weather changes and aviation flights. Cloud height data provides key data for aviation safety and can be used to guide route planning and take-off and landing decisions. Therefore, measuring cloud height has extremely important practical significance.

[0003] Existing technologies use machine vision to measure cloud height, but the shape of clouds is affected by various environmental factors, resulting in unclear cloud edges that cannot provide accurate information for machine vision, thus leading to low accuracy in cloud height measurement.

[0004] Therefore, improving the accuracy of cloud height measurement has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention provides a cloud height measurement method and system based on distributed cameras to solve the technical problem that the cloud shape is affected by various environmental factors, resulting in unclear cloud edges and an inability to provide accurate information for machine vision, thus leading to low accuracy in cloud height measurement.

[0006] To address the aforementioned technical problems, embodiments of the present invention provide a cloud height measurement method based on distributed cameras.

[0007] Each image acquisition device in the distributed camera system is used to capture images of the target cloud area, thus obtaining images of each target cloud layer;

[0008] A two-level filtering process is performed based on the saturation features, neighborhood texture features, and grayscale difference features of each pixel in the extracted target cloud image to obtain the classification result of each pixel in each target cloud image, wherein the classification result represents the region discrimination result of each pixel.

[0009] Based on the classification results, all pixels in each target cloud image are differentiated to obtain optimized cloud images. The differentiation process is designed to smooth the pixels located in the interference area and enhance the pixels located in the cloud area.

[0010] Feature points are extracted from each of the cloud optimization images to obtain each feature point of each cloud optimization image; the first position data of each cloud target is obtained based on the matching degree of the feature points between any two cloud optimization images.

[0011] Second location data of each of the image acquisition devices is acquired, and the height data of each cloud target is calculated based on the difference features between each of the first location data and all the second location data.

[0012] As one preferred embodiment, the classification results include sky pixels, interference pixels, and cloud pixels;

[0013] The two-level filtering process, based on the saturation features, neighborhood texture features, and grayscale difference features of pixels in each target cloud image, yields the classification result for each pixel in each target cloud image, including:

[0014] Each target grayscale cloud image is constructed based on the grayscale processing result of each target cloud image;

[0015] Each smooth saturated pixel is determined based on the saturation features and neighborhood texture smoothness of each pixel in each target grayscale cloud image; the gray level of each pixel in each target grayscale cloud image is determined based on the overall grayscale distribution features of each target grayscale cloud image; the current sky background gray level is determined based on the grayscale distribution features of all smooth saturated pixels in each target grayscale cloud image.

[0016] The gray level of each pixel in each target grayscale cloud image is compared and analyzed with the current sky background gray level of each target grayscale cloud image. Based on the comparison and analysis results, all pixels in each target cloud image are subjected to first-level filtering to obtain each sky pixel and each foreground pixel. The first-level filtering is designed to separate the background and foreground.

[0017] Based on the fusion features between the gray levels of each foreground pixel and the current sky background gray level, and the neighborhood texture features of each foreground pixel, a secondary screening is performed on all foreground pixels to obtain each interference pixel and each cloud pixel. The secondary screening is designed to separate interference objects and cloud regions in the foreground.

[0018] As one preferred embodiment, determining the gray level of each pixel in each target gray-scale cloud image based on the overall gray-scale distribution characteristics of each target gray-scale cloud image includes:

[0019] The number of gray levels in each target gray cloud image is determined based on the complexity of the gray level distribution within each target gray cloud image;

[0020] The grayscale value range of all pixels in each target grayscale cloud image is divided into several grayscale distribution intervals on an average basis according to the number of grayscale levels in each target grayscale cloud image, thereby obtaining a grayscale level mapping table for each target grayscale cloud image. In the grayscale level mapping table, each grayscale distribution interval corresponds to a grayscale level, and the number of grayscale levels and the number of grayscale distribution intervals are equal to the number of grayscale levels.

[0021] The grayscale value of each pixel in each target grayscale cloud image is compared and analyzed with the corresponding grayscale level mapping table to obtain the grayscale level of each pixel in each target grayscale cloud image.

[0022] As one preferred embodiment, determining each smooth saturated pixel based on the saturation features and neighborhood texture smoothness of each pixel in each target grayscale cloud image includes:

[0023] The neighborhood gray level difference of each pixel to be tested is calculated based on the gray level difference features among all pixels within a preset neighborhood window of each pixel to be tested in each target gray level cloud image.

[0024] Extract the saturation value of each of the pixels to be tested;

[0025] In each target grayscale cloud image, the pixels whose neighborhood grayscale difference is less than a preset difference threshold and whose saturation value is greater than a preset saturation threshold are designated as smooth saturated pixels.

[0026] As one preferred embodiment, the interfering pixels include a first interfering pixel and a second interfering pixel;

[0027] The process involves performing a secondary filtering on all foreground pixels based on the fusion features between the gray levels of each extracted foreground pixel and the current sky background gray level, and the neighborhood texture features of each foreground pixel, to obtain various interfering pixels and various cloud pixels, including:

[0028] Based on the fusion feature of the gray level of each foreground pixel in each target grayscale cloud image and the corresponding current sky background gray level, a layer of filtering is performed on all foreground pixels in each target cloud image to obtain each first interference pixel and each weak fusion pixel.

[0029] Based on the regularity of the neighborhood gradient distribution of all weakly fused pixels in each target grayscale cloud image, a two-layer screening is performed on all weakly fused pixels in each target cloud image to determine each second interference pixel and each cloud pixel.

