Three-dimensional point cloud generation method and system based on multi-view remote sensing image matching

Abstract

The invention provides a three-dimensional point cloud generation method and system based on multi-view remote sensing image matching, and relates to the technical field of remote sensing. By obtaining a multi-view remote sensing image set of a target area, a multi-view dynamic association initial frame is constructed, and shooting orientation association information is associated to obtain an initial association relationship; extracting feature region association attributes to generate association factors, constructing a cross-view dynamic matching constraint rule set, implementing coordinate mapping to form a three-dimensional point cloud initial unit set, extracting spatial association attributes, dividing initial unit groups, and performing progressive spatial position adjustment to obtain a three-dimensional point cloud optimization unit set; and finally, performing spatial arrangement according to the spatial distribution logic of the target area, and integrating to form a three-dimensional point cloud result of the target area, thereby effectively improving the quality and precision of three-dimensional point cloud generation.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically, to a method and system for generating three-dimensional point clouds based on multi-view remote sensing image matching. Background Technology

[0002] 3D point cloud generation is of vital importance for many applications, including geographic information acquisition, urban planning, and disaster monitoring. Traditional 3D point cloud generation methods mainly rely on single-view remote sensing imagery or data from active sensors such as LiDAR.

[0003] Single-view remote sensing imagery, because it can only acquire information about the target area from a specific angle, suffers from incomplete information and struggles to accurately reflect the three-dimensional structure of objects. For example, in 3D reconstruction of buildings, single-view imagery cannot capture detailed information about the back of the building, resulting in missing or inaccurate 3D models.

[0004] While active sensors like LiDAR can directly acquire the three-dimensional coordinates of objects, they are expensive and highly susceptible to environmental factors. In complex environments, such as dense forests or urban areas with tall buildings, LiDAR scanning may be obstructed, leading to incomplete data. Furthermore, the processing and analysis of LiDAR data are complex and require specialized technical and software support.

[0005] Currently, although there are some methods for generating 3D point clouds based on multi-view remote sensing imagery, these methods have shortcomings in image association, feature matching, and 3D point cloud optimization. For example, in the image association stage, only simple geometric relationships are often considered, ignoring the shooting orientation association information between images from different viewpoints, leading to inaccurate association; in the feature matching process, there is a lack of effective dynamic matching constraint rules, which easily leads to mismatches; and in the 3D point cloud optimization stage, the spatial association attributes are not fully utilized, failing to effectively improve the quality and accuracy of the 3D point cloud. Summary of the Invention

[0006] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for generating three-dimensional point clouds based on multi-view remote sensing image matching, the method comprising:

[0007] A set of multi-view remote sensing images of the target area is obtained. Based on the shooting view parameters of each remote sensing image, an initial framework for multi-view dynamic association is constructed. The initial association relationship of multi-view images is obtained by associating the shooting orientation association information between images from different viewpoints.

[0008] Based on the initial association relationship of the multi-view images, the association attributes of corresponding feature regions in the images from different viewpoints are extracted, feature region association factors are generated, and a set of cross-view dynamic matching constraint rules is constructed by combining the change law of feature region association factors.

[0009] Using the cross-view dynamic matching constraint rule set, coordinate mapping operation is performed on the feature regions in images from different viewpoints to convert the two-dimensional coordinates of each feature region into three-dimensional spatial coordinates, forming an initial set of three-dimensional point cloud units.

[0010] For the set of initial units of the three-dimensional point cloud, the spatial association attributes of each initial unit of the three-dimensional point cloud are extracted, and the initial units of the three-dimensional point cloud are divided into groups based on the attribute similarity. The spatial position of each group of initial units of the three-dimensional point cloud is adjusted progressively to obtain the set of optimized units of the three-dimensional point cloud.

[0011] According to the spatial distribution logic of the target area, the three-dimensional point cloud optimization units in the set of three-dimensional point cloud optimization units are arranged in space and integrated to form the three-dimensional point cloud result of the target area.

[0012] Furthermore, embodiments of the present invention also provide a three-dimensional point cloud generation system based on multi-view remote sensing image matching, characterized in that it includes:

[0013] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to perform the above-described method for generating 3D point clouds based on multi-view remote sensing image matching by executing the machine-executable instructions.

[0014] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, and the processor executing the machine-executable instructions, causing the computer device to execute the above-described method for generating three-dimensional point clouds based on multi-view remote sensing image matching.

[0015] Based on the above, by constructing a multi-view dynamic association initial framework and associating the shooting orientation association information between images from different viewpoints, the initial association relationship of multi-view images can be obtained comprehensively and accurately. In the process of extracting the association attributes of feature regions and constructing a set of cross-view dynamic matching constraint rules, the variation law of the association factors of feature regions is fully considered, so that the feature regions in images from different viewpoints can be matched more accurately, greatly reducing the probability of mismatch and improving the accuracy of the initial unit set of 3D point cloud. By extracting spatial association attributes and dividing based on attribute similarity for the initial unit set of 3D point cloud, and implementing progressive spatial position adjustment, the spatial position of 3D point cloud can be further optimized, making the generated optimized unit set of 3D point cloud more consistent with the actual spatial distribution of the target area. Finally, the spatial arrangement and integration are carried out according to the spatial distribution logic of the target area to form the 3D point cloud result, which significantly improves the quality and accuracy of 3D point cloud. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the execution flow of the three-dimensional point cloud generation method based on multi-view remote sensing image matching provided in the embodiments of the present invention.

[0017] Figure 2 This is a schematic diagram of exemplary hardware and software components of a 3D point cloud generation system based on multi-view remote sensing image matching provided in an embodiment of the present invention. Detailed Implementation

[0018] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for generating 3D point clouds based on multi-view remote sensing image matching, provided in one embodiment of the present invention. The following is a detailed description of this method for generating 3D point clouds based on multi-view remote sensing image matching.

[0019] Step S110: Obtain a set of multi-view remote sensing images of the target area, construct an initial framework for multi-view dynamic association based on the shooting view parameters of each remote sensing image, and obtain the initial association relationship of multi-view images by associating the shooting orientation association information between images from different viewpoints.

[0020] In this embodiment, a specific urban area is selected as the target area. This urban area encompasses various terrain types, such as residential areas, commercial areas, and industrial areas. First, a set of multi-view remote sensing images of the urban area is acquired using satellite remote sensing technology and UAV aerial photography technology. These images were captured in different seasons, and the shooting angles cover all four directions (north, south, east, and west) and different altitudes. Each remote sensing image is accompanied by detailed shooting perspective parameters, including latitude and longitude coordinates, altitude, camera pitch angle, yaw angle, and roll angle.

[0021] After acquiring the multi-view remote sensing image set, the initial framework for multi-view dynamic association was constructed. First, the shooting view parameters of each remote sensing image were organized and standardized, unifying the parameter format and units. For example, latitude and longitude coordinates were converted to relative coordinates with the center point of the target area as the origin, and angle parameters were uniformly converted to radians. Then, based on these standardized shooting view parameters, a three-dimensional spatial coordinate system was established to represent the shooting position and orientation of each image.

[0022] Next, the shooting orientation information between images from different perspectives is correlated. For any two images, their relative orientation relationship, such as relative distance and relative angle, is calculated by comparing their shooting position coordinates and attitude parameters. This relative orientation relationship is stored as correlation information to construct the initial correlation relationship for multi-view images. For example, for two images taken from due east and due north, the distance between their shooting positions and the angle between their shooting directions are calculated, and this information is recorded in the initial correlation relationship.

[0023] Step S120: Based on the initial association relationship of the multi-view images, extract the association attributes of the corresponding feature regions in the images from different viewpoints, generate feature region association factors, and construct a set of cross-view dynamic matching constraint rules by combining the change rules of the feature region association factors.

[0024] In the multi-view remote sensing images of the aforementioned urban area, based on the initial association relationship of the multi-view images obtained in step S110, the extraction of feature region association attributes and the construction of a set of cross-view dynamic matching constraint rules begin.

[0025] Step S121: Select corresponding image pairs from the initial association relationship of the multi-view images. Each image pair includes a first-view image and a second-view image. Select multiple feature regions in the first-view image. Extract the contour attributes and texture attributes of each feature region based on the pixel distribution information of the image. The contour attributes include the boundary pixel arrangement pattern of the feature region, and the texture attributes include the pixel grayscale change pattern inside the feature region.

[0026] From the initial correlation of multi-view images, image pairs with obvious spatial correspondence are selected. For example, an image taken vertically from 500 meters above a city area is selected as the first-view image, and an image taken from 300 meters northeast of the city area at a 30-degree downward angle is selected as the second-view image. These two images constitute an image pair.

[0027] In the first-person view image, several representative feature regions are selected. These feature regions can be iconic buildings in the city, such as skyscrapers and large stadiums, or natural terrain areas with specific shapes, such as rivers and parks. For each selected feature region, contour and texture attributes are extracted based on the pixel distribution information of the image.

[0028] Step S1211: When extracting the contour attributes of the feature region in the first-view image, the boundary pixels of the feature region are sampled point by point, the coordinate information of each boundary pixel is recorded, and a boundary pixel coordinate sequence is formed. Based on the boundary pixel coordinate sequence, the connection angle between adjacent boundary pixels is calculated, the distribution characteristics of all connection angles are statistically analyzed, and the distribution characteristics are used as the core content of the contour attributes.

[0029] Taking a feature region of a tall building in a first-person view image as an example, the boundary pixels of this region are sampled point by point. Starting from the top left boundary pixel of the building, the coordinate information of each boundary pixel is recorded sequentially in a clockwise direction, forming a boundary pixel coordinate sequence. For example, the resulting coordinate sequence may be (x1, y1), (x2, y2), (x3, y3)...(xn, yn).

[0030] Based on this sequence of boundary pixel coordinates, the connection angle between adjacent boundary pixels is calculated. For two adjacent pixels (xi, yi) and (xi+1, yi+1), the connection angle is obtained by calculating the angle between the line connecting the two points and the horizontal direction. Then, the distribution characteristics of all these connection angles are statistically analyzed, such as the range of angles and the proportion of pixels in different angle intervals. These distribution characteristics are used as the core content of the contour attributes of this feature region.

[0031] Step S1212: When extracting the texture attributes of the feature region in the first-view image, multiple sampling regions are uniformly selected within the feature region. Each sampling region contains multiple pixels. The number of times the pixel gray value changes in each sampling region is calculated. The distribution of the number of times the gray value changes in all sampling regions is statistically analyzed. The distribution result is used as the core content of the texture attribute.

[0032] Within the feature area of ​​the aforementioned high-rise building, multiple sampling regions are uniformly selected. For example, the feature area is divided into a 10×10 grid, with each grid serving as a sampling region, and each sampling region containing multiple pixels. For each sampling region, the number of changes in the grayscale value of its pixels is calculated. Specifically, starting from the top-left pixel of the sampling region, the grayscale values ​​of adjacent pixels are compared sequentially from left to right and from top to bottom. When the difference in grayscale value between adjacent pixels exceeds a preset threshold, it is recorded as a grayscale value change. The number of grayscale value changes within each sampling region is counted, and then the distribution of the number of grayscale value changes across all sampling regions is analyzed, such as the proportion of sampling regions within different intervals of change counts. This distribution result is used as the core content of the texture attributes.

[0033] Step S122: Find the feature regions in the second view image that correspond to the multiple feature regions selected in the first view image respectively. Extract the contour attributes and texture attributes of the feature regions in the second view image corresponding to each feature region in the first view image in the same way as extracting the feature region attributes in the first view image. Compare the difference in contour attributes between each feature region in the first view image and the feature region in the second view image corresponding to that feature region. Calculate the overlap ratio of the contour attributes and use the overlap ratio as the first association parameter.

[0034] In the second-view image, based on the position and shape information of the feature regions in the first-view image, and combined with the relative orientation information in the initial correlation relationship of the multi-view images, the feature regions corresponding to each feature region in the first-view image are found. For example, the feature region of tall buildings in the first-view image may appear in different shapes and positions in the second-view image due to different shooting angles, but by comparing their general outlines and surrounding environmental features, the corresponding feature regions can be found.

[0035] Using the same method as steps S1211 and S1212, the contour and texture attributes of the corresponding feature regions in the second-view image are extracted. Then, the differences in the contour attributes of the corresponding feature regions in the first-view image and the second-view image are compared.

[0036] Specifically, the boundary pixel coordinate sequences of two feature regions are matched to find the number of overlapping pixels. The ratio of the number of overlapping pixels to the average of the total number of boundary pixels of the two feature regions is calculated to obtain the overlap ratio of the contour attribute, which is used as the first association parameter. For example, if the total number of boundary pixels of the feature region in the first-view image is 100, and the total number of boundary pixels of the corresponding feature region in the second-view image is 120, and the number of overlapping pixels is 80, then the overlap ratio of the contour attribute is 80 / ((100+120) / 2), and the calculated result is the first association parameter.

[0037] Step S123: Compare the texture attribute differences between each feature region in the first-view image and the feature region in the corresponding second-view image, calculate the similarity ratio of the texture attributes, use the similarity ratio as the second association parameter, combine the first association parameter and the second association parameter to form a feature region association factor, and generate a feature region association factor for each feature region in the first-view image and the feature region in the corresponding second-view image.

[0038] Compare the texture attribute differences of corresponding feature regions in the first-view and second-view images. For each sampled region, compare the number of grayscale value changes of corresponding sampled regions in the two feature regions. If the difference in the number of grayscale value changes between the two sampled regions is within a preset allowable range, the texture attributes of the two sampled regions are considered similar. Count the number of similar sampled regions, calculate the ratio of the number of similar sampled regions to the total number of sampled regions, obtain the texture attribute similarity ratio, and use this similarity ratio as the second correlation parameter.

[0039] The first correlation parameter obtained in step S122 and the second correlation parameter obtained in this step are combined to form a feature region correlation factor. For example, if the first correlation parameter is 0.7 and the second correlation parameter is 0.65, then the feature region correlation factor can be expressed as (0.7, 0.65). Such a feature region correlation factor is generated for each feature region pair formed by each feature region in the first-view image and the corresponding feature region in the second-view image.

