Document automatic generation method and system based on multi-source spatial data fusion

Abstract

The application provides a document automatic generation method and system based on multi-source spatial data fusion, and relates to the fields of data processing and document generation; solves the technical problem that the position is often deviated when the existing building element is projected into the geographical background; includes: acquiring a multi-view view sequence containing a target object and a reference background projection; determining a spatial consistency quantization index of the target object and the reference background based on the multi-view view sequence, which is used to represent the spatial relative position deviation degree under different view projections; in response to the quantization index not meeting the preset condition, establishing a feedback closed-loop mechanism driven by the quantization index, iteratively adjusting the projection parameters to correct the deviation and regenerating the multi-view view sequence until the quantization index meets the preset condition; generating a target document based on the multi-view view sequence meeting the preset condition; the automatic and accurate compensation of the multi-view projection deviation is realized, and the spatial consistency of the multi-view view and the document reliability are improved.

Description

Technical Field

[0001] This application relates to the fields of data processing and document generation, and in particular to a method and system for automatic document generation based on multi-source spatial data fusion. Background Technology

[0002] In the research field of architecture and geographic information integration, exploring how to combine architectural design with geographic environmental data to demonstrate the comprehensiveness and accuracy of spatial information is of paramount value. Integrating architectural information with geospatial information can provide an intuitive and reliable basis for multi-party collaboration. However, current research and application methods often reveal some deep-seated shortcomings when facing actual business needs.

[0003] Many existing solutions lack support for precise spatial matching when dealing with the relationship between architectural details and the geographical environment, leading to potential biases in information presentation, especially when observation and analysis from different angles are required, making it difficult to guarantee spatial consistency of the data. This deficiency stems not merely from limitations in technical tools, but from the failure to effectively address the correspondence and transformation of spatial locations during information integration, thus affecting the reliability of the final results. The main problem lies in achieving precise spatial alignment between architectural details and the geographical environment. Since architectural design data is typically based on coordinates within a local area, while geographical environment data relies on a broad global coordinate system, the difference in spatial references between the two becomes a primary obstacle. This difference often results in positional offsets when projecting architectural details onto the geographical background. Furthermore, this offset can lead to inaccurate alignment between architectural elements and the geographical background when generating multi-angle views. For example, in a project report, the location of a building may not match the actual terrain or road layout, causing misleading information. Summary of the Invention

[0004] This application provides a method and system for automatic document generation based on multi-source spatial data fusion, which solves the technical problem that often occurs when building elements are projected onto a geographic background in the prior art.

[0005] To achieve the above objectives, this application adopts the following technical solution: In the first aspect, a method for automatic document generation based on multi-source spatial data fusion is provided, including: obtaining a multi-view view sequence containing spatial projections of a target object and a reference background; the multi-view view sequence includes a set of views generated by spatial projections of the target object and the reference background from different viewpoints; Based on the multi-view sequence, a spatial consistency quantification index between the target object and the reference background is determined. The spatial consistency quantification index is used to characterize the degree of spatial relative position deviation between the target object and the reference background under different view projections. In response to the failure of the spatial consistency quantification index to meet the preset conditions, a feedback closed-loop mechanism driven by the spatial consistency quantification index is established to iteratively adjust the projection parameters of the multi-view view sequence to correct the spatial relative position deviation, and the multi-view view sequence is regenerated based on the adjusted projection parameters until the spatial consistency quantification index meets the preset conditions. A target document is generated based on the multi-view sequence that meets the preset conditions.

[0006] Based on the above technical solution, the document automatic generation method based on multi-source spatial data fusion provided in this application realizes automatic detection and iterative correction of spatial relative position deviation in multi-view projection by establishing a feedback closed loop driven by quantitative indicators, thereby ensuring that the final generated document has a high degree of spatial consistency under different views, overcoming the defects of existing technologies where viewpoint switching causes offset and is difficult to automatically compensate.

[0007] In conjunction with the first aspect above, in one possible implementation, determining the spatial consistency quantification index between the target object and the reference background includes: Based on the multi-view sequence, feature information of the target object and the reference background is extracted; Based on the degree of matching of the feature information, the spatial consistency quantification index is determined.

[0008] By quantifying spatial consistency through the degree of matching of feature information, the abstract degree of fit is transformed into a calculable indicator, providing a precise data-driven foundation for subsequent feedback loops.

[0009] In conjunction with the first aspect above, in one possible implementation, the iterative adjustment of the projection parameters of the multi-view sequence includes: Based on the spatial consistency quantification index, the correction offset of the projection transformation matrix of the multi-view view sequence is determined; based on the correction offset, the projection parameters are iteratively adjusted.

[0010] By clarifying the correlation mechanism between quantitative indicators and projection parameter adjustments, and correcting the offset as a bridge, the feedback loop can accurately locate and compensate for deviations in projection transformation, thereby improving the pertinence and efficiency of automatic correction.

[0011] In conjunction with the first aspect above, in one possible implementation, determining the spatial consistency quantification index based on the degree of matching of the feature information includes: Calculate the feature point offset variance or overlap area ratio between the target object and the reference background; based on the feature point offset variance or overlap area ratio, generate a consistency score and mark it as a spatial consistency quantification index.