[0030] As one preferred embodiment, the first-level screening is designed as follows:

[0031] The gray level fusion index of each pixel is calculated based on the difference between the gray level of each pixel in each target gray cloud image and the gray level of the current sky background in each target gray cloud image;

[0032] Replace the gray value of each pixel in each target grayscale cloud image with the corresponding grayscale fusion index to obtain each sky fusion image;

[0033] The Otsu's method is used to perform binary segmentation on each sky fusion image to obtain a fused binary image. All foreground pixels with a pixel value of 255 in each fused binary image are taken as first interference pixels, and all foreground pixels with a pixel value of 0 in each fused binary image are taken as weak fusion pixels.

[0034] As one preferred embodiment, the two-layer screening is designed as follows:

[0035] A neighborhood window is constructed for each of the weakly fused pixels, centered on the weakly fused pixel.

[0036] Construct an oriented gradient histogram for each neighborhood window, and use the average value of all elements in each oriented gradient histogram as a reference value for each oriented gradient histogram;

[0037] The number of elements in each directional gradient histogram that are greater than the reference value is taken as the peak number, and the ratio of the peak number of each directional gradient histogram to the length of each directional gradient histogram is taken as the neighborhood texture complexity index of the corresponding weakly fused pixel.

[0038] The neighborhood texture complexity index of all weakly fused pixels is compared and analyzed with a preset threshold: weakly fused pixels with a neighborhood texture complexity index less than or equal to the preset threshold are designated as second interference pixels; weakly fused pixels with a neighborhood texture complexity index greater than the preset threshold are designated as cloud pixels.

[0039] As a preferred embodiment, the step of performing differential processing on all pixels in each target cloud image based on the classification result to obtain optimized cloud images includes:

[0040] Sky region filling processing is performed on the pixels at the corresponding positions of each interference pixel in each target grayscale cloud image to construct an interference denoising image. The sky region filling processing is designed to calculate the fused grayscale value of each interference pixel based on the grayscale values ​​of all the sky pixels in the neighborhood of each interference pixel, and replace the grayscale value of the corresponding interference pixel with the fused grayscale value.

[0041] Image enhancement processing is performed on all cloud pixels in each of the noise-reduced interference images, and the result of the image enhancement processing is used as the cloud-optimized image.

[0042] As one preferred embodiment, the step of calculating the fused grayscale value of each interfering pixel based on the grayscale values ​​of all sky pixels within the neighborhood of each interfering pixel includes:

[0043] For each interference pixel in each target grayscale cloud image, calculate the Euclidean distance between each sky pixel in the current target grayscale cloud image and the current interference pixel;

[0044] Based on the Euclidean distance between all the sky pixels and the current interference pixels in the current target grayscale cloud image, all the sky pixels are filtered to obtain the neighboring sky pixels of the current interference pixels.

[0045] The Euclidean distance between all the neighboring sky pixels of the current interfering pixel and the current interfering pixel is normalized to obtain the normalized distance of each of the neighboring sky pixels of the current interfering pixel, wherein the sum of the normalized distances of all the neighboring sky pixels of the current interfering pixel is 1.

[0046] The grayscale values ​​of all neighboring sky pixels of each interference pixel are weighted and summed based on the normalized distance of all neighboring sky pixels of each interference pixel to obtain the fused grayscale value of each interference pixel.

[0047] Another embodiment of the present invention provides a cloud height measurement system based on distributed cameras, comprising:

[0048] The distributed image acquisition module is used to capture images of the target cloud area using each image acquisition device in the distributed camera system, thereby obtaining images of each target cloud layer.

[0049] The interference region discrimination module is used to perform a two-level filtering process based on the saturation features, neighborhood texture features and grayscale difference features of the pixels in each target cloud image, to obtain the classification result of each pixel in each target cloud image, wherein the classification result represents the region discrimination result of each pixel.

[0050] The differentiation processing module is used to perform differentiation processing on all the pixels in each target cloud image according to the classification result to obtain each cloud optimized image. The differentiation processing is designed to smooth the pixels located in the interference area and enhance the pixels located in the cloud area.

[0051] The cloud location acquisition module is used to extract feature points from each of the cloud optimization images to obtain each feature point of each cloud optimization image; and to obtain the first location data of each cloud target based on the matching degree of the feature points between any two cloud optimization images.

[0052] The cloud height measurement module is used to acquire second position data of each of the image acquisition devices, and calculate the height data of each cloud target based on the difference characteristics between each of the first position data and all the second position data.

[0053] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following:

[0054] By employing a distributed camera system to acquire images of various target cloud layers, the optimal baseline length can be adaptively selected based on cloud height. Considering the differences in the color distribution characteristics of the sky background under different weather conditions, the saturation features, neighborhood texture features, and grayscale difference features of pixels in each target cloud layer image are first analyzed to dynamically determine the grayscale level of each pixel in each target grayscale cloud layer image and the current grayscale level of the sky background. Then, the grayscale level of each pixel in each target grayscale cloud layer image and the current grayscale level of the sky background are compared and analyzed to achieve a first-level screening that separates the background and foreground, obtaining each sky pixel and each foreground pixel. Finally, the fusion features between the grayscale level of each foreground pixel and the current sky background grayscale level, the neighborhood texture features of each foreground pixel, and the fusion features between all foreground pixels are extracted. A two-stage filtering process is used to separate interfering objects in the foreground from cloud regions. Through this two-stage filtering, the classification results of each pixel in each target cloud image are accurately obtained. Based on the classification results, all pixels in each target cloud image are differentiated. Pixels located in interference areas are smoothed, and pixels located in cloud regions are enhanced, resulting in optimized cloud images. This avoids interference from interfering objects in cloud detection and enhances the edge information of the clouds, improving the accuracy of subsequent cloud height detection. Finally, the first position data of each cloud target is obtained based on the matching degree of feature points between any two optimized cloud images. Triangulation technology is used in conjunction with the second position data of each image acquisition device to analyze the differential features and calculate the height data of each cloud target, thus comprehensively improving the accuracy of cloud height detection. Attached Figure Description