[0040] Step S124: Collect the feature region correlation factors of all feature region pairs, record the numerical changes of each feature region correlation factor under different shooting time periods, summarize the changing trends of all feature region correlation factors, and set a reasonable fluctuation range for each feature region correlation factor based on the changing trends of the feature region correlation factors, using the fluctuation range as a basic constraint condition.

[0041] Feature region correlation factors were collected for all feature region pairs in the aforementioned multi-view images of the urban area. Since the acquired remote sensing images may have been taken at different times, such as different seasons or time periods, the numerical changes of each feature region correlation factor under different shooting times were recorded. For example, for the same set of feature region pairs, the correlation factor in images taken in summer may differ from that in images taken in winter, because seasonal changes may lead to variations in factors such as vegetation cover and lighting conditions, thus affecting the attributes of the feature regions.

[0042] By analyzing the correlation factors of feature regions from a large number of shooting periods, their changing trends were summarized. For example, some feature region correlation factors may show a gradual upward or downward trend over time, while others may fluctuate within a certain range. Based on these trends, a reasonable fluctuation range was set for each feature region correlation factor. For example, a narrower fluctuation range was set for feature region correlation factors with relatively stable trends, and a wider fluctuation range was set for those with more drastic changes. This fluctuation range was used as a basic constraint.

[0043] Step S125: Extract environmental impact parameters when shooting images from different perspectives. The environmental impact parameters include information on changes in illumination and atmospheric transmission during shooting. Convert the environmental impact parameters into constraint adjustment coefficients. Combine the basic constraint conditions with the constraint adjustment coefficients. By adjusting the range boundary of the basic constraint conditions, dynamic constraint rules that change with environmental parameters are formed.

[0044] Extract environmental impact parameters from images taken at different perspectives. Information on changes in illumination can be indirectly reflected through parameters such as image brightness and contrast; for example, stronger lighting during shooting may result in a higher overall image brightness. Atmospheric transmittance information can be assessed by analyzing the sharpness of distant objects in the image; lower atmospheric transmittance leads to poorer sharpness of distant objects.

[0045] The aforementioned environmental impact parameters are converted into constraint adjustment coefficients. For example, when the light intensity increases, it may cause changes in the texture attributes of the feature region. In this case, the light change information can be converted into a constraint adjustment coefficient greater than 1 to expand the fluctuation range of the basic constraint conditions. When the atmospheric transmittance decreases, it may affect the extraction accuracy of the contour attributes. In this case, the atmospheric transmittance information can be converted into a constraint adjustment coefficient less than 1 to narrow the fluctuation range of the basic constraint conditions.

[0046] By combining basic constraints with constraint adjustment coefficients, dynamic constraint rules are formed by adjusting the range boundaries of the basic constraints. For example, if the fluctuation range of the basic constraints is (0.6, 0.8), when the constraint adjustment coefficient is 1.1, the new fluctuation range becomes (0.6 × 1.1, 0.8 × 1.1).

[0047] Step S126: Collect all dynamic constraint rules, classify them according to the type of feature region, form a subset of constraint rules covering different types of feature regions, combine each subset of constraint rules with the changing trend of the feature region association factors, and construct a cross-perspective dynamic matching constraint rule set that includes constraint triggering conditions, constraint adjustment methods, and constraint scope.

[0048] Collect all the dynamic constraint rules mentioned above and classify them according to the type of feature region. For example, feature regions can be divided into different types such as building type, natural terrain type, and road type. For each type of feature region, form a corresponding subset of constraint rules, which cover the dynamic constraint rules for different types of feature regions.

[0049] The constraints are combined with the changing trends of the feature region association factors. For example, for building-type feature regions, the changing trends of their feature region association factors may be less affected by seasonal changes but more affected by lighting conditions. Therefore, when constructing constraint rules, the focus should be on adjusting constraints related to lighting changes.

[0050] Construct a set of cross-view dynamic matching constraint rules, including constraint triggering conditions, constraint adjustment methods, and constraint scope. Constraint triggering conditions can be environmental influence parameters exceeding a certain threshold, or the values ​​of feature region correlation factors exceeding the fluctuation range of the basic constraint conditions. Constraint adjustment methods refer to the specific methods for adjusting the boundary of the constraint condition range based on constraint adjustment coefficients. The constraint scope clarifies which types of feature regions the rule applies to and under which shooting conditions.

[0051] Step S127: Perform an association check on multiple rules in the cross-view dynamic matching constraint rule set, identify and handle conflicts between rules to enhance the coordination of multiple rules working together in the feature region matching process; based on the historical matching result data of the feature region, adjust the order in which each rule in the cross-view dynamic matching constraint rule set is applied to form a rule priority sequence, wherein the rule at the beginning of the sequence is applied first in the matching operation; store the cross-view dynamic matching constraint rule set that has completed the association check and determined the rule priority sequence to the rule management module.

[0052] Perform correlation checks on multiple rules in the cross-view dynamic matching constraint rule set. For example, check if there are two rules that contradict each other in their constraint adjustment methods for the same feature region under the same conditions. If conflicts exist, the rules need to be adjusted and optimized, such as modifying the constraint triggering condition or adjustment method of one of the rules, to eliminate the conflict and enhance the coordination of multiple rules working together in the feature region matching process.

[0053] Based on historical matching results of feature regions in multi-view imagery of the urban area, the effectiveness and applicability of different rules in the matching process are analyzed. For example, some rules are more effective in improving matching accuracy, while others perform better in accelerating matching speed. Based on these analysis results, the order in which the rules in the cross-view dynamic matching constraint rule set are applied is adjusted to form a rule priority sequence. In subsequent feature region matching operations, the rules that appear earlier in the sequence are applied first.

[0054] The set of cross-view dynamic matching constraint rules that has completed the correlation check and determined the rule priority sequence is stored in the rule management module so that it can be called in subsequent steps such as coordinate mapping.

[0055] Step S128: When extracting the contour attributes of the feature region in the first-view image, the boundary pixels of the feature region are sampled point by point, the coordinate information of each boundary pixel is recorded, and a boundary pixel coordinate sequence is formed. Based on the boundary pixel coordinate sequence, the connection angle between adjacent boundary pixels is calculated, the distribution characteristics of all connection angles are statistically analyzed, and the distribution characteristics are used as the core content of the contour attributes.

[0056] This step is the same as step S1211, and is a further emphasis and refinement of step S1211. In actual operation, it can be performed according to the method of step S1211.

[0057] Step S129: When extracting the texture attributes of the feature region in the first-view image, multiple sampling regions are uniformly selected within the feature region. Each sampling region contains multiple pixels. The number of times the pixel gray value changes in each sampling region is calculated. The distribution of the number of times the gray value changes in all sampling regions is statistically analyzed. The distribution result is used as the core content of the texture attribute.

[0058] This step is the same as step S1212, and is a further emphasis and refinement of step S1212. In actual operation, it can be performed according to the method of step S1212.

[0059] Step S1210: When comparing the contour attributes of a feature region in the first-view image with the corresponding feature region in the second-view image, the boundary pixel coordinate sequences of the two are matched, the coordinate deviation of the corresponding boundary pixels is calculated, and the overlap ratio of the contour attributes is calculated based on the coordinate deviation. When comparing the texture attributes of a feature region in the first-view image with the corresponding feature region in the second-view image, the sampling areas of the two are matched one-to-one, the deviation of the number of gray value changes in the corresponding sampling areas is calculated, and the similarity ratio of the texture attributes is calculated based on the deviation.

[0060] When comparing contour attributes, the boundary pixel coordinate sequences of corresponding feature regions in the first-view and second-view images are precisely matched. For each pixel in the sequence, its coordinate deviation in the two sequences is calculated, such as horizontal and vertical deviation. Based on these coordinate deviations, it is determined whether a pixel is an overlapping pixel. For example, when the coordinate deviation is less than a preset threshold, the pixel is considered to be overlapping. Then, the overlap ratio of contour attributes is calculated according to the method in step S122.

[0061] When comparing texture attributes, the sampling areas of two feature regions are mapped one-to-one. For example, the sampling area in row i and column j of a feature region in the first-view image corresponds to the sampling area in row i and column j of a feature region in the second-view image. The deviation of the number of grayscale value changes in the corresponding sampling areas is calculated. When the deviation is less than a preset threshold, the texture attributes of the sampling areas are considered similar. The number of similar sampling areas is counted, and the similarity ratio of texture attributes is calculated.

[0062] Step S1211: When recording the changes in the correlation factors of the feature region, arrange the correlation factors of the feature region in different time periods according to the shooting time sequence, draw the trend curve of the correlation factors of the feature region changing with time, and when setting a reasonable fluctuation range of the correlation factors of the feature region, analyze the stable segment in the trend curve, and use the maximum and minimum values ​​of the correlation factors of the feature region in the stable segment as the initial boundary of the fluctuation range.

[0063] The feature region correlation factors of the same feature region pairs acquired at different time periods are arranged chronologically according to the shooting time. For example, the correlation factors of the same feature region pairs in images taken each month of the year are arranged in monthly order. Based on these arranged data, a trend curve of the feature region correlation factors changing over time is plotted.

[0064] Analyze the trend curve and identify the stable segment. The stable segment refers to a period where the values ​​of the correlation factors in the characteristic region change relatively little and fluctuate smoothly. For example, in periods without extreme weather or significant seasonal variations, the correlation factors in the characteristic region may be in a stable segment. Use the maximum and minimum values ​​of the correlation factors in the characteristic region within the stable segment as the initial boundaries of the fluctuation range. For example, if the maximum value in the stable segment is 0.85 and the minimum value is 0.65, then the initial fluctuation range is (0.65, 0.85).

[0065] Step S1212: When converting environmental impact parameters into constraint adjustment coefficients, the illumination adjustment coefficient is determined based on illumination change information. The illumination adjustment coefficient increases synchronously with the increase of illumination intensity. The transmission adjustment coefficient is determined based on atmospheric transmission information. The transmission adjustment coefficient decreases synchronously with the increase of transmittance. The illumination adjustment coefficient and the transmission adjustment coefficient are combined to obtain a comprehensive constraint adjustment coefficient, which is used to adjust the boundary of the fluctuation range of the correlation factor in the characteristic region.

[0066] The illumination adjustment factor is determined based on illumination change information. The relative change in illumination intensity is assessed by analyzing parameters such as image brightness. When illumination intensity increases, the illumination adjustment factor increases accordingly. For example, if the illumination intensity becomes 1.2 times its original value, the illumination adjustment factor can be set to 1.2.

[0067] The transmission adjustment factor is determined based on atmospheric transmission information. The relative change in atmospheric transmittance is assessed by analyzing the sharpness of distant objects in the image. When atmospheric transmittance increases, the transmission adjustment factor decreases accordingly. For example, when atmospheric transmittance becomes 1.1 times its original value, the transmission adjustment factor can be set to 0.9.

[0068] The illumination adjustment coefficient and the transmission adjustment coefficient are combined, for example, by multiplication, to obtain the comprehensive constraint adjustment coefficient. Comprehensive constraint adjustment coefficient = Illumination adjustment coefficient × Transmission adjustment coefficient. This comprehensive constraint adjustment coefficient is used to adjust the boundaries of the fluctuation range of the correlation factor in the feature region, such as by multiplying the upper and lower boundaries of the initial fluctuation range by the comprehensive constraint adjustment coefficient.

[0069] Step S1213: When constructing dynamic constraint rules, when the change of environmental influence parameters exceeds the preset range, the constraint adjustment operation is triggered. The boundary range of the constraint conditions is adjusted according to the preset calculation method. When classifying dynamic constraint rules, they are classified according to the contour complexity of the feature region. Feature regions with different contour complexities correspond to different constraint rules.

[0070] When constructing dynamic constraint rules, preset ranges are set for environmental impact parameters. For example, the preset range for variation in light intensity is ±10%, and the preset range for variation in atmospheric transmittance is ±5%. When the variation of environmental impact parameters exceeds these preset ranges, constraint adjustment operations are triggered.

[0071] The boundary range of the constraint conditions is adjusted according to a preset calculation method, which is a method of adjusting the boundary of the fluctuation range based on the comprehensive constraint adjustment coefficient. For example, when the comprehensive constraint adjustment coefficient is 1.1, the boundary range (a, b) of the constraint conditions is adjusted to (a×1.1, b×1.1).

[0072] When classifying dynamic constraint rules, they are categorized according to the contour complexity of the feature regions. Contour complexity can be measured by metrics such as the number of boundary pixels and the frequency of changes in connection angles. For example, contour complexity can be divided into three levels: simple, medium, and complex, each corresponding to different constraint rules. For feature regions with high contour complexity, the constraint rules may be more stringent to ensure matching accuracy.

[0073] Step S1214: Combine the classified constraint rules with the changing trend of the feature region association factors, so that the constraint rule set can dynamically adjust the constraint strength according to the changes in the feature region association factors and environmental parameters. Test the adjusted constraint rule set, verify the effectiveness of the rule set through feature region matching operation, further optimize the rule content based on the verification results, and determine the optimized constraint rule set as the final cross-view dynamic matching constraint rule set.

[0074] The categorized constraint rules are combined with the changing trends of the correlation factors in the feature regions. For feature regions with different contour complexities, the changing trends of their correlation factors are analyzed, and the strength of the constraint rules is adjusted according to the characteristics of the trends. For example, for feature regions with high contour complexity and unstable trends in the changing trends of their correlation factors, the adjustment frequency and magnitude of the constraint rules are increased to dynamically adapt to their changes.

[0075] The adjusted set of constraint rules is tested. A certain number of feature region pairs are selected, and cross-view feature region matching is performed using this set of constraint rules. The effectiveness of the rule set is verified by comparing the matching results with the actual spatial correspondence. For example, the proportion of correctly matched feature region pairs to the total number of matches is counted. If the proportion reaches a preset threshold, such as 90%, the rule set is considered effective.

[0076] Based on the verification results, the rules were further optimized. For feature region types with a high number of matching errors, the reasons were analyzed and the corresponding constraint rules were adjusted, such as modifying the constraint triggering conditions and adjusting the calculation method of the constraint adjustment coefficients. After multiple tests and optimizations, the final set of constraint rules was determined as the cross-view dynamic matching constraint rule set.