[0012] In conjunction with the first aspect above, in one possible implementation, determining the correction offset of the projection transformation matrix of the multi-view view sequence based on the spatial consistency quantification index includes: Based on the spatial coordinates of the target object and the texture features of the reference background, viewpoint depth information is extracted to construct an initial projection transformation matrix and determine the degree of geometric deformation of the multi-view sequence. Based on the degree of geometric deformation, the perspective distortion value is calculated by the lighting rendering intensity to obtain the scene fusion weight of the multi-view sequence; Based on the scene fusion weights, the correction offset of the projection transformation matrix is ​​determined using the pixel mapping relationship.

[0013] By introducing the calculation of perspective distortion and scene fusion weights, the deep geometric and optical mechanisms of deviations caused by perspective switching are revealed, thereby making the determination of the correction offset more accurate and effectively eliminating the nonlinear distortion effect caused by perspective transformation.

[0014] In conjunction with the first aspect above, in one possible implementation, the iterative adjustment of the projection parameters includes: The projection parameters are iteratively updated using a gradient descent-based optimization algorithm. The iterative update is terminated when the fusion error of the multi-view sequence converges to less than a preset error threshold.

[0015] In conjunction with the first aspect described above, in one possible implementation, before obtaining the multi-view sequence containing the target object and the spatial projection of the reference background, the method further includes: Requirements for obtaining multi-angle views from a merged dataset; Determine whether the view angles in the multi-angle view requirements exceed a preset threshold, and obtain the view sequence that needs to be optimized as the multi-view view sequence.

[0016] By filtering view angles, perspectives that are outside the normal range and prone to spatial deviations can be identified in advance, allowing subsequent closed-loop optimization to focus on the view sequences that truly need to be processed, thus improving the overall processing efficiency.

[0017] Secondly, this application provides a document automatic generation system based on multi-source spatial data fusion, comprising: The acquisition module is configured to acquire a multi-view sequence containing the spatial projection of the target object and the reference background; the index determination module is configured to determine a spatial consistency quantification index between the target object and the reference background based on the multi-view sequence. The parameter adjustment module is configured to iteratively adjust the projection parameters of the multi-view view sequence in response to the spatial consistency quantification index not meeting the preset conditions, so as to regenerate the multi-view view sequence until the spatial consistency quantification index meets the preset conditions. The document generation module is configured to generate a target document based on the multi-view sequence that meets the preset conditions.

[0018] In conjunction with the first aspect above, in one possible implementation, the parameter adjustment module includes: The offset determination unit is configured to determine the correction offset of the projection transformation matrix of the multi-view view sequence based on the spatial consistency quantification index. The parameter update unit is configured to iteratively adjust the projection parameters based on the correction offset.

[0019] Thirdly, this application provides an automatic document generation device based on multi-source spatial data fusion. The device includes a processor and a memory storing a computer program. When the processor executes the computer program, it implements the steps of the automatic document generation method based on multi-source spatial data fusion as described above.

[0020] This application provides a method and system for automatic document generation based on multi-source spatial data fusion. By introducing a spatial consistency quantification index, the degree of conformity between the target object and the reference background in multi-view projection is quantitatively characterized. This changes the limitation of existing technologies that rely solely on static coordinate alignment, enabling the spatial relative position deviation caused by viewpoint switching to be accurately detected and measured.

[0021] A feedback closed-loop mechanism driven by spatial consistency quantification indicators was established. In response to the failure of the quantification indicators to meet preset conditions, the projection parameters were iteratively adjusted to correct the deviation. This closed-loop mechanism achieves automatic and accurate compensation for multi-view projection deviations through the dynamic correlation between quantification indicators and projection parameter adjustments. It can eliminate the offset caused by viewpoint switching without manual intervention, significantly improving the spatial consistency and visual continuity of multi-view views.

[0022] The target document is generated based on a multi-view sequence that meets preset conditions, ensuring that the final output document has a high degree of spatial alignment reliability from different perspectives. This provides a precise and consistent basis for decision-making based on spatial information, and greatly improves the quality and practical value of automatic document generation.

[0023] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0024] Figure 1 A flowchart illustrating the document automatic generation method based on multi-source spatial data fusion provided in this application embodiment; Figure 2 A flowchart illustrating another document automatic generation method based on multi-source spatial data fusion provided in this application embodiment; Figure 3 A flowchart illustrating another document automatic generation method based on multi-source spatial data fusion provided in this application embodiment; Figure 4 A schematic diagram of the structure of the document automatic generation device based on multi-source spatial data fusion provided in the embodiments of this application; Figure 5 This is a schematic diagram of the hardware structure of the document automatic generation system based on multi-source spatial data fusion provided in an embodiment of this application. Detailed Implementation

[0025] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0026] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0027] Please see Figure 1-3 As shown, in order to solve the technical problem that the positional offset often occurs when building elements are projected into the geographic background in the prior art, this application provides the following embodiments.

[0028] Example 1: like Figure 1 As shown, this embodiment provides a method for automatic document generation based on multi-source spatial data fusion. The core of this method lies in establishing a feedback closed-loop mechanism driven by quantitative indicators to automatically and accurately correct spatial relative position deviations in multi-view projections, ultimately generating a target document with high spatial consistency. It should be understood that the following description is illustrative only and not restrictive, intended to help those skilled in the art better understand and implement the invention, and should not constitute any undue limitation on the scope of protection of the invention.