[0055] Figure 1 This is a flowchart of a cloud height measurement method based on a distributed camera in one embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of a target grayscale cloud image in one embodiment of the present invention;

[0057] Figure 3 for Figure 2 A schematic diagram of smoothed saturated pixels within a target grayscale cloud image;

[0058] Figure 4 This is a structural block diagram of a cloud height measurement system based on a distributed camera according to one embodiment of the present invention;

[0059] Figure label:

[0060] 11. Distributed image acquisition module; 12. Interference area discrimination module; 13. Differentiation processing module; 14. Cloud location acquisition module; 15. Cloud height measurement module. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0062] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0063] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0064] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0065] Because the shape of clouds is affected by various environmental factors, the edges of clouds are not clear enough to provide accurate information for machine vision, resulting in low accuracy of existing visual inspection technologies in measuring cloud height.

[0066] One embodiment of the present invention provides a cloud height measurement method based on distributed cameras. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 The flowchart shown is a cloud height measurement method based on a distributed camera according to one embodiment of the present invention.

[0067] A flowchart of a cloud height measurement method based on a distributed camera in one embodiment of the present invention includes the following steps S1 to S5, as detailed below:

[0068] Step S1: Each image acquisition device in the distributed camera system takes a picture of the target cloud area to obtain images of each target cloud.

[0069] It should be noted that existing technologies measure cloud height using binocular vision. The basic principle is based on stereo vision technology, which uses two cameras to capture images of clouds from different perspectives and uses parallax to calculate the three-dimensional position of the clouds to obtain their height. However, relying on two fixed cameras results in a limited baseline, a small coverage area, and difficulty in adapting to clouds at different heights. To improve the accuracy of cloud height measurement, this invention uses a distributed camera system to measure cloud height. By dynamically adjusting the position of the image acquisition devices in the distributed camera system, the optimal baseline length is adaptively selected according to the cloud height, and statistical methods are used to reduce random errors, thereby improving the accuracy of cloud height measurement.

[0070] Step S2: A two-level filtering process is performed based on the saturation features, neighborhood texture features, and grayscale difference features of pixels in each target cloud image to obtain the classification result of each pixel in each target cloud image. The classification result represents the discrimination result of the region where each pixel is located.

[0071] It should be noted that the classification results include sky pixels, interference pixels, and cloud pixels.

[0072] Specifically, the two-stage screening process includes steps S201 to S206:

[0073] Step S201: Construct each target grayscale cloud image based on the grayscale processing result of each target cloud image.

[0074] For details, please see Figure 2 , Figure 2 The diagram shown is a schematic representation of a target grayscale cloud image in one embodiment of the present invention.

[0075] Step S202: Determine smooth saturated pixels based on the saturation features of each pixel in each target grayscale cloud image and the smoothness of the neighborhood texture.

[0076] Considering that the color distribution characteristics of the sky background differ under different weather conditions, in order to accurately extract the clouds, it is necessary to first dynamically identify the background area. Since the saturation of the sky area is high, the pixels in the sky background area often have relatively smooth neighborhood textures and occupy a large area. Therefore, the saturation and neighborhood texture smoothness of the pixels are analyzed first.

[0077] Specifically, the neighborhood grayscale difference of each pixel is calculated based on the grayscale difference features among all pixels within a preset neighborhood window of each pixel to be tested in each target grayscale cloud image; the saturation value of each pixel to be tested is extracted, and pixels in each target grayscale cloud image whose neighborhood grayscale difference is less than a preset difference threshold and whose saturation value is greater than a preset saturation threshold are taken as smoothed saturated pixels.

[0078] In one embodiment, the average of the absolute values ​​of the differences in grayscale values ​​between all pixels within a preset neighborhood window of each pixel to be tested in each target grayscale cloud image is taken as the neighborhood grayscale difference of each pixel to be tested.

[0079] It should be noted that smooth saturated pixels represent pixels that may be located in sky areas with flat textures and high saturation.

[0080] For details, please see Figure 3 , Figure 3 Shown as Figure 2 A schematic diagram of smoothed saturated pixels within a target grayscale cloud image. Figure 3 In the image, white pixels represent smooth, saturated pixels.

[0081] It should be noted that the preset difference threshold, preset saturation threshold, and preset neighborhood window length are all manually preset values. Practical analysis shows that the preset difference threshold ranges from 1 to 20, the preset neighborhood window length is an odd number (in this embodiment, the preset difference threshold is 4, the preset neighborhood window length is 9, and the preset saturation threshold is 60). When the difference in grayscale values ​​within a pixel's neighborhood window is smaller and the saturation is higher, it indicates that the pixel is more likely to be located in a flat, highly saturated sky region. To avoid interference from other flat-textured objects, further filtering is performed based on the grayscale distribution characteristics of smooth, saturated pixels.

[0082] Step S203: Determine the gray level of each pixel in each target gray-scale cloud image based on the overall gray-scale distribution characteristics of each target gray-scale cloud image.

[0083] Specifically, a gray-level mapping table for each target gray-level cloud image is dynamically determined based on the discrete features of gray-level distribution extracted from each target gray-level cloud image. The discrete features of gray-level distribution include the complexity of gray-level distribution and the range of gray-level values.