[0077] Step S130: Using the cross-view dynamic matching constraint rule set, coordinate mapping operation is performed on the feature regions in images from different viewpoints to convert the two-dimensional coordinates of each feature region into three-dimensional spatial coordinates, forming an initial set of three-dimensional point cloud units.

[0078] In the multi-view remote sensing image processing of the aforementioned urban area, based on the cross-view dynamic matching constraint rule set constructed in step S120, coordinate mapping operations are performed on the feature regions in the images from different viewpoints to convert the two-dimensional coordinates into three-dimensional spatial coordinates.

[0079] Step S131: Extract constraint rules for different types of feature regions from the cross-view dynamic matching constraint rule set, determine the coordinate mapping parameters corresponding to each type of feature region, select a feature region in the first view image, determine the type of the feature region, and find the constraint rules and coordinate mapping parameters corresponding to the type of feature region.

[0080] From the set of cross-view dynamic matching constraint rules, constraint rules for different types of feature regions in urban areas are extracted, such as building type, road type, and natural terrain type. For each type of feature region, the corresponding coordinate mapping parameters are determined according to its imaging characteristics and spatial geometric relationship in multi-view images. The coordinate mapping parameters include the image's interior orientation elements (such as focal length, principal point coordinates, etc.) and exterior orientation elements (such as shooting position coordinates, attitude parameters, etc.).

[0081] Select a feature region in the first-view image, for example, a feature region of the building type. Determine that the type of the feature region is building, and then find the constraint rules and coordinate mapping parameters corresponding to the building type feature region from the cross-view dynamic matching constraint rule set.

[0082] Step S132: Extract the two-dimensional coordinates of the feature region in the first-view image, and convert the horizontal coordinate values ​​in the two-dimensional coordinates into horizontal axis coordinate values ​​in three-dimensional space according to the coordinate mapping parameters corresponding to the feature region of the corresponding type. The two-dimensional coordinates include horizontal coordinate values ​​and vertical coordinate values.

[0083] Extract the two-dimensional coordinates of the building's feature area in the first-view image. The two-dimensional coordinates take the upper left corner of the image as the origin, the horizontal direction to the right as the horizontal direction, and the vertical direction downward as the vertical direction, and include the horizontal coordinate values ​​and the vertical coordinate values.

[0084] Based on the coordinate mapping parameters corresponding to the building-type feature areas, the horizontal coordinate values ​​in two-dimensional coordinates are converted into horizontal axis coordinate values ​​in three-dimensional space. The specific conversion process is based on the collinearity equation principle in photogrammetry. Through the interior and exterior orientation elements in the coordinate mapping parameters, a mathematical relationship is established between the two-dimensional coordinates of the image and the three-dimensional spatial coordinates, thereby achieving the conversion of the horizontal coordinate values.

[0085] Step S133: Convert the vertical coordinate values ​​in the two-dimensional coordinate system into vertical axis coordinate values ​​in the three-dimensional space, and combine them with the shooting height parameters of the first-view image to determine the height axis coordinate values ​​of the feature region in the three-dimensional space.

[0086] Using a method similar to step S132, based on the coordinate mapping parameters and the collinearity equation principle, the vertical coordinate values ​​in the two-dimensional coordinate system are converted into vertical axis coordinate values ​​in the three-dimensional space.

[0087] For example, step S1331: obtain the shooting height parameter when the first-view image is captured. The shooting height parameter includes the vertical distance between the image sensor and the ground. Extract the gray value of the feature area in the first-view image. The gray value reflects the reflectivity of the ground object corresponding to the feature area.

[0088] The shooting height parameter at the time of first-person view image capture can be obtained from the image's metadata, which includes the vertical distance between the image sensor and the ground. For example, a shooting height parameter of 500 meters indicates that the image sensor was 500 meters above the ground at the time of capture.

[0089] The grayscale values ​​of the building's feature areas are extracted from the first-person view image. The magnitude of the grayscale value reflects the light reflection characteristics of the ground objects corresponding to that feature area. Different ground objects, such as the building's wall materials and colors, have different reflective characteristics, thus exhibiting different grayscale values.

[0090] Step S1332: Based on the grayscale value of the feature region, determine the height of the ground object corresponding to the feature region through the pre-calibrated grayscale-height mapping relationship, and subtract the height of the ground object corresponding to the feature region from the vertical distance in the shooting height parameter to obtain the initial height direction axis coordinate value of the feature region in three-dimensional space.

[0091] The pre-calibrated grayscale-height mapping relationship is obtained by collecting a large number of grayscale value samples from ground objects of known heights and performing data analysis and fitting. Based on the grayscale values ​​of the building's feature area, the pre-calibrated grayscale-height mapping relationship is queried to determine the height of the ground object corresponding to that feature area.

[0092] Subtracting the height of the ground object from the vertical distance in the shooting height parameter yields the initial height axis coordinate value of the feature region in three-dimensional space. For example, if the vertical distance in the shooting height parameter is 500 meters and the height of the ground object is 100 meters, then the initial height axis coordinate value is 500 - 100 = 400 meters.

[0093] Step S1333: Obtain the tilt angle parameters when the first-view image is captured. The tilt angle parameters include the angle between the image sensor and the vertical direction. Based on the tilt angle parameters, calculate the correction amount of the tilt angle to the height axis coordinate value through the geometric projection model. The correction amount increases synchronously when the tilt angle value increases.

[0094] The tilt angle parameter during first-person view image capture is obtained. This tilt angle parameter can also be obtained from the image's metadata, including the angle between the image sensor and the vertical direction. Based on the tilt angle parameter, a geometric projection model is used to calculate the correction amount of the tilt angle to the height axis coordinate value.

[0095] The geometric projection model considers the influence of the image sensor's tilt angle on the imaged position of ground objects. As the tilt angle increases, the imaged position of ground objects in the image shifts, leading to errors in the calculation of the height axis coordinates. This model can calculate the amount of correction needed to the initial height axis coordinates, and the correction amount increases proportionally with the tilt angle.

[0096] Step S1334: Add the initial height direction axis coordinate value to the correction amount to obtain the adjusted height direction axis coordinate value, and extract the position information of the feature region in the first view image. The position information includes the distance between the feature region and the center of the image.

[0097] The initial altitude axis coordinate value obtained in step S1332 is added to the correction amount obtained in step S1333 to obtain the adjusted altitude axis coordinate value. For example, if the initial altitude axis coordinate value is 400 meters and the correction amount is 5 meters, then the adjusted altitude axis coordinate value is 405 meters.

[0098] The location information of the feature region in the first-view image is extracted. This location information can be represented by the distance between the center coordinates of the feature region and the center coordinates of the image. For example, if the center coordinates of the feature region are (x, y) and the center coordinates of the image are (cx, cy), then the distance between the feature region and the image center can be obtained by calculating the Euclidean distance between the two points.

[0099] Step S1335: Based on the location information, calculate the secondary correction amount of the position relative to the height axis coordinate value through the image distortion correction model. The secondary correction amount increases synchronously when the distance between the feature region and the image center increases. Add the adjusted height axis coordinate value to the secondary correction amount to obtain the final height axis coordinate value.

[0100] The image distortion correction model is used to correct the impact of image distortion caused by factors such as lens optical characteristics on coordinate calculations. Based on the distance between the feature region and the image center, the image distortion correction model calculates the secondary correction amount of the position with respect to the height axis coordinate values.

[0101] Generally, the greater the distance between the feature region and the image center, the greater the impact of image distortion; therefore, the secondary correction value also increases accordingly. The adjusted height axis coordinate value is added to the secondary correction value to obtain the final height axis coordinate value. For example, if the adjusted height axis coordinate value is 405 meters and the secondary correction value is 3 meters, then the final height axis coordinate value is 408 meters.

[0102] Step S1336: Record the correspondence between the final height axis coordinate values ​​and the shooting height parameters, tilt angle parameters, and the location information of the feature area. Compare the height axis coordinate values ​​of different feature areas in the same viewpoint image and analyze whether the distribution of height axis coordinate values ​​meets the preset judgment criteria of spatial features under the corresponding viewpoint.

[0103] Record the correspondence between the final height axis coordinate values ​​and the shooting height parameters, tilt angle parameters, and feature area location information, so as to facilitate subsequent data tracing and analysis.

[0104] The height coordinates of different feature areas in an image from the same viewpoint are compared. For example, the height values ​​of adjacent building feature areas are compared, and their distribution is analyzed to see if it conforms to a preset judgment standard for spatial characteristics under that viewpoint. The preset judgment standard may be the approximate range of building heights in different areas obtained from urban planning data, or the height variation pattern of natural terrain, etc.

[0105] Step S1337: If there are cases where the height axis coordinate values ​​do not meet the preset judgment criteria for spatial features under this viewpoint, then re-examine the calculation process, adjust the influence coefficient and correction coefficient, so that the height axis coordinate values ​​of each feature region can accurately reflect the vertical position of the feature region in three-dimensional space.

[0106] If the analysis in step S1336 finds that the height direction axis coordinate values ​​of certain feature areas do not meet the preset judgment criteria, for example, if the height calculation result of a certain building is significantly beyond the reasonable height range of buildings in that area, then it is necessary to re-examine the entire process from obtaining the gray value to calculating the secondary correction amount.

[0107] Adjust the influence and correction coefficients in the calculation process, such as the fitting coefficient in the gray-height mapping relationship, the parameters in the geometric projection model, and the coefficients in the image distortion correction model, and recalculate the height axis coordinate values ​​until the height axis coordinate values ​​of each feature region can accurately reflect its vertical position in three-dimensional space.

[0108] Step S1338: Combine the corrected height axis coordinate values ​​with the horizontal axis coordinate values ​​and the vertical axis coordinate values ​​to form the three-dimensional spatial coordinates of the feature region, and store the three-dimensional spatial coordinates in association with the feature region's identification information.

[0109] The corrected and verified height axis coordinate values ​​are combined with the horizontal axis coordinate values ​​and vertical axis coordinate values ​​obtained in step S132 to form the three-dimensional spatial coordinates (X, Y, Z) of the feature region, where X is the horizontal axis coordinate value, Y is the vertical axis coordinate value, and Z is the height axis coordinate value.

[0110] The three-dimensional spatial coordinates are associated with and stored with the identification information of the feature region. The identification information can be the feature region's number, type, etc., so as to facilitate the subsequent management and application of the three-dimensional point cloud data.

[0111] Step S134: Combine the horizontal axis coordinate values, vertical axis coordinate values, and height axis coordinate values ​​to form the three-dimensional spatial coordinates of the feature region, which serves as an initial unit of the three-dimensional point cloud. Select the next feature region in the first-view image and repeat the above coordinate transformation steps to generate the initial unit of the three-dimensional point cloud corresponding to the next feature region.

[0112] In the first-view image of the aforementioned urban area, after calculating the 3D spatial coordinates of a building feature region, the horizontal, vertical, and height axis coordinate values ​​are combined to form the 3D spatial coordinates of that feature region. These 3D spatial coordinates constitute a 3D point cloud initial unit. Next, the first-view image is used to select the next feature region, such as a road intersection. Following steps S131 to S133, the type of the road intersection feature region is determined to be a road, the corresponding constraint rules and coordinate mapping parameters are found, its 2D coordinates are extracted and converted into 3D spatial coordinates, generating the 3D point cloud initial unit corresponding to the road intersection feature region. This process is repeated until all selected feature regions in the first-view image have completed the generation of 3D point cloud initial units.

[0113] Step S135: After completing the coordinate transformation of all feature regions in the first-view image, perform the same operation on the feature regions in the second-view image, extract the two-dimensional coordinates of each feature region in the second-view image, and convert the two-dimensional coordinates into three-dimensional spatial coordinates based on the constraint rules and coordinate mapping parameters corresponding to the type of each feature region, and generate the initial three-dimensional point cloud unit corresponding to each feature region.

[0114] After coordinate transformation and initial 3D point cloud units are generated for all feature regions in the first-view image, the process shifts to the second-view image. For each feature region in the second-view image, such as buildings, roads, and parks in an image taken from the northeast, the same operational procedure as for the first-view image is performed. First, the type of each feature region is determined; for example, if a feature region is a park-type natural terrain, then the constraint rules and coordinate mapping parameters corresponding to the park-type feature region are extracted from the cross-view dynamic matching constraint rule set. The two-dimensional coordinates of the park feature region in the second-view image are extracted, and based on the coordinate mapping parameters, the two-dimensional coordinates are converted into three-dimensional spatial coordinates to generate the corresponding initial 3D point cloud units. This process must strictly follow the same coordinate transformation method as the first-view image to ensure the consistency of the initial 3D point cloud units generated from different view images in the spatial coordinate system.

[0115] Step S136: Repeat the above process until all feature regions in all viewpoint images have completed coordinate transformation, generate the initial 3D point cloud unit corresponding to each feature region, collect all generated initial 3D point cloud units, and classify them according to the type of feature region.

[0116] For other multi-view images of the urban area, such as images taken from different angles like northwest and southeast, steps S131 to S135 are repeated to perform coordinate transformation on each feature region in the image, generating corresponding initial 3D point cloud units. After all view images have been processed, all generated initial 3D point cloud units are collected. These initial 3D point cloud units are then categorized according to the type of feature region, such as buildings, roads, natural terrain (parks, rivers, etc.), and vegetation. For example, initial 3D point cloud units generated from all building feature regions are grouped into a building set, those generated from road feature regions are grouped into a road set, and so on, preparing for the formation of subsequent optimized 3D point cloud unit sets.

[0117] Step S137: Organize the coordinates of the initial 3D point cloud units corresponding to the same type of feature region to keep the coordinate format consistent, and combine the categorized and organized initial 3D point cloud units to form a set of initial 3D point cloud units.

[0118] For each type of feature region, the coordinate format of the initial 3D point cloud units is standardized. The coordinate representation of each initial 3D point cloud unit is checked to ensure that the units of the horizontal, vertical, and height axis coordinates are consistent (e.g., all in meters), the number of decimal places is uniform (e.g., three decimal places), and the origin and axis directions conform to the preset 3D spatial coordinate system definition. For initial 3D point cloud units that do not meet the format requirements, the coordinate format is converted and adjusted. All the categorized and standardized initial 3D point cloud units are combined to form a set of initial 3D point cloud units containing the 3D coordinate information of all feature regions within the urban area.