[0029] Step S100: Obtain a multi-view sequence containing the spatial projection of the target object and the reference background; The multi-view sequence includes a set of views generated by spatially projecting the target object and the reference background from different viewpoints.

[0030] Specifically, the "multi-view sequence" referred to in this application is not a static image from a single perspective, but rather a collection of spatial projections formed on the projection plane when the viewpoint switches around or across the target object and the reference background.

[0031] The target object and the reference background are two core elements that constitute multi-source spatial data. The target object usually refers to the main element that needs to be focused on or inserted, such as the design details of a building or a three-dimensional model of a mechanical part. The reference background refers to the macroscopic environment or base data that carries the target object, such as the geographical environment and terrain, or the overall assembly framework.

[0032] It should be noted that although this embodiment uses the integration of architecture and geographical environment as an example for easy understanding, in other embodiments, the target object can also be an underground pipeline model, interior furniture layout, etc., and the reference background can also be an urban underground space map, interior wall structure, etc., as long as the requirement of spatial projection superposition and matching between the two from multiple perspectives is met. By acquiring this set of views, the system can capture the spatial relative relationship between the target object and the reference background in dynamic perspective changes, providing a data foundation for subsequent consistency quantitative assessment.

[0033] Step S200: Based on the multi-view sequence, determine the spatial consistency quantification index between the target object and the reference background; Among them, the spatial consistency quantification index is used to characterize the degree of spatial relative position deviation between the target object and the reference background under different viewpoint projections.

[0034] Specifically, the "spatial consistency quantification index" referred to in this application characterizes the degree of alignment or deviation between the relative positions of the target object and the reference background in the projection rendering result during dynamic viewpoint switching. Therefore, it is fundamentally different from traditional static "coordinate alignment." Coordinate alignment is merely a static benchmark operation in the data preprocessing stage to unify data from different sources into the same coordinate system. Spatial consistency, on the other hand, focuses on whether the data, after coordinate alignment, can still maintain the correct relative spatial relationship in visual presentation after undergoing dynamic rendering transformations such as perspective projection and viewpoint rotation. For example, even if the target object and the reference background are perfectly aligned in 3D space, under a projection at a large angle, due to perspective distortion, the target object may appear to detach from the surface of the reference background or exhibit unreasonable overlap. This relative positional deviation generated under dynamic projection is precisely what the spatial consistency quantification index aims to capture and measure.

[0035] By quantifying the degree of deviation, the originally vague visual consistency is transformed into a calculable and comparable precise value, thus providing an objective driving basis for subsequent automatic correction.

[0036] Step S300: In response to the spatial consistency quantification index not meeting the preset conditions, a feedback closed-loop mechanism driven by the spatial consistency quantification index is established to iteratively adjust the projection parameters of the multi-view view sequence to correct the spatial relative position deviation, and the multi-view view sequence is regenerated based on the adjusted projection parameters until the spatial consistency quantification index meets the preset conditions.

[0037] Specifically, this step is the core dynamic correction logic of the present invention, namely the feedback closed-loop mechanism. When the quantitative index calculated in step S200 indicates that the spatial relative position deviation in the current view sequence is too large, i.e., the preset conditions are not met, the system does not directly abandon or manually intervene, but automatically triggers closed-loop iteration. The driving force of this closed loop comes directly from the aforementioned quantitative index. Based on the deviation direction and magnitude reflected by the index, the system adjusts the projection parameters of the multi-view view sequence in a targeted manner, such as adjusting the rotation component, translation component, or perspective scaling factor in the projection transformation matrix.

[0038] Based on the adjusted projection parameters, the system re-renders and generates a multi-view sequence, and recalculates the spatial consistency quantification index, forming an "evaluation-adjustment-re-evaluation" cycle. The iteration terminates only when the quantification index meets the preset conditions, i.e., the deviation has been corrected to an acceptable range.

[0039] This feedback loop mechanism ensures that the correction process is data-driven and gradually approximates the target, avoiding oscillations or overcorrection caused by blind adjustments, and achieving automatic and accurate compensation for deviations caused by perspective switching.

[0040] Step S400: Generate a target document based on the multi-view sequence that satisfies the preset conditions.

[0041] Specifically, once the feedback loop mechanism successfully optimizes the spatial consistency of the multi-view sequence to meet the requirements, these view sequences possess reliable spatial alignment and visual continuity. At this point, the system maps and outputs these optimized view sequences as the final target document.

[0042] The logic for generating the target document is to record the spatial distribution data, alignment status, and optimization parameters contained in the view sequence, and organize and render them according to the preset document structure framework. For example, it can generate planning reports or review documents that include multi-angle consistency analysis charts, spatial alignment reliability verification records, and optimized view displays.

[0043] It should be noted that the target document can be in the format of a PDF report, an XML structured data file, or an interactive web page document, as long as it carries multi-view information that meets the spatial consistency requirements.

[0044] Through the above steps, this embodiment forms a complete technical chain from quantitative evaluation to closed-loop correction and then to document output, which completely overcomes the defects of existing technologies where perspective switching causes offset and is difficult to automatically compensate for, and significantly improves the generation quality and reliability of multi-source spatial data fusion documents.