[0084] It should be noted that, considering the varying complexity of elements contained in different areas of the sky, when the target cloud image corresponds to the edge of the sky, the grayscale cloud image contains more complex elements, such as buildings, vegetation, and mountain peaks. However, when the target cloud image corresponds to the center of the sky, the grayscale cloud image contains simpler elements, possibly only including the sky, clouds, and a few interfering objects, such as birds, plastic bags, and other floating objects.

[0085] To improve the real-time performance of cloud height detection while ensuring accuracy, preferably, the number of gray levels in each target gray-scale cloud image is determined based on the complexity of the gray-scale distribution within each target gray-scale cloud image. The gray-scale value range of all pixels in each target gray-scale cloud image is then divided into several gray-scale distribution intervals based on the number of gray-scale levels in each target gray-scale cloud image, resulting in a gray-scale mapping table for each target gray-scale cloud image. In this gray-scale mapping table, each gray-scale distribution interval corresponds to a gray-scale level, and the number of gray-scale levels, the number of gray-scale distribution intervals, and the number of gray-scale levels are all equal.

[0086] In one embodiment, the standard deviation of the gray values ​​of all pixels in each target gray-scale cloud image is used as the gray-scale distribution complexity index of each target gray-scale cloud image, and the rounded value of the product of the normalized value of the gray-scale distribution complexity index and the preset level is used as the number of gray-scale levels of each target gray-scale cloud image.

[0087] It should be noted that the number of grayscale levels is a positive integer, and the preset number of levels is a manually preset value that can be set according to the complexity of the application scenario. In this embodiment, the preset number of levels is 10. For other implementation methods, the implementer can choose the value of the preset number of levels as needed. When the standard deviation of the grayscale values ​​of all pixels in the target grayscale cloud image is larger, it indicates that the grayscale value distribution of the pixels in the target grayscale cloud image is more discrete, and a larger number of grayscale levels should be set to ensure that the grayscale levels of different objects in the target grayscale cloud image can be distinguished, thereby improving the accuracy of subsequent processing. When the standard deviation of the grayscale values ​​of all pixels in the target grayscale cloud image is smaller, it indicates that the grayscale value distribution of the pixels in the target grayscale cloud image is more concentrated, and a smaller number of grayscale levels should be set to ensure the speed of subsequent processing, thereby improving the real-time performance of cloud height detection.

[0088] For example, when the normalized value of the grayscale distribution complexity index of the target grayscale cloud image is 0.6, the preset level is 100, and the grayscale value range of all pixels in the target grayscale cloud image is 0 to 255, the corresponding number of grayscale levels is calculated to be 60, and the number of grayscale values ​​allocated to each level is 256 / 60 = 4, with a remainder of 16.

[0089] Therefore, the grayscale level mapping table is as follows: For the first 44 grayscale levels, each grayscale level includes 4 grayscale values; for grayscale levels 45 to 59, each grayscale level includes 5 grayscale values; and for grayscale level 60, each grayscale level includes 1 grayscale value. See the table below:

[0090]

[0091]

[0092] Furthermore, the grayscale value of each pixel in each target grayscale cloud image is compared and analyzed with the corresponding grayscale level mapping table to determine the grayscale level of each pixel in each target grayscale cloud image.

[0093] For example, when the gray value of pixel a is 177, then the gray level of pixel a is 44.

[0094] Step S204: Determine the current sky background gray level based on the gray level distribution characteristics of all smooth saturated pixels in each target gray level cloud image.

[0095] It should be noted that a grayscale histogram is a function of grayscale distribution. By counting the frequency of occurrence of all pixels in a digital image according to their grayscale values, it is possible to statistically analyze the overall grayscale distribution characteristics of the image. Each data point in the grayscale histogram represents the probability of occurrence of each grayscale level. Since the sky background occupies a large area in the target grayscale cloud image, the grayscale level of the current sky background has the highest probability of occurrence.

[0096] In one embodiment, a grayscale histogram is drawn based on the grayscale levels of all smooth saturated pixels in each target grayscale cloud image, and the grayscale level with the highest frequency in the grayscale histogram is taken as the current sky background grayscale level.

[0097] Step S205: Compare and analyze the gray level of each pixel in each target grayscale cloud image with the current sky background gray level of each target grayscale cloud image. Based on the comparison and analysis results, perform a first-level screening on all pixels in each target cloud image to obtain each sky pixel and each foreground pixel. The first-level screening is designed to separate the background and foreground.

[0098] In one embodiment, all pixels in each target grayscale cloud image that have the same grayscale level as the current sky background are designated as sky pixels, and the remaining pixels are designated as foreground pixels.

[0099] Step S206: Based on the fusion features between the gray levels of each foreground pixel and the current sky background gray levels, and the neighborhood texture features of each foreground pixel, a secondary screening is performed on all foreground pixels to obtain each interference pixel and each cloud pixel. The secondary screening is designed to separate the interference objects in the foreground from the cloud region.

[0100] It should be noted that, in order to accurately measure cloud height, feature points in multiple target cloud images need to be matched to accurately obtain the location data of each cloud target. However, because the shape of clouds is affected by various environmental factors, the edges of clouds are not clear enough to provide accurate information for machine vision, resulting in low accuracy of cloud height measurement. To improve the accuracy of cloud feature point extraction, foreground pixels located in areas with interfering objects are smoothed, and foreground pixels located in cloud areas are enhanced.