[0119] Step S138: Perform coordinate verification on each 3D point cloud initial unit in the 3D point cloud initial unit set. By comparing the 3D coordinates of the 3D point cloud initial units corresponding to the feature regions adjacent to the feature regions corresponding to the 3D point cloud initial units, correct the 3D point cloud initial units whose coordinate deviation exceeds the preset allowable deviation range, and store the corrected 3D point cloud initial unit set in the initial unit database.

[0120] To ensure the accuracy of the initial 3D point cloud unit set, coordinate verification is performed on each initial 3D point cloud unit in the set. For each initial 3D point cloud unit, the initial 3D point cloud units corresponding to its adjacent feature regions in the image are found. For example, the initial 3D point cloud unit of a building feature region may have adjacent feature regions such as another building or road nearby. The initial 3D point cloud units corresponding to these adjacent feature regions are found. The spatial distance between the initial 3D point cloud unit to be verified and its adjacent initial 3D point cloud units is calculated. Combined with the actual spatial layout knowledge of the urban area, it is determined whether the distance is within a reasonable range, thereby assessing whether there is a deviation in the coordinates of the initial 3D point cloud unit to be verified. If the coordinate deviation exceeds the preset allowable deviation range (e.g., the distance deviation exceeds 5 meters), the coordinate transformation process of the initial 3D point cloud unit is re-examined, the relevant parameters are adjusted, and the coordinates are recalculated until the deviation is within the allowable range. All verified and corrected initial 3D point cloud units are stored in the initial unit database. The database uses a structured storage method to record the coordinate information, the type of feature region to which each initial 3D point cloud unit belongs, and the corresponding image identifier, etc., for easy retrieval and management in subsequent steps.

[0121] Step S140: For the set of initial units of the three-dimensional point cloud, extract the spatial association attributes of each initial unit of the three-dimensional point cloud, divide the initial units of the three-dimensional point cloud into groups based on the attribute similarity, and perform progressive spatial position adjustment on each group of initial units of the three-dimensional point cloud to obtain the set of optimized units of the three-dimensional point cloud.

[0122] Based on the initial set of 3D point cloud units in the aforementioned urban area, the construction of an optimized set of 3D point cloud units begins. By extracting spatial correlation attributes, grouping, and adjusting positions, the spatial distribution accuracy of the 3D point cloud is improved.

[0123] Step S141: Extract the three-dimensional spatial coordinate information of each three-dimensional point cloud initial unit from the set of three-dimensional point cloud initial units. The three-dimensional spatial coordinate information includes the horizontal axis coordinate value, the vertical axis coordinate value, and the height axis coordinate value. Use the three-dimensional spatial coordinate information as the basic content of spatial association attributes.

[0124] A set of initial 3D point cloud units is read from the initial unit database. For each initial 3D point cloud unit in the set, its 3D spatial coordinate information is extracted. This 3D spatial coordinate information specifically includes the horizontal axis coordinate value (X), the vertical axis coordinate value (Y), and the height axis coordinate value (Z). These three coordinate values ​​precisely define the position of the initial unit in 3D space. These three coordinate values ​​are used as the basic content of spatial association attributes because they are the most basic elements describing the spatial position of the initial unit. Subsequent spatial distance calculations, similarity assessments, etc., are all based on this.

[0125] Step S142: Calculate the spatial distance between each initial 3D point cloud unit and all other initial 3D point cloud units. Use the spatial distance as a supplement to the spatial association attribute. Combine the 3D spatial coordinate information and spatial distance of each initial 3D point cloud unit to form the spatial association attribute.

[0126] For each 3D point cloud initial cell, it is necessary to calculate the spatial distance between it and all other 3D point cloud initial cells in the set of 3D point cloud initial cells.

[0127] Step S1421: Obtain the three-dimensional spatial coordinates of each three-dimensional point cloud initial unit. The three-dimensional spatial coordinates include the horizontal axis coordinate values, the vertical axis coordinate values, and the height axis coordinate values. Select the first three-dimensional point cloud initial unit as the target unit, and then select other three-dimensional point cloud initial units as comparison units in turn.

[0128] Obtain the 3D spatial coordinates of all elements in the current set of initial 3D point cloud elements. Each coordinate consists of a horizontal axis coordinate, a vertical axis coordinate, and a height axis coordinate. Select the first 3D point cloud initial element from the set as the target element, denoted as element A, with coordinates (X_A, Y_A, Z_A). Then, sequentially select other 3D point cloud initial elements from the set (excluding element A) as comparison elements, such as element B, element C, element D, etc., with coordinates (X_B, Y_B, Z_B), (X_C, Y_C, Z_C), (X_D, Y_D, Z_D), etc., respectively.

[0129] Step S1422: Calculate the coordinate difference between the target unit and the comparison unit in the horizontal axis direction, and take the absolute value of the difference as the horizontal axis deviation. Calculate the coordinate difference between the target unit and the comparison unit in the vertical axis direction, and take the absolute value of the coordinate difference as the vertical axis deviation. Calculate the coordinate difference between the target unit and the comparison unit in the height axis direction, and take the absolute value of the coordinate difference as the height axis deviation.

[0130] For target unit A and the current comparison unit B, calculate the coordinate difference between them in the horizontal axis direction, i.e., X_A minus X_B, and then take the absolute value of this difference to obtain the horizontal axis deviation ΔX_AB = |X_A - X_B|. Similarly, calculate the coordinate difference in the vertical axis direction, Y_A minus Y_B, and take the absolute value to obtain the vertical axis deviation ΔY_AB = |Y_A - Y_B|. Calculate the coordinate difference in the height axis direction, Z_A minus Z_B, and take the absolute value to obtain the height axis deviation ΔZ_AB = |Z_A - Z_B|.

[0131] Step S1423: Based on the horizontal axis deviation, vertical axis deviation and height axis deviation, the spatial distance between the target unit and the comparison unit is obtained by spatial distance calculation. The spatial distance between the target unit and all comparison units is recorded to form a list of spatial distances of the target unit.

[0132] Based on the horizontal axis deviation ΔX_AB, vertical axis deviation ΔY_AB, and height axis deviation ΔZ_AB obtained in step S1422, the spatial distance between target unit A and comparison unit B is calculated using the three-dimensional spatial distance calculation formula. The spatial distance is calculated by taking the square root of the sum of the squares of the horizontal axis deviation, vertical axis deviation, and height axis deviation, i.e., distance D_AB = √(ΔX_AB² + ΔY_AB² + ΔZ_AB²). Following this method, the spatial distances D_AC, D_AD, etc., between target unit A and all other comparison units (unit C, unit D, etc.) are calculated sequentially, and these distance values ​​are recorded to form a spatial distance list for target unit A. This list contains the spatial distance information between unit A and all other units in the set.

[0133] Step S1424: Combine the spatial distance list of the target unit with the three-dimensional spatial coordinate information of the target unit to form the spatial association attribute of the target unit. Select the next three-dimensional point cloud initial unit as the new target unit. Repeat the above steps of calculating spatial distance and constructing spatial association attribute until all three-dimensional point cloud initial units have completed spatial distance calculation and spatial association attribute construction.

[0134] The 3D spatial coordinates (X_A, Y_A, Z_A) of target unit A are combined with its spatial distance list (including D_AB, D_AC, D_AD, etc.) to form the spatial association attribute of target unit A. This spatial association attribute comprehensively describes the spatial location of unit A and its spatial proximity relationship with other surrounding units. After completing the construction of the spatial association attribute of target unit A, the next unit in the initial set of 3D point cloud units is selected as the new target unit, such as unit B. Steps S1421 to S1424 are repeated to calculate the spatial distance between unit B and all other units in the set (including unit A, unit C, unit D, etc.), construct the spatial distance list of unit B, and combine its 3D spatial coordinates to form the spatial association attribute. This process is repeated until the spatial distance calculation and spatial association attribute construction of all initial 3D point cloud units in the set are completed.

[0135] Step S1425: Sort the spatial distance list of each 3D point cloud initial unit in ascending order of spatial distance value, check whether the spatial distance list in each spatial association attribute is complete, and confirm that the spatial distance between the target unit and all other 3D point cloud initial units is recorded.

[0136] For each initial 3D point cloud cell, the spatial distance list is sorted according to the numerical value of the spatial distance, from smallest to largest. For example, the sorted spatial distance list for target cell A might be [D_AC (smallest), D_AB, D_AD, ...]. This sorted list facilitates quick searching of neighboring cells. Simultaneously, the completeness of the spatial distance list in each spatial association attribute is checked, confirming that the spatial distances between the target cell and all other initial 3D point cloud cells in the set have been accurately recorded without omissions. If any distances are missing, they must be recalculated and supplemented to ensure the completeness and accuracy of the spatial association attributes.

[0137] Step S1426: Standardize the format of coordinate information and spatial distance in the spatial association attribute, and store the standardized spatial association attribute in the specified data area.

[0138] The coordinate information and spatial distance in the spatial association attributes of all initial 3D point cloud units are formatted uniformly. For coordinate information, the units and decimal places of the coordinate values ​​along the horizontal, vertical, and height axes are ensured to be consistent; similarly, the units and decimal places of spatial distances are standardized. For example, both coordinate and distance values ​​are retained to three decimal places, with the unit being meters. The formatted spatial association attributes are then stored in a designated data area according to a preset data structure, such as a specific array in computer memory or a specific data table in a database, so that subsequent grouping operations based on attribute similarity can be efficiently read and processed.

[0139] Step S143: Calculate the spatial association attribute similarity between any two initial 3D point cloud units. The similarity calculation is based on the proximity of their 3D spatial coordinates and the similarity of their spatial distance lists. The proximity of the 3D spatial coordinates is evaluated by calculating the difference in coordinate values, and the similarity of the spatial distance lists is evaluated by calculating the consistency of the distance distribution. The evaluation results are combined into a spatial association attribute similarity. Initial 3D point cloud units with spatial association attribute similarity exceeding the attribute similarity threshold are grouped into the same group to form multiple groups of initial 3D point cloud units.

[0140] After constructing the spatial association attributes of all initial 3D point cloud units, the similarity of spatial association attributes between any two initial 3D point cloud units is calculated. The similarity calculation is performed from two aspects: the proximity of 3D spatial coordinates and the similarity of spatial distance lists. For the proximity of 3D spatial coordinates, the differences in the horizontal, vertical, and height axes of the two units are calculated. The magnitude of these differences is used to assess the proximity of the coordinates; the smaller the difference, the higher the proximity. For the similarity of spatial distance lists, the distribution of distance values ​​in the spatial distance lists of the two units is compared. For example, the similarity of the first few minimum distances in the lists is compared, and the overall distribution trend of the distance values ​​is consistent. This assesses the consistency of the distance distribution; the higher the consistency, the higher the similarity. The evaluation results of the proximity of coordinates and the evaluation results of the similarity of distance lists are combined according to a certain weight ratio to obtain the spatial association attribute similarity between the two units. A similarity threshold is set, and the initial 3D point cloud units with spatial correlation attributes that exceed the threshold are grouped into the same group. After pairwise comparison and grouping of all units, multiple initial 3D point cloud unit groups are finally formed, and the units in each group have high similarity in spatial correlation attributes.

[0141] Step S144: For each group of initial 3D point cloud units, select the initial 3D point cloud unit with the center position in the spatial coordinates of the group as the reference unit, calculate the coordinate deviation between each other initial 3D point cloud unit in the group and the reference unit, and determine the adjustment direction of each initial 3D point cloud unit based on the coordinate deviation, with the adjustment direction pointing towards the direction of the reference unit.

[0142] For each initial group of three-dimensional point cloud units formed, a reference unit needs to be selected so that the spatial positions of other units in the group can be adjusted later.

[0143] Step S1441: Extract the three-dimensional spatial coordinates of all three-dimensional point cloud initial units within each three-dimensional point cloud initial unit group, and separate the horizontal axis coordinate values, vertical axis coordinate values, and height axis coordinate values ​​of each three-dimensional point cloud initial unit.

[0144] Extract the 3D spatial coordinates of all initial 3D point cloud units within the initial unit group. For each initial unit, decompose its 3D spatial coordinates into three independent components: horizontal axis coordinate (X), vertical axis coordinate (Y), and height axis coordinate (Z). For example, if a group contains units G1, G2, G3, etc., extract their coordinate components X_G1, Y_G1, Z_G1, X_G2, Y_G2, Z_G2, X_G3, Y_G3, Z_G3, etc.

[0145] Step S1442: Calculate the average value of the horizontal axis coordinates of all initial 3D point cloud units in the group, and use it as the center coordinate of the horizontal axis in the group. Calculate the average value of the vertical axis coordinates of all initial 3D point cloud units in the group, and use it as the center coordinate of the vertical axis in the group. Calculate the average value of the height axis coordinates of all initial 3D point cloud units in the group, and use it as the center coordinate of the height axis in the group.

[0146] The horizontal axis coordinates (X) of all initial 3D point cloud elements within a group are summed, and then divided by the number of elements to obtain the average horizontal axis coordinate values. This average value is used as the horizontal axis center coordinate (X_center) of the group. Similarly, the average vertical axis coordinates (Y) are calculated and used as the vertical axis center coordinate (Y_center) of the group; the average height axis coordinates (Z) are calculated and used as the height axis center coordinate (Z_center) of the group. For example, if there are 3 elements in the group with horizontal axis coordinate values ​​X1, X2, and X3, then X_center = (X1 + X2 + X3) / 3, and Y_center and Z_center are calculated in the same way.

[0147] Step S1443: Combine the center coordinates of the horizontal axis, the center coordinates of the vertical axis, and the center coordinates of the height axis to form the center coordinates within the group. Calculate the deviation between the three-dimensional spatial coordinates of each initial unit of the three-dimensional point cloud within the group and the center coordinates within the group. The deviation includes the deviation of the horizontal axis, the deviation of the vertical axis, and the deviation of the height axis.