[0045] Example 2: Based on Example 1, this example further elaborates on the specific calculation logic for determining the spatial consistency quantification index in step S200. Based on the multi-view sequence, feature information of the target object and the reference background is extracted; based on the degree of matching of the feature information, the spatial consistency quantification index is determined.

[0046] Specifically, feature extraction is a crucial step in transforming the abstract visual representation in a view sequence into computable structured data. The feature information referred to in this application is not limited to single pixel color distributions, but focuses more on stable features that can characterize spatial topology and geometric structure.

[0047] In one implementation, the system can extract the edge contour features of the target object and the reference background through an edge detection algorithm. These contour features visually outline the shape boundaries of the two objects in the projection plane. In another implementation, the system can also extract key feature points, such as corner points, vertices, or extreme points of curvature change of the target object, as well as corresponding positioning reference points in the reference background, through feature point recognition algorithms.

[0048] It should be noted that although the above lists two specific forms—edge contours and key feature points—in other embodiments, the feature information can also be texture gradient distribution, region centroid coordinates, etc., as long as it can reflect the spatial occupancy and relative relationship between the target object and the reference background from a specific viewpoint. By extracting this feature information, the system provides data anchors for subsequent matching degree calculations.

[0049] The matching degree compares the feature information of the target object with the feature information of the reference background to measure the degree of overlap between the two in spatial position. Specifically, the higher the matching degree, the better the target object and the background are aligned from different perspectives; the lower the matching degree, the greater the deviation.

[0050] Further, based on the degree of matching of the feature information, the spatial consistency quantification index is determined, including: Calculate the variance of the feature point offset or the percentage of the overlapping area between the target object and the reference background; Based on the feature point offset variance or the overlap area ratio, a consistency score is generated and marked as a spatial consistency quantification index.

[0051] This step provides two parallel optional computational dimensions, designed to accurately capture the degree of spatial relative positional deviation through different mathematical paths, thereby expanding the protective net of equivalent alternatives.

[0052] As the first optional calculation dimension, the system calculates the feature point offset variance. Specifically, after extracting multiple pairs of key feature points between the target object and the reference background, the system first calculates the Euclidean distance deviation of each pair of feature points in the current projection view coordinate system, that is, the vector difference magnitude between the coordinates of the feature points of the target object and the corresponding feature points of the reference background. Subsequently, the system calculates the statistical variance based on the deviation values ​​of all feature point pairs. An example of the variance formula can be expressed as: Where N is the number of feature point pairs. Let be the deviation value of the i-th pair of feature points. The variance S is the mean of all deviation values, and the summation range is the number of all feature point pairs. This variance value S directly reflects the degree of dispersion of the target object relative to the reference background at different local positions. If the variance S is small, it indicates that the offset is uniform and small, and the spatial consistency is good; if the variance S exceeds the preset variance threshold (for example, setting the threshold to 2.5 pixels), it indicates that there is a serious relative position distortion or misalignment in the local area, and the spatial consistency is poor. The system then converts the calculated variance value S into a consistency score between 0 and 100 through a mapping function (such as negative correlation linear mapping or piecewise threshold mapping), with a higher score indicating better consistency.

[0053] As a second optional calculation dimension, the system calculates the percentage of overlapping area.

[0054] Specifically, the system performs pixel-level superposition comparison between the projected area of ​​the target object from the current viewpoint and the projected area of ​​the reference background from the corresponding viewpoint. Through Boolean logic operations or contour intersection calculations, the system extracts the actual overlapping pixel area on the projection surface. Simultaneously calculate the union area of ​​the sum of the projected areas of the target object and the reference background, minus the overlapping area. .

[0055] An example of calculating the percentage of overlapping area can be expressed as follows: .

[0056] The percentage R directly represents the degree of visual alignment between the target object and the reference background. If the percentage R is close to 1, it means that the two are almost completely overlapped on the projection plane, with no obvious relative misalignment or separation. If the percentage R is lower than the preset overlap threshold (e.g., the threshold is set to 0.75), it means that a large proportion of the target object is outside the reference background or has unreasonable interweaving, resulting in impaired spatial consistency.

[0057] The system can also directly use the percentage R as the consistency score, or transform it into a score within the standard scoring range through normalization mapping.

[0058] This embodiment introduces two specific calculation logics: feature point offset variance and overlapping area ratio. This transforms the relatively abstract concept of "spatial relative position deviation" from Embodiment 1 into a quantifiable value with clear physical meaning and mathematical support. Offset variance focuses on evaluating consistency from the perspective of local dispersion of the geometric structure, sensitive to minor deformations and misalignments. Overlapping area ratio, on the other hand, focuses on evaluating consistency from the perspective of macroscopic consistency of overall visual occupancy, sensitive to large-scale deviations and omissions. These two dimensions complement each other, enabling the system to make accurate quantitative responses to different types of spatial deviations. This provides a highly reliable driving basis for the subsequent feedback loop mechanism, ensuring the targeted and convergent nature of iterative adjustments.

[0059] Example 3: Building upon Example 1, this example further elaborates on the correlation mechanism and offset determination method of iteratively adjusting projection parameters in step S300. Specifically, this example reveals the deep geometric and optical mechanisms underlying the deviation caused by viewpoint switching, providing a solid causal chain support for the calculation of offset correction.