[0101] It should be noted that since there may be interfering objects in the sky, such as power lines and buildings, whose colors differ significantly from the clouds, or objects like airplanes and plastic bags, whose colors differ less significantly from the clouds, it is necessary to filter the interfering objects based on their different characteristics. Interfering pixels are divided into first interference pixels and second interference pixels. The first interference pixel represents the interference pixel corresponding to the interference object with a significant color difference from the clouds, and the second interference pixel represents the interference pixel corresponding to the interference object with a less significant color difference from the clouds.

[0102] Specifically, based on the fusion features of the gray level of each foreground pixel in each target grayscale cloud image and the corresponding current sky background gray level, a layer of filtering is performed on all foreground pixels in each target grayscale cloud image to obtain each first interference pixel and each weak fusion pixel.

[0103] It should be noted that weakly blended pixels are foreground pixels whose gray values ​​are close to those of the sky background area, such as clouds, white or light gray drones, white doves, and plastic bags.

[0104] In one embodiment, firstly, the gray level fusion index of each pixel is calculated based on the difference between the gray level of each pixel in each target gray-scale cloud image and the gray level of the current sky background in each target gray-scale cloud image; then, the gray value of each pixel in each target gray-scale cloud image is replaced with the corresponding gray level fusion index to obtain each sky fusion image; then, the Otsu's method is used to perform binary segmentation on each sky fusion image to obtain a fused binary image, and all foreground pixels with a pixel value of 255 in each fused binary image are taken as first interference pixels, and all foreground pixels with a pixel value of 0 in each fused binary image are taken as weak fusion pixels.

[0105] In one embodiment, the gray level fusion index of each pixel is calculated based on the difference between the gray level of each pixel in each target gray-scale cloud image and the current sky background gray level of each target gray-scale cloud image:

[0106] G_con(x)=exp(-|GrayL(x)-GSkyL|)

[0107] Where G_con(x) is the gray level fusion index of pixel x, exp() is the exponential function with the natural constant as the base, GrayL(x) is the gray level of pixel x, and GSkyL is the gray level of the current sky background.

[0108] It should be noted that the grayscale fusion index represents the degree of fusion between a pixel and the sky region. It amplifies the difference between grayscale levels through an exponential function. When the difference between the grayscale level of a pixel and the current grayscale level of the sky background is small, it means that the area where the pixel is located is very likely to correspond to the sky background area or the cloud area. When the difference between the grayscale level of a pixel and the current grayscale level of the sky background is large, it means that the area where the pixel is located is very likely to correspond to the interfering object.

[0109] It should be noted that, considering the possibility of some interfering objects in the sky that are similar in color to the background area, clouds are composed of water vapor, ice crystals, etc., and are affected by factors such as airflow and temperature. Furthermore, clouds are constantly deformed with wind speed and atmospheric movement, resulting in constantly changing textures, making it difficult to form a stable gradient direction aggregation. In contrast, interfering objects are mostly industrially manufactured and, due to limitations in processing technology, often have regular edges. Therefore, by analyzing the regularity of the neighborhood gradient distribution of pixels, it is possible to further distinguish cloud areas from interfering objects.

[0110] Specifically, based on the regularity of the neighborhood gradient distribution of all weakly fused pixels in each target grayscale cloud image, a two-layer screening is performed on all weakly fused pixels in each target cloud image to determine each second interference pixel and each cloud pixel.

[0111] It should be noted that the second interference pixel represents the interference pixel corresponding to an object whose color is relatively similar to that of the cloud layer, but whose texture distribution is relatively regular.

[0112] In one embodiment, firstly, a neighborhood window is constructed centered on each weakly fused pixel, and an orienting gradient histogram is constructed for each neighborhood window. The average value of all elements in each orienting gradient histogram is used as the reference value for each orienting gradient histogram. Then, the number of elements in each orienting gradient histogram that are greater than the reference value is used as the peak value, and the ratio of the peak value to the length of each orienting gradient histogram is used as the neighborhood texture complexity index of the corresponding weakly fused pixel. Then, the neighborhood texture complexity index of all weakly fused pixels is compared and analyzed with a preset threshold: weakly fused pixels with a neighborhood texture complexity index less than or equal to the preset threshold are designated as second interference pixels; weakly fused pixels with a neighborhood texture complexity index greater than the preset threshold are designated as cloud pixels.

[0113] It should be noted that the Histogram of Oriented Gradients (HGP) constructs a histogram to describe the texture features of an image by calculating the gradient direction and intensity of each local region. The construction of the HGP is a well-known technique. When the HGP shows sparse peaks, it indicates that the texture distribution within the neighborhood window of the corresponding weakly fused pixel is simpler, and it is more likely to be located in an area corresponding to industrially manufactured interference objects. When the HGP shows multi-peak or uniform distribution, it indicates that the texture distribution within the neighborhood window of the corresponding weakly fused pixel is more complex, and it is more likely to be located in a cloud region. The construction process of the HGP is not described in detail in this embodiment.

[0114] As one embodiment of this application, the side length of the neighborhood window is 9, and the preset threshold value is 0.7. As for the setting of the side length of the neighborhood window and the value of the preset threshold, the implementer can choose for himself in other implementation methods, and this application does not impose any special restrictions here.

[0115] Step S3: Based on the classification results, perform differential processing on all pixels in each target cloud image to obtain optimized cloud images. The differential processing is designed to smooth pixels located in interference areas and enhance pixels located in cloud areas.