[0148] The horizontal axis center coordinates X_center, vertical axis center coordinates Y_center, and height axis center coordinates Z_center calculated in step S1442 are combined to form the group's center coordinates (X_center, Y_center, Z_center). For each initial 3D point cloud unit in the group, such as unit G1, the deviation of its 3D spatial coordinates (X_G1, Y_G1, Z_G1) from the group's center coordinates (X_center, Y_center, Z_center) is calculated. The horizontal axis deviation ΔX_G1 = X_G1 - X_center, the vertical axis deviation ΔY_G1 = Y_G1 - Y_center, and the height axis deviation ΔZ_G1 = Z_G1 - Z_center. Similarly, the deviations in each direction for all other units in the group are calculated.

[0149] Step S1444: Calculate the total deviation value of each 3D point cloud initial unit. The total deviation value is obtained by combining the deviation of the horizontal axis, the vertical axis, and the height axis. Arrange all 3D point cloud initial units in the group in ascending order of total deviation value.

[0150] For each initial 3D point cloud element within the group, such as element G1, the total deviation value is calculated based on its horizontal axis deviation ΔX_G1, vertical axis deviation ΔY_G1, and height axis deviation ΔZ_G1. The total deviation value can be calculated by taking the square root of the sum of the squares of the deviations in each direction, i.e., total deviation value D_total_G1 = √(ΔX_G1² + ΔY_G1² + ΔZ_G1²). This total deviation value reflects the overall degree of deviation of element G1's coordinates from the group's center coordinates. After calculating the total deviation value for all elements within the group, the elements are arranged in ascending order of their total deviation values. This arrangement reflects the distance of each element from the group's center coordinates.

[0151] Step S1445: Select the 3D point cloud initial unit with the smallest total deviation value as the candidate reference unit. If there are multiple 3D point cloud initial units with the same and smallest total deviation value, then further calculate the average spatial distance between all 3D point cloud initial units with the same and smallest total deviation value and other 3D point cloud initial units in the group.

[0152] In the list of cells within a group, arranged from smallest to largest total deviation value, the initial 3D point cloud cell with the smallest total deviation value is selected as the candidate reference cell. For example, if the cell with the smallest total deviation value in the list is G2, then G2 is the candidate reference cell. If multiple initial 3D point cloud cells have the same and smallest total deviation value, such as cells G2 and G3 having the same and smallest total deviation value, then further evaluation of the candidate cells is required. The evaluation method is to calculate the spatial distance between each candidate cell and all other initial 3D point cloud cells in the group, then sum these distances and divide by the number of cells in the group (minus itself) to obtain the average spatial distance of each candidate cell.

[0153] Step S1446: Select the 3D point cloud initial unit with the smallest average spatial distance value as the final reference unit, and record the 3D spatial coordinates of the reference unit as reference coordinates for subsequent adjustment of other 3D point cloud initial units.

[0154] For the candidate units with the same and smallest total deviation value in step S1445, compare their average spatial distances. The candidate unit with the smallest average spatial distance value indicates that its overall spatial distribution is closest to that of other units in the group, making it more suitable as the reference unit. Select this candidate unit as the final reference unit. For example, if the average spatial distance of unit G2 is less than that of G3, then G2 is selected as the reference unit. Record the three-dimensional spatial coordinates (X_G2, Y_G2, Z_G2) of the reference unit G2. These three-dimensional spatial coordinates will be used as reference coordinates for subsequent adjustments to the initial spatial positions of other three-dimensional point cloud units in the group.

[0155] Step S1447: Associate and store the coordinate information of the reference unit with the coordinate information of other 3D point cloud initial units in the group, analyze the position of the reference unit in the group, so that the reference unit can represent the overall spatial distribution trend of the 3D point cloud initial units in the group.

[0156] The 3D spatial coordinates of the reference unit are associated and stored with the coordinates of all other initial 3D point cloud units within the group using unit identifiers. For example, a field is created in the database table linking the reference unit ID with other unit IDs. Simultaneously, the spatial location of the reference unit within the group is analyzed. For instance, it is determined whether the reference unit is located near the geometric center of the group's unit distribution and whether it reflects the overall distribution trend of the group's units in the horizontal, vertical, and height directions. This analysis ensures the representativeness of the reference unit, enabling subsequent adjustments to effectively optimize the spatial distribution of units within the group.

[0157] Step S1448: Mark the selected reference unit to avoid the reference unit being incorrectly adjusted during subsequent adjustment. Store the mark information of the reference unit and the coordinate information together in the unit management module.

[0158] To prevent accidental modification of the reference unit's coordinates during subsequent progressive spatial position adjustments, the selected reference unit is specially marked. For example, a "Reference Unit Marker" field is set in the unit attributes, with a value of "Yes" or "1". This mark information, along with the reference unit's 3D spatial coordinate information and unit identifier, is stored in the unit management module. The unit management module is responsible for the unified management and maintenance of information for all initial 3D point cloud units, ensuring the stability of the reference units during the adjustment process.

[0159] Step S145: Determine the adjustment range of each 3D point cloud initial unit based on the magnitude of the coordinate deviation. When the coordinate deviation value increases, the adjustment range value increases synchronously. According to the determined adjustment direction and adjustment range, perform the first spatial position adjustment on each 3D point cloud initial unit in the group to obtain the first adjusted 3D point cloud initial unit.

[0160] For each initial 3D point cloud unit within the group, excluding the reference unit, such as unit G1, its adjustment direction has been determined in step S144 to be pointing towards the reference unit. The adjustment magnitude is determined based on the coordinate deviation between the unit and the reference unit. The larger the coordinate deviation, the farther the unit deviates from the reference unit, and the larger the adjustment magnitude required; that is, the adjustment magnitude is positively correlated with the coordinate deviation. For example, the adjustment magnitude can be set as a certain proportion of the coordinate deviation, such as adjustment magnitude = coordinate deviation × adjustment coefficient, where the adjustment coefficient is a positive number less than 1, and the specific value can be set according to the actual optimization requirements. According to the determined adjustment direction (pointing towards the reference unit) and adjustment magnitude, the horizontal axis coordinates, vertical axis coordinates, and height axis coordinates of unit G1 are adjusted respectively. For example, the adjusted horizontal axis coordinate X_G1_new = X_G1 - ΔX_G1 * adjustment coefficient (where ΔX_G1 is the horizontal axis deviation between unit G1 and the reference unit; since the adjustment direction points towards the reference unit, subtraction is used). The adjustment methods for the vertical axis and height axis coordinates are similar. After the adjustment is completed, the initial 3D point cloud element G1_new after the first adjustment is obtained. The same adjustment operation is performed on all other non-baseline elements in the group.

[0161] Step S146: Calculate the spatial distance between all initial 3D point cloud units in the group after the first adjustment, statistically analyze the distribution characteristics of the spatial distance, and determine whether a second adjustment is needed based on the distribution characteristics of the spatial distance. If the spatial distance distribution still exceeds the preset uniform distribution range, then redetermine the reference unit and calculate the new coordinate deviation, adjustment direction, and adjustment range.

[0162] After the first spatial position adjustment, calculate the spatial distance between each pair of initial 3D point cloud units (including the reference unit and adjusted non-reference units) within the group. Statistically analyze the distribution characteristics of these spatial distances, such as calculating the average, variance, maximum, and minimum distances, to determine if the distance distribution is uniform. Preset a range for uniform distance distribution; for example, the variance of the distances must be less than a certain threshold, and the ratio of the maximum to the minimum distance must be less than a certain threshold. If the spatial distance distribution after the first adjustment exceeds this preset range, it indicates that the spatial distribution of units within the group is still not uniform enough, requiring a second adjustment. At this point, re-execute the reference unit selection process in step S144, determining a new reference unit based on the unit coordinates after the first adjustment. Then, calculate the coordinate deviation, adjustment direction (pointing to the new reference unit), and adjustment magnitude (based on the new deviation) between other units within the group and the new reference unit.

[0163] Step S147: Adjust the spatial position of the initial 3D point cloud units in the group a second time according to the new adjustment parameters to obtain the initial 3D point cloud units after the second adjustment. Repeat the above adjustment process until the spatial distance distribution between all the initial 3D point cloud units in the group conforms to the preset uniform distance distribution range.

[0164] Based on the newly determined reference unit, new coordinate deviation, adjustment direction, and adjustment range in step S146, a second spatial position adjustment is performed on the initial 3D point cloud units within the group, resulting in the second adjusted initial 3D point cloud units. The adjustment method is the same as the first adjustment, i.e., the coordinates in each direction are corrected according to the new coordinate deviation and adjustment range. After completing the second adjustment, the spatial distance distribution characteristics of the units within the group are calculated and statistically analyzed again to determine whether they conform to the preset uniform distance distribution range. If they still do not conform, the adjustment process is repeated, the reference unit is redefined, the adjustment parameters are calculated, and adjustments are made until the spatial distance distribution between all initial 3D point cloud units within the group conforms to the preset range. At this point, the spatial position adjustment of the units within the group is considered to have achieved the target.

[0165] Step S148: Integrate all the adjusted initial 3D point cloud units, remove duplicate initial 3D point cloud units between groups, form a set of optimized 3D point cloud units, and store the coordinate information of each optimized 3D point cloud unit in the set of optimized 3D point cloud units into the coordinate database.

[0166] After all initial 3D point cloud units have undergone progressive spatial position adjustments, these adjusted groups are integrated. Since different groups may be generated based on different viewpoint images or feature region extractions, some initial 3D point cloud units may appear repeatedly in multiple groups. By comparing the 3D spatial coordinates and feature region identification information of the units, these duplicate initial 3D point cloud units between groups are identified and removed, retaining only unique units. All the adjusted initial 3D point cloud units after deduplication are combined to form an optimized 3D point cloud unit set. The units in this optimized 3D point cloud unit set have a more uniform spatial distribution and higher coordinate accuracy. The coordinate information (horizontal axis, vertical axis, and height axis coordinate values) of each optimized 3D point cloud unit in the optimized 3D point cloud unit set is stored in a coordinate database. The database uses an efficient index structure to support rapid data retrieval during subsequent spatial arrangement operations based on spatial distribution logic.

[0167] Step S149: When extracting the spatial association attributes of each 3D point cloud initial unit, in addition to the 3D spatial coordinates and spatial distance, the coordinate change trend of each 3D point cloud initial unit is also extracted. The coordinate change trend is determined based on the coordinate change sequence of the 3D point cloud initial unit when the coordinates are mapped from images from different viewpoints. The coordinate change trend is added as a supplementary attribute to the spatial association attributes to form a complete spatial association attribute containing coordinates, distance and change trend, which is used to improve the accuracy of attribute similarity calculation.

[0168] Building upon the spatial association attributes extracted in step S141, the content of these attributes is further enriched. For each initial 3D point cloud unit, its coordinate change trend is extracted. The determination of the coordinate change trend is based on the coordinate change sequence generated by the unit during coordinate mapping from images from different viewpoints. For example, a certain feature region may be extracted and mapped in multiple images from different viewpoints, generating multiple 3D coordinate values. These coordinate values ​​are arranged in the order of image capture to form a coordinate change sequence. The variation patterns of the horizontal, vertical, and height axis coordinate values ​​in this sequence are analyzed, such as the direction of change (increasing or decreasing) and the rate of change, to determine the coordinate change trend. This coordinate change trend is added as a supplementary attribute to the original spatial association attributes, making the spatial association attributes include not only 3D spatial coordinates and spatial distance but also coordinate change trends, forming a more complete spatial association attribute. This helps to more comprehensively evaluate the spatial association between two units when calculating the spatial association attribute similarity in step S143, thereby improving the accuracy of attribute similarity calculation.

[0169] Step S1410: When setting the attribute similarity threshold, refer to the correspondence between attribute similarity and 3D point cloud matching effect in historical matching data to determine a reasonable range of thresholds that can guarantee the matching effect, and select the middle value from the reasonable range of thresholds as the initial attribute similarity threshold.

[0170] In step S143, when grouping the initial 3D point cloud units, an attribute similarity threshold needs to be set. This threshold is set with reference to historical matching data. The historical matching data records the grouping results of the initial 3D point cloud units under different attribute similarity thresholds in the past, along with corresponding 3D point cloud matching performance evaluation metrics (such as matching accuracy and point cloud density uniformity). By analyzing the historical data, the correspondence between attribute similarity and matching performance is identified, and a reasonable range of attribute similarity thresholds is determined that ensures the matching performance meets preset requirements (such as a matching accuracy of no less than 95%). For example, if analysis shows that the matching performance is good when the threshold is between 0.6 and 0.8, this range is taken as the reasonable threshold range. An intermediate value, such as 0.7, is selected from this reasonable range as the initial attribute similarity threshold for preliminary grouping operations.

[0171] Step S1411: After dividing the initial 3D point cloud into initial unit groups based on the initial attribute similarity threshold, analyze the division results. If the number of initial 3D point cloud unit groups exceeds the preset maximum group number threshold and the average number of initial 3D point cloud units in each group is lower than the preset minimum unit number threshold, it is determined that the number of unit groups is too large and the number of units is too small. At this time, the attribute similarity threshold is lowered. If the number of initial 3D point cloud unit groups is lower than the preset minimum group number threshold and the average number of initial 3D point cloud units in each group exceeds the preset maximum unit number threshold, it is determined that the number of unit groups is too small and the number of units is too large. At this time, the attribute similarity threshold is increased.

[0172] After dividing the 3D point cloud into initial unit groups based on an initial attribute similarity threshold, the division results are analyzed. The number of initial 3D point cloud unit groups and the number of initial 3D point cloud units within each group are counted, and the average number of units within each group is calculated. Preset maximum and minimum group number thresholds are used. If the actual number of groups exceeds the maximum threshold, and the average number of units per group is lower than the minimum threshold, it indicates that the current threshold is too high, leading to over-subdivision of units. The attribute similarity threshold needs to be lowered so that more units are grouped together due to similarity exceeding the threshold, thus reducing the number of groups and increasing the number of units per group. Conversely, preset minimum and maximum group number thresholds are used. If the actual number of groups is lower than the minimum threshold, and the average number of units per group exceeds the maximum threshold, it indicates that the current threshold is too low, leading to over-merging of units. The attribute similarity threshold needs to be increased so that some units originally in the same group are grouped into different groups due to similarity not reaching the new threshold, thus increasing the number of groups and reducing the number of units per group.