[0060] In step S300, the iterative adjustment of the projection parameters of the multi-view sequence includes: Based on the spatial consistency quantification index, determine the correction offset of the projection transformation matrix of the multi-view view sequence; The projection parameters are iteratively adjusted based on the corrected offset.

[0061] This step clarifies the core correlation between the quantification index and the adjustment of projection parameters. In the feedback loop mechanism, the spatial consistency quantification index acts as a deviation sensor, indicating the direction and degree of deviation between the target object and the reference background in the spatial relative position in the current view sequence; while the correction offset of the projection transformation matrix is ​​the output of the deviation compensator, which transforms the abstract quantification index into specific matrix parameter adjustment values.

[0062] By correcting the offset, the closed-loop mechanism can accurately locate and compensate for deviations in the projection transformation, avoiding parameter oscillations or overcorrection caused by blind adjustment, and significantly improving the targeting and efficiency of automatic correction.

[0063] Furthermore, determining the correction offset of the projection transformation matrix of the multi-view view sequence based on the spatial consistency quantification index includes the following in-depth mechanism derivation steps: Step S301: Based on the spatial coordinates of the target object and the texture features of the reference background, extract viewpoint depth information to construct an initial projection transformation matrix and determine the degree of geometric deformation of the multi-view sequence.

[0064] Specifically, when the viewing angle changes, the relative depth relationship between the target object and the reference background in three-dimensional space is the fundamental geometric reason for the projection deviation.

[0065] The system extracts depth information from the current viewpoint by analyzing the spatial coordinates of the target object (e.g., its Z-axis depth value in 3D space) and the texture features of the reference background (e.g., the distance relationships implied by variations in texture density). Based on this depth information, the system constructs an initial projection transformation matrix, which defines the mapping rules from 3D spatial points to 2D projection planes.

[0066] However, due to the non-uniform scaling effect caused by the change of viewpoint, the target object and the reference background will undergo different degrees of geometric deformation after being projected through the initial matrix. For example, the edges of the target object closer to the viewpoint will be stretched, while the reference background area farther away will be compressed. This difference in the degree of deformation directly leads to the misalignment of their relative spatial positions on the projection plane.

[0067] Step S302: For the degree of geometric deformation, the perspective distortion value is calculated by the lighting rendering intensity to obtain the scene fusion weight of the multi-view sequence.

[0068] Specifically, geometric deformation in visual rendering is often accompanied by changes in lighting distribution. The system cleverly utilizes this optical correlation mechanism to convert the degree of geometric deformation into a calculable photometric index.

[0069] Under perspective projection, areas with severe distortion are usually accompanied by abnormal attenuation or enhancement of lighting intensity. The system calculates the difference in lighting intensity between the target object and the reference background at the current viewpoint and quantifies it as a perspective distortion value.

[0070] It's important to note that there's a non-linear relationship between perspective distortion and geometric deformation: at small skew angles, the distortion value increases approximately linearly with the degree of deformation; however, when the skew angle increases to a critical range, perspective compression causes the distortion value to jump exponentially. This is the underlying reason why views at large skew angles are prone to severe spatial deviations. Subsequently, the system maps this perspective distortion value to scene blending weights. Blending weights are essentially an allocation of projection credibility for different regions. Regions with larger distortion values ​​have lower spatial credibility in their projection results; therefore, they need to be assigned higher adjustment weights during subsequent blending corrections to ensure the consistency of the overall view.

[0071] Step S303: Based on the scene fusion weights, determine the correction offset of the projection transformation matrix using the pixel mapping relationship.

[0072] Specifically, scene fusion weights ultimately need to be implemented through specific parameter interventions on the projection transformation matrix. The system utilizes a pre-established pixel mapping relationship, which records the coordinate correspondence between 3D spatial points and 2D projected pixels, to convert the fusion weights of each region into correction offsets for the matrix parameters. For example, for high-distortion regions with high fusion weights, the system assigns a larger offset step size to the corresponding matrix translation or rotation components when calculating the correction offset; while for low-distortion regions with low fusion weights, a smaller fine-tuning step size is assigned. This non-uniform offset allocation method based on pixel mapping relationships allows the correction operation to precisely focus on the local areas with the most severe deviations, rather than performing a uniform adjustment across the entire view, thus ensuring global convergence while achieving precise compensation for local details.

[0073] This embodiment fully illustrates the deep mechanism chain from viewpoint depth extraction to geometric deformation perception, then to perspective distortion normalization and fusion weight mapping, ultimately transforming into matrix correction of offset through the above steps. This mechanism proves that the correction logic of this invention is not a simple conventional coordinate translation, but rather profoundly conforms to the nonlinear physical optics laws under perspective projection. Thus, while resisting attacks based on common knowledge, it ensures reliable convergence and high-precision compensation of the feedback closed-loop mechanism in complex multi-view scenarios.

[0074] It should be noted that although this embodiment describes in detail the method for determining the offset based on perspective distortion and lighting rendering intensity, in other embodiments, the determination of the corrected offset can also be achieved based on the calculation of affine distortion rate of pure geometry or gradient mapping based on feature point alignment error, as long as it can establish a dynamic relationship between the quantization index and the matrix parameters.