[0116] Specifically, sky region filling processing is performed on the pixels at the corresponding positions of each interference pixel in each target grayscale cloud image to construct an interference-denoising image. The sky region filling processing is designed to calculate the fused grayscale value of each interference pixel based on the grayscale values ​​of all sky pixels in the neighborhood of each interference pixel, and replace the grayscale value of the corresponding interference pixel with the fused grayscale value.

[0117] Simultaneously, image enhancement processing is performed on each cloud pixel in each noise-reduced image, and the result of the image enhancement processing is used as the cloud optimization image.

[0118] In one embodiment, a nonlocal mean filtering algorithm is used to perform image enhancement processing on the pixels at the corresponding positions of each cloud pixel in the noise-reduced image.

[0119] It should be noted that the nonlocal mean filtering algorithm uses two fixed-size windows: a large search window and a small neighborhood window. The neighborhood window slides within the search window, and the influence (i.e., weight) of the corresponding center pixel on the current pixel is determined based on the similarity between neighbors. Nonlocal mean filtering, through global similarity mining and adaptive weighting, preserves key details of the cloud layer while removing noise, improving the accuracy of subsequent cloud height detection. The nonlocal mean filtering algorithm is a well-known technique and will not be elaborated upon in this embodiment.

[0120] In one embodiment, the fused grayscale value of each interfering pixel is calculated based on the grayscale values ​​of all sky pixels within the neighborhood of each interfering pixel, including steps S301 to S304:

[0121] Step S301: For each interfering pixel in each target grayscale cloud image, calculate the Euclidean distance between each sky pixel in the current target grayscale cloud image and the current interfering pixel.

[0122] Step S302: Based on the Euclidean distance between all sky pixels and the current interference pixel in the current target grayscale cloud image, filter all sky pixels to obtain the neighboring sky pixels of the current interference pixel.

[0123] Step S303: Normalize the Euclidean distance between all neighboring sky pixels of the current interfering pixel and the current interfering pixel to obtain the normalized distance of each neighboring sky pixel of the current interfering pixel, wherein the sum of the normalized distances of all neighboring sky pixels of the current interfering pixel is 1.

[0124] Step S304: Based on the normalized distance of all neighboring sky pixels of each interfering pixel, perform a weighted sum of the gray values ​​of all neighboring sky pixels of each interfering pixel to obtain the fused gray value of each interfering pixel.

[0125]

[0126] Where G_replaced(y) is the fused grayscale value of the interfering pixel y, and Norm_d(y,s) is the normalized distance between the s-th neighboring sky pixel of the interfering pixel y and the interfering pixel y. Let y be the grayscale value of the s-th neighboring sky pixel of the interfering pixel y, and n be the number of neighboring sky pixels of each interfering pixel.

[0127] It should be noted that the fused grayscale value represents the grayscale value after fusing the interfering pixels and the sky background area. A larger weight is assigned to sky pixels closer to the interfering pixels, and a smaller weight is assigned to sky pixels farther away from the interfering pixels, to avoid grayscale jumps at the boundaries. The number n of neighboring sky pixels for each interfering pixel is a preset value. In this embodiment, the number of neighboring sky pixels is 5. For other implementation methods, the implementer can choose the appropriate value for the number of neighboring sky pixels; this application does not impose any special restrictions here.

[0128] It should be further explained that by replacing the grayscale values ​​of interfering pixels with the grayscale values ​​of pixels in the neighborhood that correspond to the current grayscale level of the sky background, instead of directly replacing the grayscale values ​​of interfering pixels with fixed values, uniform color blocks are prevented, the natural transition of the background is maintained, the "patch effect" caused by fixed value filling is avoided, and the replaced area is not mistakenly detected as a cloud as a whole in the subsequent cloud detection process, thus improving the accuracy of subsequent cloud feature point acquisition.

[0129] Step S4: Extract feature points from each cloud layer optimized image to obtain each feature point of each cloud layer optimized image; obtain the first position data of each cloud layer target based on the matching degree of feature points between any two cloud layer optimized images.

[0130] As an embodiment of this application, the SIFT corner detection algorithm is used to extract each feature point in the optimized image of each cloud layer.

[0131] Step S5: Obtain the second position data of each image acquisition device, and calculate the height data of each cloud target based on the difference characteristics between each first position data and all second position data.

[0132] It should be noted that triangulation is a distance measurement method based on the principles of geometric optics. It involves measuring the same target from two or more observation points, such as cameras or laser sensors, and calculating the target's distance or three-dimensional position using the similarity of triangles. When multiple image acquisition devices observe the same cloud point from different locations, the pixel position of that point in multiple images of the target cloud will exhibit a horizontal shift, known as parallax. This shift can be used to deduce the vertical height of the cloud layer.

[0133] The cloud height measurement method based on distributed cameras provided in this invention uses a distributed camera system to acquire images of various target cloud layers, and can adaptively select the optimal baseline length according to the cloud height. Considering that the color distribution characteristics of the sky background differ under different weather conditions, the saturation features, neighborhood texture features, and grayscale difference features of pixels in each target cloud layer image are first analyzed to dynamically determine the grayscale level of each pixel in each target grayscale cloud layer image and the current grayscale level of the sky background. Then, the grayscale level of each pixel in each target grayscale cloud layer image and the current grayscale level of the sky background are compared and analyzed to achieve a first-level screening to separate the background and foreground, obtaining each sky pixel and each foreground pixel. Finally, the fusion features between the grayscale level of each foreground pixel and the current grayscale level of the sky background, as well as the neighboring features of each foreground pixel, are extracted. The system utilizes domain texture features and performs secondary filtering on all foreground pixels to separate interfering objects and cloud regions. Through this two-stage filtering, it accurately obtains the classification results for each pixel in each target cloud image. Based on the classification results, it performs differential processing on all pixels in each target cloud image, smoothing pixels located in interference areas and enhancing pixels located in cloud regions to obtain optimized cloud images. This avoids interference from interfering objects in cloud detection and enhances the edge information of the clouds, improving the accuracy of subsequent cloud height detection. Finally, based on the matching degree of feature points between any two optimized cloud images, it obtains the first position data of each cloud target. It then uses triangulation technology combined with the second position data of each image acquisition device to analyze differential features and calculate the height data of each cloud target, thus comprehensively improving the accuracy of cloud height detection.