[0173] Step S1412: After the first spatial position adjustment of the initial 3D point cloud units in the group, calculate the spatial association attribute similarity of the initial 3D point cloud units in the group after the adjustment, and compare it with the spatial association attribute similarity of the initial 3D point cloud units in the group before the adjustment. If the similarity improvement does not reach the preset improvement requirement, adjust the adjustment range calculation parameter in the spatial position adjustment and recalculate the adjustment range.

[0174] After the first spatial position adjustment is completed in step S145, the spatial association attribute similarity between all pairwise initial units of the 3D point cloud within the adjusted group is calculated, and the average of these similarities is calculated. The adjusted average similarity is compared with the average spatial association attribute similarity within the group before adjustment, and the similarity improvement is calculated (improvement = (adjusted average similarity - original average similarity) / original average similarity). A preset improvement requirement is set, such as an improvement of 10%. If the actual improvement does not meet this requirement, it indicates that the current adjustment calculation parameters (such as the adjustment coefficient in step S145) are set unreasonably and need to be adjusted. For example, if the improvement is insufficient, the adjustment coefficient can be appropriately increased to increase the unit adjustment range, thereby potentially improving the spatial association attribute similarity of units within the group to a greater extent. After adjusting the parameters, the adjustment range of each unit is recalculated, and the units within the group are adjusted for the first time according to the new adjustment range. Then, the similarity improvement is calculated again until the preset requirement is met.

[0175] Step S1413: When performing progressive spatial position adjustments, after each adjustment, calculate the spatial distance between each pair of initial units of all three-dimensional point clouds in the group, statistically analyze the distribution of spatial distances, and plot the spatial distance distribution curve.

[0176] During the progressive spatial position adjustment process in steps S145, S146, and S147, after each adjustment (including the first, second, and subsequent adjustments), the spatial distance of all initial 3D point cloud units within the group must be calculated. The spatial distance between any two units is calculated, and then the distribution of these spatial distances is statistically analyzed, such as the percentage of unit pairs within different distance intervals. Based on the statistical results, a spatial distance distribution curve is plotted with spatial distance as the abscissa and the percentage of unit pairs corresponding to that distance as the ordinate. This 3D point cloud optimized unit curve can intuitively reflect the distribution characteristics of the spatial distances within the group after adjustment.

[0177] Step S1414: By comparing the spatial distance distribution curves obtained after two adjacent adjustments, calculate the difference between the curves. When the difference is less than the preset stability threshold, it is determined that the adjustment process has become stable, and the adjustment of the initial unit of the three-dimensional point cloud is stopped.

[0178] During the progressive adjustment process, for two adjacent adjustments, such as after the first adjustment and after the second adjustment, their respective spatial distance distribution curves are obtained. The difference between these two curves is calculated. The difference can be calculated by comparing the sum of the squares of the differences in the ordinates (quantity percentage) of the two curves at the same abscissa (spatial distance), or by using the difference in the area under the curves, etc. A stabilization threshold is preset. When the difference between the spatial distance distribution curves after two adjacent adjustments is less than this stabilization threshold, it indicates that the spatial distance distribution of the units within the group has basically stabilized after adjustment, and further adjustment will not significantly improve the distribution uniformity. At this point, the adjustment process is considered to have stabilized, and further adjustment of the initial units of the 3D point cloud group is stopped.

[0179] Step S1415: When integrating all adjusted initial 3D point cloud units to form an optimized 3D point cloud unit set, check whether there are initial 3D point cloud units with a spatial distance less than a preset overlap threshold between different groups. If so, it is determined to be spatial overlap. For initial 3D point cloud units with spatial overlap, based on the similarity between their spatial association attributes and the average attributes of each group, they are reassigned to the initial 3D point cloud unit group with the highest similarity to ensure that each initial 3D point cloud unit belongs to only one group and that the spatial distribution of the initial 3D point cloud units within the group after reassignment conforms to the known spatial characteristics of the target area.

[0180] In step S148, when integrating and adjusting the initial 3D point cloud units, in addition to removing duplicate units within a group, it is also necessary to check whether there are spatially overlapping units between groups. For any two initial 3D point cloud units in different groups, the spatial distance between them is calculated. If this distance is less than a preset overlap threshold (e.g., a small distance value represents that the two units almost overlap spatially), then it is determined that the two units have spatial overlap. For units with spatial overlap, the similarity between the spatial association attribute of each unit and the average spatial association attribute of its group, as well as the similarity between the spatial association attribute of each unit and the average spatial association attribute of the groups of other overlapping units, is calculated. The unit is then reassigned to the group with the highest similarity to its spatial association attribute. After reassignment, it is ensured that each initial 3D point cloud unit belongs to only one group, and the spatial distribution of the initial 3D point cloud units within the reassigned group conforms to the known spatial characteristics of the target area. For example, in an urban area, units in the building group should be distributed within the urban building area, and units in the road group should be distributed along the road direction, etc.

[0181] Step S1416: Perform a unit count on the integrated 3D point cloud optimization unit set, compare the total number of units with the required number of units calculated based on the target area's spatial range and point cloud density requirements. If the total number of units does not reach the required number of units, supplement the missing area with multi-view remote sensing images, generate new 3D point cloud initial units, and re-execute the spatial association attribute extraction, grouping, and progressive adjustment process. Finally, store the obtained 3D point cloud optimization unit set in the optimization unit database.

[0182] After integrating and forming a set of optimized 3D point cloud units, the total number of units in the set is counted. Simultaneously, based on the spatial range of the target area (e.g., the area of ​​an urban area) and the preset point cloud density requirement (e.g., the number of point cloud units required per square meter), the required number of 3D point cloud units for the target area is calculated. The total number of units is compared with the required number. If the total number of units does not reach the required number, it indicates that some feature areas within the target area have not been fully collected and converted into 3D point cloud units, i.e., there are data gaps. For these gap areas, supplementary multi-view remote sensing images are collected. Then, following steps S110 to S148, the feature areas in the supplementary images are mapped to generate new initial 3D point cloud units. The processes of spatial correlation attribute extraction, attribute similarity-based grouping, and progressive spatial position adjustment are re-executed, and the newly generated 3D point cloud units are added to the original set. This process is repeated until the total number of units in the 3D point cloud optimized unit set reaches the required number. Finally, the final 3D point cloud optimized unit set is stored in the optimized unit database.

[0183] Step S150: According to the spatial distribution logic of the target area, arrange the three-dimensional point cloud optimization units in the set of three-dimensional point cloud optimization units in space and integrate them to form the three-dimensional point cloud result of the target area.

[0184] After obtaining the set of 3D point cloud optimization units, the 3D point cloud optimization units are spatially arranged and integrated according to the spatial distribution logic of the urban area to form the final 3D point cloud result.

[0185] Step S151: Obtain the spatial distribution logic of the target area. The spatial distribution logic includes the distribution order and positional relationship of different terrain types within the target area. Classify the three-dimensional point cloud optimization units in the set of three-dimensional point cloud optimization units according to the terrain type corresponding to each three-dimensional point cloud optimization unit. Each terrain type corresponds to a set of three-dimensional point cloud optimization units.

[0186] First, the spatial distribution logic of the target area (urban area) is obtained. This spatial distribution logic is determined based on urban planning data, geographic information system data, etc., and includes the distribution order and positional relationship of different terrain types within the urban area. For example, the urban center is dominated by commercial buildings, with residential areas and industrial areas distributed outwards in sequence, while natural terrain areas (such as mountains and rivers) and farmland are distributed on the urban edge. There are specific positional adjacency relationships between different terrain types. According to the terrain type to which the feature region corresponding to the 3D point cloud optimization unit belongs, all units in the 3D point cloud optimization unit set are classified. For example, units corresponding to commercial buildings are classified into the commercial building group, units corresponding to residential buildings into the residential building group, units corresponding to roads into the road group, units corresponding to natural terrain areas into the natural terrain group, etc. Each terrain type corresponds to a set of 3D point cloud optimization units.

[0187] Step S152: According to the distribution order of terrain types in the target area, determine the arrangement order of each group of three-dimensional point cloud optimization units in three-dimensional space, select the three-dimensional point cloud optimization unit group corresponding to the first type of terrain, and extract the three-dimensional spatial coordinates of all three-dimensional point cloud optimization units in the three-dimensional point cloud optimization unit group.

[0188] Based on the distribution order of different terrain types within the target area, such as the order from the city center to the edge being commercial buildings, residential buildings, roads, and natural terrain, the arrangement order of each group of 3D point cloud optimization units in 3D space is determined, i.e., spatial arrangement according to the order of commercial buildings, residential buildings, roads, and natural terrain. The 3D point cloud optimization unit group corresponding to the first terrain type in the arrangement order is selected, such as the commercial buildings group, and the 3D spatial coordinates (X, Y, Z) of all 3D point cloud optimization units within this group are extracted from the optimization unit database.

[0189] Step S153: Based on the spatial reference system of the target area, the three-dimensional spatial coordinates are standardized to align the coordinate values ​​with the spatial distribution logic.

[0190] The spatial reference system for the target area is a unified three-dimensional coordinate system used to accurately locate all spatial elements within the area. The three-dimensional spatial coordinates extracted in step S152 are transformed to this spatial reference system and standardized. For example, if the original coordinates are based on the principal point of an image, they need to be converted to coordinates based on the city center point; if the coordinate units are inconsistent, they need to be uniformly converted to meters. Through coordinate transformation operations such as translation, rotation, and scaling, the three-dimensional spatial coordinate values ​​are aligned with the spatial distribution logic of the target area, ensuring that the optimized three-dimensional point cloud units of the commercial building group are located at their corresponding positions in the city center area within the spatial reference system.

[0191] Step S154: Sort the 3D point cloud optimization units within the 3D point cloud optimization unit group in ascending order of standardized horizontal axis coordinate values. Based on the first sort, sort the 3D point cloud optimization units within the 3D point cloud optimization unit group in ascending order of standardized vertical axis coordinate values. Based on the second sort, sort the 3D point cloud optimization units within the 3D point cloud optimization unit group in ascending order of standardized height axis coordinate values. After sorting, arrange the 3D point cloud optimization units within the 3D point cloud optimization unit group in an orderly manner in 3D space according to the sorting results to form point cloud fragments of the first type of terrain.

[0192] The optimized 3D point cloud units of the standardized commercial building group are sorted. First, they are sorted in ascending order of horizontal axis coordinate (X) values, arranging the units from left to right horizontally. Based on this first sort, units with similar or identical horizontal axis coordinate values ​​are sorted in ascending order of vertical axis coordinate (Y) values, arranging them from front to back vertically. Based on this second sort, units with similar or identical horizontal and vertical coordinate values ​​are sorted in ascending order of height axis coordinate (Z) values, arranging them from bottom to top vertically. After these three sorts, the optimized 3D point cloud units of the commercial building group are arranged in an orderly fashion in 3D space according to the sorting results. These orderly arranged units together constitute the point cloud segment of the first type of terrain (commercial buildings), which accurately reflects the spatial distribution of commercial buildings in the city center.

[0193] Step S155: Select the 3D point cloud optimization unit group corresponding to the second type of terrain, repeat the above sorting and arrangement steps to form point cloud fragments of the second type of terrain, and stitch together the point cloud fragments of all terrain types according to the distribution order of terrain types; adjust the position of the 3D point cloud optimization units of adjacent fragments so that the 3D point cloud optimization units of adjacent fragments form a continuous connection in spatial distribution. During the stitching process, calculate the spatial distance between the 3D point cloud optimization units of adjacent fragments, adjust the spatial distance value so that the spatial distance distribution between fragments conforms to the preset uniform distribution range.

[0194] Select the 3D point cloud optimization unit group corresponding to the second type of terrain in the arrangement order, such as the residential building group. Repeat steps S153 to S154 to perform coordinate standardization, three-dimensional sorting (horizontal, vertical, and height directions), and orderly arrangement of the 3D point cloud optimization units of the residential building group to form the point cloud fragment of the second type of terrain (residential buildings). According to the distribution order of terrain types, perform the same operation on subsequent terrain type groups (road group, natural terrain group, etc.) to generate their respective point cloud fragments. Then, stitch together the point cloud fragments of all terrain types according to the distribution order, such as stitching residential building fragments after commercial building fragments, and stitching road fragments after residential building fragments, etc. When stitching adjacent fragments, calculate the spatial distance between the 3D point cloud optimization units at the edge of adjacent fragments. If the distance is too large or too small, adjust the position of the edge units to make the units of adjacent fragments form a continuous connection in spatial distribution. At the same time, ensure that the spatial distance distribution of the 3D point cloud optimization units between adjacent fragments conforms to the preset uniform distribution range and is consistent with the distance distribution within the fragments to avoid abrupt changes in point cloud density at the stitching point.

[0195] Step S156: After stitching together all segments, a preliminary three-dimensional point cloud structure of the target area is formed. The coordinates of the three-dimensional point cloud optimization units in the preliminary three-dimensional point cloud structure are extracted, and the spatial continuity of the preliminary three-dimensional point cloud structure is analyzed. Based on the spatial continuity analysis results, the missing areas in the preliminary three-dimensional point cloud structure are identified. According to the coordinate values ​​of the three-dimensional point cloud optimization units around the missing areas, the coordinate values ​​of the three-dimensional point cloud optimization units that should exist in the missing areas are calculated, and supplementary three-dimensional point cloud optimization units are generated.

[0196] After stitching together point cloud fragments of all terrain types, a preliminary 3D point cloud structure for the target area is formed. The coordinates of all 3D point cloud optimization units in this structure are extracted, and their spatial continuity is analyzed. Spatial continuity analysis is achieved by checking whether the spatial distance between adjacent units is within a preset continuity threshold. If the distance between units at multiple consecutive locations within a certain area exceeds the continuity threshold, the area is determined to be a void area. For identified void areas, the coordinate values ​​of the 3D point cloud optimization units that should exist in the void area are calculated using methods such as interpolation, based on the coordinate values ​​of the surrounding 3D point cloud optimization units (up, down, left, right, front, back, etc.). For example, for planar void areas, the average coordinates of the surrounding units can be used for filling; for void areas with height variations, the height variation trend of the surrounding units can be fitted and calculated. Supplementary 3D point cloud optimization units are generated based on the calculated coordinate values.