[0075] Example 4: Building upon the iterative adjustment steps of Embodiment 1 or Embodiment 3, this embodiment further elaborates on the termination logic and specific optimization algorithm of the feedback closed-loop mechanism. The core of the feedback closed-loop mechanism lies in the "evaluation-adjustment-re-evaluation" cycle, and how to efficiently and stably break out of this cycle directly determines the balance between the final quality of multi-view optimization and the consumption of computational resources.

[0076] The iterative adjustment of the projection parameters of the multi-view sequence includes: The projection parameters are iteratively updated using a gradient descent-based optimization algorithm. The iterative update is terminated when the fusion error of the multi-view sequence converges to less than a preset error threshold.

[0077] Specifically, the gradient descent-based optimization algorithm is the mathematical engine that drives the projection transformation matrix parameters to approximate the direction of minimizing deviation. In each iteration, the system calculates the error gradient of the projection parameters based on the spatial consistency quantification index (or the fusion error function derived from it) of the current multi-view sequence. Subsequently, the system updates the projection parameters along the gradient descent direction with a preset step size. For example, in a preferred parameter setting example, the system sets the initial step size of gradient descent to 0.1 and the preset error threshold to 0.1. Therefore, in the early stages of iteration, the parameter adjustment is relatively large, which can quickly reduce significant spatial position deviations; as the iteration progresses, the error gradually decreases, and the magnitude of parameter updates naturally converges, achieving a smooth transition between coarse and fine adjustments. When the system calculates that the fusion error of the current multi-view sequence has dropped below 0.1, it indicates that the spatial relative position deviation between the target object and the reference background has been corrected to an acceptable accuracy range. At this point, the system automatically terminates the iterative update to avoid overcomputation.

[0078] To demonstrate the close relationship between the non-obviousness of the parameter range selected in this application and the technical effect, it must be pointed out that if the parameter settings deviate from the preferred range of this application, unexpected negative effects will occur.

[0079] As one of the out-of-bounds comparison methods, if the preset error threshold is set too high, such as 1.0, the system will prematurely terminate the iteration when the fusion error is still as high as 1.0. At this point, although the closed-loop mechanism stops operating, the spatial consistency quantification index between the target object and the reference background in the multi-view sequence is actually far from meeting the visual matching requirements. Figure 1 The consistency is still very poor, resulting in obvious spatial misalignment defects in the final generated target document, which defeats the purpose of closed-loop optimization.

[0080] As the second example of out-of-bounds comparison, if the step size of gradient descent is set too large, such as 1.5, the parameter update amplitude in each iteration will be too drastic. Since there is often a non-linear relationship between the spatial consistency quantification index and the projection parameters (such as the non-linear relationship between perspective distortion and offset as described in Example 3), an excessively large step size will cause the parameter update to exceed the minimum point of the error function, causing the fusion error to oscillate during the iteration process and fail to converge. It may even cause the position of the target object in the view to jump drastically, making it impossible to achieve automatic and accurate compensation.

[0081] This embodiment introduces a gradient descent-based optimization algorithm and a termination condition with a preset error threshold, which not only makes the feedback closed-loop mechanism logically complete and executable, but also clarifies, through specific parameter settings and comparison with out-of-bounds comparison ratios, that "coordinated setting of step size and threshold" is a key technical means to ensure that the closed loop is neither underfitting nor oscillating.

[0082] This precise parameter control enables the system to automatically find the optimal timing for terminating the iteration while ensuring optimization accuracy, thus balancing high quality in multi-view projection space consistency with computational efficiency in document generation.

[0083] It should be noted that although this embodiment uses a step size of 0.1 and an error threshold of 0.1 as a preferred example, in other embodiments, the step size can also be selected in the range of 0.01 to 0.5, and the error threshold can also be selected in the range of 0.05 to 0.2, depending on the complexity of the target object and the texture density of the reference background, as long as it can meet the dual constraints of avoiding oscillation and ensuring consistency.

[0084] Example 5: Building upon Example 1, this example further elaborates on the pre-screening logic for obtaining the multi-view sequence in step S100. In practical business scenarios involving multi-source spatial data fusion, the system typically receives a fusion dataset covering all perspectives. However, not all projections exhibit severe spatial consistency deviations. Blindly performing closed-loop iterative optimization on all perspectives would consume significant computational resources with minimal results. Therefore, this example employs a pre-screening mechanism to ensure that subsequent feedback loops accurately focus on the view sequences that truly require processing, significantly improving the overall optimization efficiency and targeting.

[0085] Before obtaining the multi-view sequence containing the target object and the reference background space projection, the method further includes: Obtain the multi-angle view requirements from the fusion dataset; determine whether the view angles in the multi-angle view requirements exceed a preset threshold, and obtain the view sequence that needs to be optimized as the multi-view view sequence.

[0086] Specifically, multi-angle view requirements refer to the view coverage requirements of the final document based on the business scenario. For example, in an urban planning report, it may be required to generate multiple specific angle projection views, such as a 0-degree overhead view, a 90-degree frontal view, and a 45-degree side view. After obtaining these view requirements, the system does not directly incorporate them all into the subsequent closed-loop optimization. Instead, it first performs threshold determination on the angle of each requirement.