[0134] Another embodiment of the present invention provides a cloud height measurement system based on distributed cameras. For details, please refer to [link to relevant documentation]. Figure 4 , Figure 4 The diagram shown illustrates a structural block diagram of a cloud height measurement system based on a distributed camera, according to one embodiment of the present invention, comprising:

[0135] The distributed image acquisition module 11 is used to capture images of the target cloud area using each image acquisition device in the distributed camera system, thereby obtaining images of each target cloud.

[0136] The interference region discrimination module 12 is used to perform a two-level screening process based on the saturation features, neighborhood texture features and grayscale difference features of pixels in each target cloud image to obtain the classification result of each pixel in each target cloud image, wherein the classification result represents the region discrimination result of each pixel.

[0137] The differential processing module 13 is used to perform differential processing on all pixels in each target cloud image according to the classification results to obtain optimized images of each cloud. The differential processing is designed to smooth the pixels located in the interference area and enhance the pixels located in the cloud area.

[0138] The cloud location acquisition module 14 is used to extract feature points from each cloud optimization image to obtain each feature point of each cloud optimization image; and to obtain the first location data of each cloud target based on the matching degree of feature points between any two cloud optimization images.

[0139] The cloud height measurement module 15 is used to acquire the second position data of each image acquisition device and calculate the height data of each cloud target based on the difference characteristics between each first position data and all second position data.

[0140] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A cloud height measurement method based on distributed cameras, characterized in that, The method includes: Each image acquisition device in the distributed camera system is used to capture images of the target cloud area, thus obtaining images of each target cloud layer; A two-level filtering process is performed based on the saturation features, neighborhood texture features, and grayscale difference features of each pixel in the extracted target cloud image to obtain the classification result of each pixel in each target cloud image, wherein the classification result represents the region discrimination result of each pixel. Based on the classification results, all pixels in each target cloud image are differentiated to obtain optimized cloud images. The differentiation process is designed to smooth the pixels located in the interference area and enhance the pixels located in the cloud area. Feature points are extracted from each of the cloud optimization images to obtain each feature point of each cloud optimization image; the first position data of each cloud target is obtained based on the matching degree of the feature points between any two cloud optimization images. Acquire second location data for each of the image acquisition devices, and calculate the height data of each cloud target based on the difference features between each of the first location data and all the second location data; The classification results include sky pixels, interference pixels, and cloud pixels. The two-level filtering process, based on the saturation features, neighborhood texture features, and grayscale difference features of pixels in each target cloud image, yields the classification result for each pixel in each target cloud image, including: Each target grayscale cloud image is constructed based on the grayscale processing result of each target cloud image; Each smooth saturated pixel is determined based on the saturation features and neighborhood texture smoothness of each pixel in each target grayscale cloud image; the gray level of each pixel in each target grayscale cloud image is determined based on the overall grayscale distribution features of each target grayscale cloud image; the current sky background gray level is determined based on the grayscale distribution features of all smooth saturated pixels in each target grayscale cloud image. The gray level of each pixel in each target grayscale cloud image is compared and analyzed with the current sky background gray level of each target grayscale cloud image. Based on the comparison and analysis results, all pixels in each target cloud image are subjected to first-level filtering to obtain each sky pixel and each foreground pixel. The first-level filtering is designed to separate the background and foreground. Based on the fusion features between the gray levels of each foreground pixel and the current sky background gray level, and the neighborhood texture features of each foreground pixel, a secondary screening is performed on all foreground pixels to obtain each interference pixel and each cloud pixel. The secondary screening is designed to separate interference objects and cloud regions in the foreground.

2. The cloud height measurement method based on distributed cameras according to claim 1, characterized in that, Determining the gray level of each pixel in each target gray-scale cloud image based on the overall gray-scale distribution characteristics of each target gray-scale cloud image includes: The number of gray levels in each target gray cloud image is determined based on the complexity of the gray level distribution within each target gray cloud image; The grayscale value range of all pixels in each target grayscale cloud image is divided into several grayscale distribution intervals on an average basis according to the number of grayscale levels in each target grayscale cloud image, thereby obtaining a grayscale level mapping table for each target grayscale cloud image. In the grayscale level mapping table, each grayscale distribution interval corresponds to a grayscale level, and the number of grayscale levels and the number of grayscale distribution intervals are equal to the number of grayscale levels. The grayscale value of each pixel in each target grayscale cloud image is compared and analyzed with the corresponding grayscale level mapping table to obtain the grayscale level of each pixel in each target grayscale cloud image.

3. The cloud height measurement method based on distributed cameras according to claim 1, characterized in that, The step of determining each smooth saturated pixel based on the saturation features and neighborhood texture smoothness of each pixel in each target grayscale cloud image includes: The neighborhood gray level difference of each pixel to be tested is calculated based on the gray level difference features among all pixels within a preset neighborhood window of each pixel to be tested in each target gray level cloud image. Extract the saturation value of each of the pixels to be tested; In each target grayscale cloud image, the pixels whose neighborhood grayscale difference is less than a preset difference threshold and whose saturation value is greater than a preset saturation threshold are designated as smooth saturated pixels.