[0197] Step S157: Add the supplementary 3D point cloud optimization units to the preliminary 3D point cloud structure of the target area to fill in the missing areas and form the 3D point cloud structure of the target area. Format the 3D point cloud structure of the target area and store the 3D point cloud optimization unit information according to the standard format of 3D point cloud data. Read the formatted 3D point cloud data and check whether the 3D point cloud data completely reflects the spatial characteristics of the target area. If the 3D point cloud data meets the preset requirements of the spatial characteristics of the target area, it is used as the 3D point cloud result of the target area. If the 3D point cloud data does not meet the preset requirements of the spatial characteristics of the target area, it is re-integrated and adjusted.

[0198] The generated supplementary 3D point cloud optimization units are added to the initial 3D point cloud structure of the target area to fill in missing areas and form a complete 3D point cloud structure for the target area. This structure is then formatted by organizing and storing the coordinate information, unit identifier, and terrain type of the 3D point cloud optimization units according to standard 3D point cloud data formats (such as PLY, PCD, etc.). The formatted 3D point cloud data is read and compared with the spatial distribution logic of the target area and known geographic feature data (such as building height, road width, terrain undulation, etc.) to check whether the 3D point cloud data completely and accurately reflects the spatial characteristics of the target area. If it meets the preset requirements (such as a spatial feature conformity of over 98%), the 3D point cloud data is used as the 3D point cloud result for the target area; if it does not meet the requirements, such as significant spatial feature deviations or missing data, the process returns to step S151 to re-integrate and adjust the terrain type classification, sorting, or filling in missing areas until the 3D point cloud data meets the preset requirements.

[0199] Step S158: Store the final 3D point cloud results of the target area to the designated data storage area.

[0200] The 3D point cloud results of the target area that meet the preset requirements are stored in a designated data storage area. This storage area can be a high-performance disk array, a distributed file system, or a cloud storage service, ensuring data security and accessibility. During storage, the 3D point cloud results are compressed and indexed to improve data storage efficiency and the performance of subsequent retrieval, visualization, and other operations. Simultaneously, metadata information of the 3D point cloud results is recorded, such as data acquisition time, coordinate system information, point cloud density, and accuracy indicators. At this point, the process of generating a 3D point cloud of the target area based on multi-view remote sensing image matching is complete.

[0201] Based on the same inventive concept, please refer to Figure 2The diagram shows a schematic block diagram of a three-dimensional point cloud generation system 100 based on multi-view remote sensing image matching, provided in an embodiment of this application, for performing the above-described three-dimensional point cloud generation method based on multi-view remote sensing image matching. The three-dimensional point cloud generation system 100 based on multi-view remote sensing image matching may include a communication unit 110, a machine-readable storage medium 120, and a processor 130.

[0202] In this embodiment, the machine-readable storage medium 120 can also be integrated into the processor 130 and can communicate and interact with external systems through the communication unit 110. The machine-readable storage medium 120 stores machine-executable instructions for executing the scheme of this application, and the processor 130 executes the machine-executable instructions stored in the machine-readable storage medium 120 to implement the three-dimensional point cloud generation method based on multi-view remote sensing image matching provided in the aforementioned method embodiments.

[0203] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for generating 3D point clouds based on multi-view remote sensing image matching, characterized in that, The method includes: A set of multi-view remote sensing images of the target area is obtained. Based on the shooting view parameters of each remote sensing image, an initial framework for multi-view dynamic association is constructed. The initial association relationship of multi-view images is obtained by associating the shooting orientation association information between images from different viewpoints. Based on the initial association relationship of the multi-view images, the association attributes of corresponding feature regions in the images from different viewpoints are extracted, feature region association factors are generated, and a set of cross-view dynamic matching constraint rules is constructed by combining the change law of feature region association factors. Using the cross-view dynamic matching constraint rule set, coordinate mapping operation is performed on the feature regions in images from different viewpoints to convert the two-dimensional coordinates of each feature region into three-dimensional spatial coordinates, forming an initial set of three-dimensional point cloud units. For the set of initial units of the three-dimensional point cloud, the spatial association attributes of each initial unit of the three-dimensional point cloud are extracted, and the initial units of the three-dimensional point cloud are divided into groups based on the attribute similarity. The spatial position of each group of initial units of the three-dimensional point cloud is adjusted progressively to obtain the set of optimized units of the three-dimensional point cloud. According to the spatial distribution logic of the target area, the three-dimensional point cloud optimization units in the set of three-dimensional point cloud optimization units are arranged in space and integrated to form the three-dimensional point cloud result of the target area.

2. The method for generating 3D point clouds based on multi-view remote sensing image matching according to claim 1, characterized in that, Based on the initial association relationship of the multi-view images, the association attributes of corresponding feature regions in the images from different viewpoints are extracted, feature region association factors are generated, and a set of cross-view dynamic matching constraint rules is constructed by combining the variation law of the feature region association factors, including: Image pairs with corresponding relationships are selected from the initial association relationship of the multi-view images. Each image pair includes a first-view image and a second-view image. Multiple feature regions are selected in the first-view image. Based on the pixel distribution information of the image, the contour attributes and texture attributes of each feature region are extracted. The contour attributes include the boundary pixel arrangement pattern of the feature region, and the texture attributes include the pixel grayscale change pattern inside the feature region. Find the feature regions in the second view image that correspond to the multiple feature regions selected in the first view image respectively. Extract the contour attributes and texture attributes of the feature regions in the second view image corresponding to each feature region in the first view image in the same way as extracting the feature region attributes in the first view image. Compare the difference in contour attributes between each feature region in the first view image and the feature region in the second view image corresponding to that feature region. Calculate the overlap ratio of the contour attributes and use the overlap ratio as the first association parameter. By comparing the texture attribute differences between each feature region in the first-view image and the corresponding feature region in the second-view image, the similarity ratio of the texture attributes is calculated. The similarity ratio is used as the second association parameter. The first association parameter and the second association parameter are combined to form a feature region association factor. A feature region association factor is generated for each feature region in the first-view image and the feature region in the corresponding feature image in the second-view image. Collect feature region correlation factors for all feature region pairs, record the numerical changes of each feature region correlation factor under different shooting time periods, summarize the changing trends of all feature region correlation factors, and based on the changing trends of feature region correlation factors, set a reasonable fluctuation range for each feature region correlation factor, and use the fluctuation range as a basic constraint. Environmental impact parameters are extracted from images taken from different perspectives. These environmental impact parameters include information on changes in illumination and atmospheric transmission during the shooting process. The environmental impact parameters are then converted into constraint adjustment coefficients. The basic constraints are combined with these constraint adjustment coefficients, and by adjusting the range boundaries of the basic constraints, dynamic constraint rules that change with the environmental parameters are formed. Collect all dynamic constraint rules, classify them according to the type of feature region, and form a subset of constraint rules covering different types of feature regions. Combine each subset of constraint rules with the changing trend of the correlation factor of the feature region to construct a cross-perspective dynamic matching constraint rule set that includes constraint triggering conditions, constraint adjustment methods, and constraint scope. The system performs correlation checks on multiple rules in the cross-view dynamic matching constraint rule set, identifies and handles conflicts between rules, and enhances the coordination of multiple rules working together in the feature region matching process. Based on the historical matching result data of the feature region, the system adjusts the order in which the rules in the cross-view dynamic matching constraint rule set are applied to form a rule priority sequence, in which the rules that appear earlier in the sequence are applied first during the matching operation. The cross-view dynamic matching constraint rule set that has completed the correlation check and determined the rule priority sequence is stored in the rule management module.

3. The method for generating 3D point clouds based on multi-view remote sensing image matching according to claim 1, characterized in that, The step of extracting the association attributes of corresponding feature regions in images from different perspectives based on the initial association relationship of the multi-view images, generating feature region association factors, and constructing a set of cross-view dynamic matching constraint rules based on the changing patterns of the feature region association factors also includes: When extracting the contour attributes of the feature region in the first-view image, the boundary pixels of the feature region are sampled point by point, the coordinate information of each boundary pixel is recorded, and a boundary pixel coordinate sequence is formed. Based on the boundary pixel coordinate sequence, the connection angle between adjacent boundary pixels is calculated, the distribution characteristics of all connection angles are statistically analyzed, and the distribution characteristics are used as the core content of the contour attributes. When extracting the texture attributes of feature regions in first-person view images, multiple sampling regions are uniformly selected within the feature regions. Each sampling region contains multiple pixels. The number of times the pixel grayscale value changes in each sampling region is calculated. The distribution of the number of times the grayscale value changes in all sampling regions is statistically analyzed, and the distribution result is used as the core content of the texture attributes. When comparing the contour attributes of a feature region in a first-view image with the corresponding feature region in a second-view image, the boundary pixel coordinate sequences of the two are matched, the coordinate deviation of the corresponding boundary pixels is calculated, and the overlap ratio of the contour attributes is calculated based on the coordinate deviation. When comparing the texture attributes of a feature region in a first-view image with the corresponding feature region in a second-view image, the sampling areas of the two are matched one-to-one, the deviation of the number of gray value changes in the corresponding sampling areas is calculated, and the similarity ratio of the texture attributes is calculated based on the deviation. When recording the changes in the correlation factors of the feature region, the correlation factors of the feature region in different time periods are arranged in the order of shooting time, and the trend curve of the correlation factors of the feature region over time is plotted. When setting a reasonable fluctuation range for the correlation factors of the feature region, the stable segment in the trend curve is analyzed, and the maximum and minimum values ​​of the correlation factors of the feature region within the stable segment are used as the initial boundaries of the fluctuation range. When converting environmental impact parameters into constraint adjustment coefficients, the illumination adjustment coefficient is determined based on illumination change information. As illumination intensity increases, the illumination adjustment coefficient increases synchronously. The transmission adjustment coefficient is determined based on atmospheric transmission information. As transmittance increases, the transmission adjustment coefficient decreases synchronously. The illumination adjustment coefficient and the transmission adjustment coefficient are combined to obtain a comprehensive constraint adjustment coefficient, which is used to adjust the boundary of the fluctuation range of the correlation factors in the characteristic region. When constructing dynamic constraint rules, when the changes in environmental influence parameters exceed the preset range, a constraint adjustment operation is triggered, and the boundary range of the constraint conditions is adjusted according to the preset calculation method. When classifying dynamic constraint rules, they are classified according to the contour complexity of the feature region. Feature regions with different contour complexities correspond to different constraint rules. By combining the categorized constraint rules with the changing trends of feature region association factors, the constraint rule set can dynamically adjust the constraint strength according to the changes in feature region association factors and environmental parameters. The adjusted constraint rule set is tested, and the effectiveness of the rule set is verified through feature region matching operations. Based on the verification results, the rule content is further optimized, and the optimized constraint rule set is determined as the final cross-view dynamic matching constraint rule set.

4. The method for generating 3D point clouds based on multi-view remote sensing image matching according to claim 1, characterized in that, For the initial set of 3D point cloud units, the spatial association attributes of each initial 3D point cloud unit are extracted. Based on attribute similarity, the initial 3D point cloud units are grouped. Progressive spatial position adjustments are performed on each group of initial 3D point cloud units to obtain an optimized set of 3D point cloud units, including: The three-dimensional spatial coordinate information of each three-dimensional point cloud initial unit is extracted from the set of three-dimensional point cloud initial units. The three-dimensional spatial coordinate information includes the horizontal axis coordinate value, the vertical axis coordinate value, and the height axis coordinate value. The three-dimensional spatial coordinate information is used as the basic content of spatial association attributes. Calculate the spatial distance between each initial 3D point cloud unit and all other initial 3D point cloud units, use the spatial distance as a supplement to the spatial association attribute, and combine the 3D spatial coordinate information and spatial distance of each initial 3D point cloud unit to form the spatial association attribute; Calculate the spatial association attribute similarity between any two initial 3D point cloud units. The similarity calculation is based on the proximity of their 3D spatial coordinates and the similarity of their spatial distance lists. The proximity of the 3D spatial coordinates is evaluated by calculating the difference in coordinate values, and the similarity of the spatial distance lists is evaluated by calculating the consistency of the distance distribution. The evaluation results are combined into a spatial association attribute similarity. 3D point cloud initial units with spatial association attribute similarity exceeding the attribute similarity threshold are grouped into the same group, forming multiple 3D point cloud initial unit groups. For each group of initial 3D point cloud units, the initial 3D point cloud unit with the center position in the spatial coordinates of the group is selected as the reference unit. The coordinate deviation between each other initial 3D point cloud unit in the group and the reference unit is calculated. Based on the coordinate deviation, the adjustment direction of each initial 3D point cloud unit is determined, and the adjustment direction points to the direction of the reference unit. The adjustment range of each 3D point cloud initial unit is determined based on the magnitude of the coordinate deviation. When the coordinate deviation value increases, the adjustment range value increases synchronously. According to the determined adjustment direction and adjustment range, the first spatial position adjustment is performed on each 3D point cloud initial unit in the group to obtain the 3D point cloud initial unit after the first adjustment. Calculate the spatial distance between all initial 3D point cloud units in the group after the first adjustment, statistically analyze the distribution characteristics of the spatial distance, and determine whether a second adjustment is needed based on the distribution characteristics of the spatial distance. If the spatial distance distribution still exceeds the preset uniform distribution range, then redetermine the reference unit and calculate the new coordinate deviation, adjustment direction, and adjustment range. The initial spatial position of the three-dimensional point cloud units in the group is adjusted a second time according to the new adjustment parameters to obtain the initial three-dimensional point cloud units after the second adjustment. The above adjustment process is repeated until the spatial distance distribution between all the initial three-dimensional point cloud units in the group conforms to the preset uniform distance distribution range. All the adjusted initial 3D point cloud units are integrated, and duplicate initial 3D point cloud units between groups are removed to form a set of optimized 3D point cloud units. The coordinate information of each optimized 3D point cloud unit in the set of optimized 3D point cloud units is stored in the coordinate database.