[0087] It should be noted that the preset threshold refers to the range of conventional observation angles set based on the geometric laws of perspective projection. For example, the threshold range for horizontal observation angles can be set to ±180 degrees (i.e., no more than a full 360 degrees), or the threshold range for vertical tilt angles can be set to ±60 degrees. When the angle of a view in a multi-angle view request exceeds this preset threshold, such as requesting the generation of an extreme bird's-eye view with a tilt angle of 75 degrees that deviates from the conventional observation range, the system will determine it as a view that is prone to serious spatial deviations and mark it as a view sequence that needs optimization.

[0088] The reason for separately marking views exceeding the threshold lies in the nonlinear escalation of perspective distortion. As explained in Example 3, at small angles of view, the geometric deformation and perspective distortion of the target object and the reference background are approximately linearly related, with relatively small spatial positional deviations that are often within acceptable limits. However, when the angle of view increases and exceeds the critical threshold, the perspective compression effect causes an exponential increase in distortion values, making it extremely easy for the target object to detach from the reference background or produce unreasonable visual overlap on the projection plane. Therefore, these views exceeding the threshold are the real culprits causing spatial consistency problems and are the targets that the feedback loop mechanism needs to focus its efforts on overcoming.

[0089] As an alternative implementation, the preset threshold is not limited to a single absolute angle value; it can also be dynamically adjusted based on the geometric complexity of the target object and the texture density of the reference background. For example, for target objects with simple structures and high symmetry, their tolerance to perspective distortion is relatively strong, and the preset threshold can be appropriately relaxed to ±75 degrees. However, for target objects with complex structures and containing many slender components, even a small viewpoint shift can lead to severe visual occlusion and misalignment. In this case, the preset threshold needs to be tightened to ±45 degrees or even smaller. This dynamic threshold setting method allows the filtering logic to adapt to data sources with different characteristics, further eliminating unnecessary iterative calculations while ensuring optimization quality.

[0090] Through the pre-screening step in this embodiment, before entering the time-consuming feedback loop iteration, the system excludes views that naturally have good spatial consistency from a conventional perspective and do not require special optimization, and only uses view sequences with high-risk deviations as the multi-view view sequences obtained in the subsequent step S100. This mechanism achieves precise allocation of computing resources, avoids ineffective iterations, and thus significantly improves the overall efficiency of multi-source spatial data fusion and automatic document generation.

[0091] It should be noted that although this embodiment describes the filtering logic based on angle thresholds in detail, in other embodiments, the filtering conditions can also be comprehensively judged in combination with dimensions such as the resolution requirements of the view and the complexity of lighting rendering, as long as it can achieve the purpose of separating the view that is prone to deviation from the full data.

[0092] Example 6: like Figure 1 As shown, this embodiment provides an automatic document generation system based on multi-source spatial data fusion. This system materializes the closed-loop method of the aforementioned embodiments into specific functional modules. Through the collaborative operation and data flow between these modules, the stable operation and efficient execution of the feedback closed-loop mechanism at the system level are ensured. It should be understood that the module division described below is merely illustrative and not restrictive, intended to help those skilled in the art better understand the system architecture of this invention. In actual engineering deployments, these modules can be implemented either through software logic or through dedicated hardware processing circuits, as long as they can perform the corresponding data processing and flow functions.

[0093] The system includes: an acquisition module, an indicator determination module, a parameter adjustment module, and a document generation module.

[0094] The acquisition module is configured to acquire a multi-view sequence that includes the target object and the projection of the reference background space.

[0095] Specifically, the acquisition module, serving as the data entry point for the entire system, is responsible for reading a set of views containing the spatial projections of the target object and reference background from different perspectives from external data sources or internal storage media. This module not only acquires the raw view data but also performs preliminary parsing and validation of the data format, ensuring that the input multi-view sequence can be correctly recognized and processed by subsequent modules. For example, the acquisition module can receive projected view streams from a 3D rendering engine via a data interface or pull preprocessed view frame sequences from a distributed database. After completing data acquisition and preliminary processing, the acquisition module passes the multi-view sequence along the data flow path to the downstream indicator determination module.

[0096] The indicator determination module is configured to determine a spatial consistency quantification indicator between the target object and the reference background based on the multi-view sequence.

[0097] Specifically, the index determination module is the core module in the system responsible for deviation perception and quantitative assessment. It receives multi-view sequences from the acquisition module and uses feature extraction and matching algorithms to calculate the degree of spatial relative positional deviation between the target object and the reference background under different viewpoint projections, thereby outputting a spatial consistency quantitative index. This index is not only a static numerical value but also a dynamic state descriptor that contains the direction and severity of the deviation.

[0098] After completing the calculations, the indicator determination module does not merely display or record the indicators, but rather uses them as key control signals to drive the feedback loop, transmitting them to the parameter adjustment module along a specific data flow path. This direct data transfer relationship from the indicator determination module to the parameter adjustment module constitutes the starting point of the system-level "evaluation-adjustment" closed loop, ensuring that subsequent corrective actions are data-driven and targeted.

[0099] The parameter adjustment module is configured to iteratively adjust the projection parameters of the multi-view view sequence in response to the spatial consistency quantification index failing to meet the preset conditions, so as to regenerate the multi-view view sequence until the spatial consistency quantification index meets the preset conditions.