4. The cloud height measurement method based on distributed cameras according to claim 1, characterized in that, The interfering pixels include a first interfering pixel and a second interfering pixel; The process involves performing a secondary filtering on all foreground pixels based on the fusion features between the gray levels of each extracted foreground pixel and the current sky background gray level, and the neighborhood texture features of each foreground pixel, to obtain various interfering pixels and various cloud pixels, including: Based on the fusion feature of the gray level of each foreground pixel in each target grayscale cloud image and the corresponding current sky background gray level, a layer of filtering is performed on all foreground pixels in each target cloud image to obtain each first interference pixel and each weak fusion pixel. Based on the regularity of the neighborhood gradient distribution of all weakly fused pixels in each target grayscale cloud image, a two-layer screening is performed on all weakly fused pixels in each target cloud image to determine each second interference pixel and each cloud pixel.

5. The cloud height measurement method based on distributed cameras according to claim 4, characterized in that, The first-level screening is designed as follows: The gray level fusion index of each pixel is calculated based on the difference between the gray level of each pixel in each target gray cloud image and the gray level of the current sky background in each target gray cloud image; Replace the gray value of each pixel in each target grayscale cloud image with the corresponding grayscale fusion index to obtain each sky fusion image; The Otsu's method is used to perform binary segmentation on each sky fusion image to obtain a fused binary image. All foreground pixels with a pixel value of 255 in each fused binary image are taken as first interference pixels, and all foreground pixels with a pixel value of 0 in each fused binary image are taken as weak fusion pixels.

6. The cloud height measurement method based on distributed cameras according to claim 4, characterized in that, The two-layer screening is designed as follows: A neighborhood window is constructed for each of the weakly fused pixels, centered on the weakly fused pixel. Construct an oriented gradient histogram for each neighborhood window, and use the average value of all elements in each oriented gradient histogram as a reference value for each oriented gradient histogram; The number of elements in each directional gradient histogram that are greater than the reference value is taken as the peak number, and the ratio of the peak number of each directional gradient histogram to the length of each directional gradient histogram is taken as the neighborhood texture complexity index of the corresponding weakly fused pixel. The neighborhood texture complexity index of all weakly fused pixels is compared and analyzed with a preset threshold: weakly fused pixels with a neighborhood texture complexity index less than or equal to the preset threshold are designated as second interference pixels; weakly fused pixels with a neighborhood texture complexity index greater than the preset threshold are designated as cloud pixels.

7. The cloud height measurement method based on distributed cameras according to claim 1, characterized in that, The step of performing differential processing on all pixels in each target cloud image based on the classification result to obtain optimized cloud images includes: Sky region filling processing is performed on the pixels at the corresponding positions of each interference pixel in each target grayscale cloud image to construct an interference denoising image. The sky region filling processing is designed to calculate the fused grayscale value of each interference pixel based on the grayscale values ​​of all the sky pixels in the neighborhood of each interference pixel, and replace the grayscale value of the corresponding interference pixel with the fused grayscale value. Image enhancement processing is performed on all cloud pixels in each of the noise-reduced interference images, and the result of the image enhancement processing is used as the cloud-optimized image.

8. The cloud height measurement method based on distributed cameras according to claim 7, characterized in that, The step of calculating the fused grayscale value of each interference pixel based on the grayscale values ​​of all sky pixels within the neighborhood of each interference pixel includes: For each interference pixel in each target grayscale cloud image, calculate the Euclidean distance between each sky pixel in the current target grayscale cloud image and the current interference pixel; Based on the Euclidean distance between all the sky pixels and the current interference pixels in the current target grayscale cloud image, all the sky pixels are filtered to obtain the neighboring sky pixels of the current interference pixels. The Euclidean distance between all the neighboring sky pixels of the current interfering pixel and the current interfering pixel is normalized to obtain the normalized distance of each of the neighboring sky pixels of the current interfering pixel, wherein the sum of the normalized distances of all the neighboring sky pixels of the current interfering pixel is 1. The grayscale values ​​of all neighboring sky pixels of each interference pixel are weighted and summed based on the normalized distance of all neighboring sky pixels of each interference pixel to obtain the fused grayscale value of each interference pixel.

9. A cloud height measurement system based on distributed cameras, characterized in that, The system employs the cloud height measurement method based on distributed cameras as described in any one of claims 1 to 8, wherein the method is executed collaboratively by the following modules to achieve cloud height measurement: The distributed image acquisition module is used to capture images of the target cloud area using each image acquisition device in the distributed camera system, thereby obtaining images of each target cloud layer. The interference region discrimination module is used to perform a two-level filtering process based on the saturation features, neighborhood texture features and grayscale difference features of the pixels in each target cloud image, to obtain the classification result of each pixel in each target cloud image, wherein the classification result represents the region discrimination result of each pixel. The differentiation processing module is used to perform differentiation processing on all the pixels in each target cloud image according to the classification result to obtain each cloud optimized image. The differentiation processing is designed to smooth the pixels located in the interference area and enhance the pixels located in the cloud area. The cloud location acquisition module is used to extract feature points from each of the cloud optimization images to obtain each feature point of each cloud optimization image; and to obtain the first location data of each cloud target based on the matching degree of the feature points between any two cloud optimization images. The cloud height measurement module is used to acquire second position data of each of the image acquisition devices, and calculate the height data of each cloud target based on the difference characteristics between each of the first position data and all the second position data.

Images