5. The method for generating 3D point clouds based on multi-view remote sensing image matching according to claim 4, characterized in that, The calculation of the spatial distance between each initial 3D point cloud unit and all other initial 3D point cloud units, using spatial distance as a supplement to the spatial association attribute, includes: Obtain the three-dimensional spatial coordinates of each initial unit of the three-dimensional point cloud. The three-dimensional spatial coordinates include the coordinate values ​​of the horizontal axis, the vertical axis, and the height axis. Select the first initial unit of the three-dimensional point cloud as the target unit, and then select other initial units of the three-dimensional point cloud as comparison units in turn. Calculate the coordinate difference between the target element and the comparison element in the horizontal axis direction, and take the absolute value of the difference as the horizontal axis deviation. Calculate the coordinate difference between the target element and the comparison element in the vertical axis direction, and take the absolute value of the coordinate difference as the vertical axis deviation. Calculate the coordinate difference between the target element and the comparison element in the height axis direction, and take the absolute value of the coordinate difference as the height axis deviation. Based on the horizontal axis deviation, vertical axis deviation and height axis deviation, the spatial distance between the target unit and the comparison unit is obtained by spatial distance calculation. The spatial distance between the target unit and all comparison units is recorded to form a list of spatial distances of the target unit. Combine the spatial distance list of the target unit with the three-dimensional spatial coordinate information of the target unit to form the spatial association attribute of the target unit. Select the next three-dimensional point cloud initial unit as the new target unit, and repeat the above steps of calculating spatial distance and building spatial association attribute until all three-dimensional point cloud initial units have completed spatial distance calculation and spatial association attribute construction. Sort the list of spatial distances for each 3D point cloud initial unit in ascending order of spatial distance values, check whether the list of spatial distances in each spatial association attribute is complete, and confirm that the spatial distances between the target unit and all other 3D point cloud initial units are recorded. Standardize the format of coordinate information and spatial distance in spatial association attributes, and store the standardized spatial association attributes in a specified data area.

6. The method for generating 3D point clouds based on multi-view remote sensing image matching according to claim 4, characterized in that, The step of selecting the initial 3D point cloud unit at the center of the group as the reference unit for each initial 3D point cloud unit group includes: Extract the three-dimensional spatial coordinates of all three-dimensional point cloud initial units within each three-dimensional point cloud initial unit group, and separate the horizontal axis coordinates, vertical axis coordinates, and height axis coordinates of each three-dimensional point cloud initial unit; Calculate the average value of the horizontal axis coordinates of all initial 3D point cloud units in the group, and use it as the center coordinate of the horizontal axis of the group. Calculate the average value of the vertical axis coordinates of all initial 3D point cloud units in the group, and use it as the center coordinate of the vertical axis of the group. Calculate the average value of the height axis coordinates of all initial 3D point cloud units in the group, and use it as the center coordinate of the height axis of the group. The center coordinates of the horizontal axis, vertical axis, and height axis are combined to form the center coordinates within the group. The deviations of the three-dimensional spatial coordinates of each initial unit of the three-dimensional point cloud within the group from the center coordinates within the group are calculated. The deviations include the deviations of the horizontal axis, vertical axis, and height axis. The total deviation value of each 3D point cloud initial unit is calculated. The total deviation value is obtained by combining the deviation of the horizontal axis, the vertical axis, and the height axis. All 3D point cloud initial units in the group are arranged in ascending order of total deviation value. The initial 3D point cloud unit with the smallest total deviation value is selected as the candidate reference unit. If there are multiple initial 3D point cloud units with the same and smallest total deviation value, the average spatial distance between all initial 3D point cloud units with the same and smallest total deviation value and other initial 3D point cloud units in the group is further calculated. The initial 3D point cloud unit with the smallest average spatial distance is selected as the final reference unit. The 3D spatial coordinates of the reference unit are recorded as reference coordinates for subsequent adjustments of other initial 3D point cloud units. The coordinate information of the reference unit is associated and stored with the coordinate information of other 3D point cloud initial units in the group. The position of the reference unit in the group is analyzed so that the reference unit can represent the overall spatial distribution trend of the 3D point cloud initial units in the group. The selected reference elements are marked to prevent them from being mistakenly adjusted during subsequent adjustments. The marking information of the reference elements and their coordinate information are stored together in the element management module.

7. The method for generating 3D point clouds based on multi-view remote sensing image matching according to claim 1, characterized in that, The process of extracting spatial association attributes for each initial 3D point cloud unit from the initial 3D point cloud unit set, dividing the initial 3D point cloud units into groups based on attribute similarity, and performing progressive spatial position adjustments on each group of initial 3D point cloud units to obtain an optimized 3D point cloud unit set further includes: When extracting the spatial association attributes of each 3D point cloud initial unit, in addition to the 3D spatial coordinates and spatial distance, the coordinate change trend of each 3D point cloud initial unit is also extracted. The coordinate change trend is determined based on the coordinate change sequence of the 3D point cloud initial unit when the coordinates are mapped from images from different viewpoints. The coordinate change trend is added as a supplementary attribute to the spatial association attributes to form a complete spatial association attribute containing coordinates, distance and change trend, which is used to improve the accuracy of attribute similarity calculation. When setting the attribute similarity threshold, refer to the correspondence between attribute similarity and 3D point cloud matching effect in historical matching data to determine a reasonable range of thresholds that can guarantee the matching effect, and select the middle value from the reasonable range of thresholds as the initial attribute similarity threshold. After dividing the 3D point cloud into initial unit groups based on the initial attribute similarity threshold, the division results are analyzed. If the number of initial 3D point cloud unit groups exceeds the preset maximum group number threshold and the average number of initial 3D point cloud units in each group is lower than the preset minimum unit number threshold, then the number of unit groups is determined to be too many and the number of units is too few, and the attribute similarity threshold is lowered. If the number of initial 3D point cloud unit groups is lower than the preset minimum group number threshold and the average number of initial 3D point cloud units in each group exceeds the preset maximum unit number threshold, then the number of unit groups is determined to be too few and the number of units is too many, and the attribute similarity threshold is increased. After the initial spatial position adjustment of the 3D point cloud initial unit in the group, the spatial association attribute similarity of the initial 3D point cloud in the group after the adjustment is calculated and compared with the spatial association attribute similarity of the initial 3D point cloud in the group before the adjustment. If the similarity improvement does not reach the preset improvement requirement, the adjustment range calculation parameter in the spatial position adjustment is adjusted and the adjustment range is recalculated. When making progressive spatial position adjustments, after each adjustment, the spatial distance between each pair of initial units of all three-dimensional point clouds in the group is calculated, the distribution of spatial distances is statistically analyzed, and a spatial distance distribution curve is plotted. By comparing the spatial distance distribution curves obtained after two adjacent adjustments, the difference between the curves is calculated. When the difference is less than the preset stability threshold, it is determined that the adjustment process has become stable, and the adjustment of the initial unit of the three-dimensional point cloud is stopped. When integrating all the adjusted initial 3D point cloud units to form an optimized 3D point cloud unit set, check whether there are 3D point cloud initial units with a spatial distance less than a preset overlap threshold between different groups. If so, it is determined to be spatial overlap. For 3D point cloud initial units with spatial overlap, they are reassigned to the 3D point cloud initial unit group with the highest similarity based on the similarity between their spatial association attributes and the average attributes of each group. This ensures that each 3D point cloud initial unit belongs to only one group, and that the spatial distribution of the 3D point cloud initial units within the group after reassignment conforms to the known spatial characteristics of the target area. The number of units in the integrated 3D point cloud optimization unit set is counted. The total number of units is compared with the required number of units calculated based on the spatial range of the target area and the point cloud density requirements. If the total number of units does not reach the required number of units, multi-view remote sensing images of the missing area are collected to generate new 3D point cloud initial units. The spatial correlation attribute extraction, grouping and progressive adjustment process is re-executed. The final 3D point cloud optimization unit set is stored in the optimization unit database.

8. The method for generating 3D point clouds based on multi-view remote sensing image matching according to claim 1, characterized in that, The method of utilizing the cross-view dynamic matching constraint rule set to perform coordinate mapping operations on feature regions in images from different viewpoints, converting the two-dimensional coordinates of each feature region into three-dimensional spatial coordinates, and forming an initial set of three-dimensional point cloud units includes: Extract constraint rules for different types of feature regions from the cross-view dynamic matching constraint rule set, determine the coordinate mapping parameters corresponding to each type of feature region, select a feature region in the first view image, determine the type of the feature region, and find the constraint rules and coordinate mapping parameters corresponding to the type of feature region. Extract the two-dimensional coordinates of the feature region in the first-view image, and convert the horizontal coordinate values ​​in the two-dimensional coordinates into horizontal axis coordinate values ​​in three-dimensional space according to the coordinate mapping parameters corresponding to the feature region of the corresponding type. The two-dimensional coordinates include horizontal coordinate values ​​and vertical coordinate values. The vertical coordinate values ​​in the two-dimensional coordinate system are converted into vertical axis coordinate values ​​in the three-dimensional space. Combined with the shooting height parameters of the first-view image, the height axis coordinate values ​​of the feature region in the three-dimensional space are determined. The horizontal axis coordinates, vertical axis coordinates, and height axis coordinates are combined to form the three-dimensional spatial coordinates of the feature region, which is used as an initial unit of the three-dimensional point cloud. The next feature region is selected in the first-view image, and the above coordinate transformation steps are repeated to generate the initial unit of the three-dimensional point cloud corresponding to the next feature region. After completing the coordinate transformation of all feature regions in the first-view image, the same operation is performed on the feature regions in the second-view image. The two-dimensional coordinates of each feature region in the second-view image are extracted. Based on the constraint rules and coordinate mapping parameters corresponding to the type of each feature region, the two-dimensional coordinates are converted into three-dimensional spatial coordinates to generate the initial three-dimensional point cloud unit corresponding to each feature region. Repeat the above process until all feature regions in all viewpoint images have completed coordinate transformation, generating initial 3D point cloud units corresponding to each feature region, collecting all generated initial 3D point cloud units, and classifying them according to the type of feature region. The coordinates of the initial 3D point cloud units corresponding to the same type of feature region are sorted to keep the coordinate format consistent. The sorted initial 3D point cloud units are then combined to form a set of initial 3D point cloud units. The coordinates of each 3D point cloud initial unit in the set of 3D point cloud initial units are verified. By comparing the 3D coordinates of the 3D point cloud initial units corresponding to the feature regions adjacent to the feature regions corresponding to the 3D point cloud initial units, the coordinate deviations of the 3D point cloud initial units that exceed the preset allowable deviation range are corrected. The corrected set of 3D point cloud initial units is then stored in the initial unit database.

9. The method for generating three-dimensional point clouds based on multi-view remote sensing image matching according to claim 1, characterized in that, The step of arranging the 3D point cloud optimization units in the 3D point cloud optimization unit set according to the spatial distribution logic of the target area, and integrating them to form the 3D point cloud result of the target area, includes: The spatial distribution logic of the target area is obtained. The spatial distribution logic includes the distribution order and positional relationship of different terrain types in the target area. The three-dimensional point cloud optimization units in the set of three-dimensional point cloud optimization units are classified according to the terrain type corresponding to each three-dimensional point cloud optimization unit. Each terrain type corresponds to a set of three-dimensional point cloud optimization units. According to the distribution order of terrain types in the target area, determine the arrangement order of each group of three-dimensional point cloud optimization units in three-dimensional space, select the three-dimensional point cloud optimization unit group corresponding to the first type of terrain, and extract the three-dimensional spatial coordinates of all three-dimensional point cloud optimization units in the three-dimensional point cloud optimization unit group. Based on the spatial reference system of the target area, the three-dimensional spatial coordinates are standardized to align the coordinate values ​​with the spatial distribution logic. The three-dimensional point cloud optimization units within the group are sorted in ascending order of their standardized horizontal axis coordinate values. Based on this first sort, the three-dimensional point cloud optimization units within the group are sorted in ascending order of their standardized vertical axis coordinate values. Based on the second sorting, the three-dimensional point cloud optimization units within the group are sorted a third time according to the standardized height axis coordinate values ​​from smallest to largest. After the sorting is completed, the three-dimensional point cloud optimization units within the group are arranged in an orderly manner in three-dimensional space according to the sorting results to form point cloud fragments of the first type of terrain. Select the 3D point cloud optimization unit group corresponding to the second type of terrain, repeat the above sorting and arrangement steps to form the point cloud fragment of the second type of terrain, and stitch together the point cloud fragments of all terrain types according to the distribution order of terrain types. Adjust the position of the 3D point cloud optimization units of adjacent segments so that the 3D point cloud optimization units of adjacent segments form a continuous connection in spatial distribution. During the splicing process, calculate the spatial distance between the 3D point cloud optimization units of adjacent segments and adjust the spatial distance value so that the spatial distance distribution between segments conforms to the preset uniform distance distribution range. After stitching together all the segments, a preliminary three-dimensional point cloud structure of the target area is formed. The coordinates of the three-dimensional point cloud optimization units in the preliminary three-dimensional point cloud structure are extracted, and the spatial continuity of the preliminary three-dimensional point cloud structure is analyzed. Based on the spatial continuity analysis results, the missing regions in the preliminary three-dimensional point cloud structure are identified. According to the coordinate values ​​of the three-dimensional point cloud optimization units around the missing regions, the coordinate values ​​of the three-dimensional point cloud optimization units that should exist in the missing regions are calculated, and supplementary three-dimensional point cloud optimization units are generated. Supplementary 3D point cloud optimization units are added to the initial 3D point cloud structure of the target area to fill the missing areas and form the 3D point cloud structure of the target area. The 3D point cloud structure of the target area is formatted and the 3D point cloud optimization unit information is stored in the standard format of 3D point cloud data. Read the processed 3D point cloud data and check whether the 3D point cloud data fully reflects the spatial characteristics of the target area. If the 3D point cloud data meets the preset requirements of the spatial characteristics of the target area, it is used as the 3D point cloud result of the target area; if the 3D point cloud data does not meet the preset requirements of the spatial characteristics of the target area, it is re-integrated and adjusted. The final 3D point cloud results of the target area are stored in the designated data storage area.

10. A three-dimensional point cloud generation system based on multi-view remote sensing image matching, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the three-dimensional point cloud generation method based on multi-view remote sensing image matching as described in any one of claims 1 to 9 by executing the machine-executable instructions.

Images