[0100] Specifically, the parameter adjustment module is the core execution unit in the system responsible for deviation compensation and iterative optimization. It receives spatial consistency quantification metrics from the metric determination module and determines whether these metrics meet preset consistency conditions. If the metrics indicate excessive deviation and fail to meet the conditions, the parameter adjustment module automatically triggers a feedback loop mechanism. Based on the deviation characteristics reflected by the quantification metrics, it calculates and adjusts the projection parameters of the multi-view sequence. Based on the adjusted projection parameters, the parameter adjustment module re-renders or calls an external rendering unit to generate new multi-view sequences, and then sends these regenerated view sequences back to the metric determination module for secondary evaluation.

[0101] The data flow loop of "indicator input - parameter adjustment - view regeneration - indicator re-evaluation" continuously occurs between the acquisition module, the indicator determination module, and the parameter adjustment module, forming a dynamic feedback data loop. Only when the quantitative indicator fed back by the indicator determination module finally meets the preset conditions will the parameter adjustment module terminate the iteration and pass the multi-view sequence that meets the conditions to the document generation module along the data flow path.

[0102] It should be noted that although this embodiment describes the parameter adjustment module directly regenerating the view sequence, in other embodiments, the parameter adjustment module may only output the adjusted projection parameter instructions, and the independent rendering engine will execute the view regeneration and then send the result back to the index determination module, as long as the complete flow of closed-loop data is ensured.

[0103] The document generation module is configured to generate a target document based on the multi-view sequence that meets the preset conditions.

[0104] Specifically, the document generation module, serving as the data output and output packaging unit for the entire system, receives multi-view sequences that have already met spatial consistency requirements from the parameter adjustment module. This module organizes and renders the spatial distribution data, alignment status, and optimization parameter records contained within the optimized view sequences according to a preset document structure framework, ultimately outputting the target document. For example, the document generation module can embed the view sequences into a PDF-formatted planning report or generate an XML structured data file containing interactive view displays. Through the encapsulation operations of the document generation module, the system transforms the closed-loop optimized high-quality spatial data into document outputs that users can directly read and use.

[0105] This embodiment fully reproduces the dynamic correction logic of the method side at the system level through the configuration relationship and data flow path of the above four modules, especially the collaborative operation mechanism of the indicator determination module transmitting the quantitative indicator as a control signal to the parameter adjustment module to drive the feedback closed loop.

[0106] This materialized modular architecture not only ensures the stable execution of the feedback loop mechanism in the actual computing system, but also improves the system's efficiency in accurately correcting projection deviations through clear data dependencies, thereby achieving automatic optimization of multi-view spatial consistency and reliable document generation.

[0107] Example 7: Building upon Example 6, this example further refines the internal structure of the parameter adjustment module to clarify the specific execution logic and collaborative mechanism of the core closed loop at the system level. The parameter adjustment module includes an offset determination unit and a parameter update unit.

[0108] Specifically, the offset determination unit is configured to determine the correction offset of the projection transformation matrix of the multi-view sequence based on the spatial consistency quantification index. This unit is the core of the deviation analysis in the parameter adjustment module. It receives the spatial consistency quantification index from the index determination module and converts this macroscopic state descriptor, which reflects the degree of spatial relative position deviation, into specific microscopic parameter adjustment values ​​in the projection transformation matrix. For example, when the quantification index shows that the target object tends to shift to the upper right relative to the reference background in the current viewpoint, the offset determination unit does not simply output a reverse translation vector. Instead, it combines the perspective distortion value and scene fusion weight described in Embodiment 3 to calculate the precise correction offset for the rotation and translation components in the projection transformation matrix.

[0109] This transformation process takes into account the nonlinear geometric deformation mechanism caused by viewpoint switching, enabling the correction offset to accurately pinpoint the true root cause of the deviation, rather than merely remaining at the level of pixel displacement. After completing the calculation, the offset determination unit transmits the correction offset as specific control command data along the internal data flow path to the downstream parameter update unit.

[0110] The parameter update unit is configured to iteratively adjust the projection parameters based on the corrected offset. This unit is the execution engine in the parameter adjustment module. It receives the corrected offset from the offset determination unit and directly applies it to the projection transformation matrix of the current multi-view sequence, completing the iterative update operation of the matrix parameters. For example, the parameter update unit superimposes the corrected offset onto the corresponding elements of the original projection transformation matrix to generate a new set of projection parameters, and drives the rendering engine to regenerate the multi-view sequence based on this new parameter set. Subsequently, the regenerated view sequence will be sent back to the index determination module for secondary evaluation, thus forming a complete feedback data loop.

[0111] It's important to note that there's a strict upstream-downstream data dependency between the offset determination unit and the parameter update unit. The output of the offset determination unit is a necessary input for the parameter update unit; they are logically decoupled but data-driven. This refined internal architecture allows for a specialized division of labor between the two distinct computational tasks: offset analysis and parameter execution. The offset determination unit focuses on complex geometric and optical mechanism calculations to ensure the accuracy of the correction direction; the parameter update unit focuses on efficient matrix operations and iterative processes to ensure the timeliness of the correction.

[0112] This division of labor and collaboration mechanism avoids computational congestion or logical confusion that may occur when a single processing module faces complex closed-loop logic, significantly improving the system's execution efficiency and stability in accurately correcting projection deviations. Although this embodiment describes the offset determination unit and the parameter update unit as two independent logical units, in other embodiments, they can also be integrated in different computing cores of the same processing chip, or deployed on different computing nodes through a microservice architecture, as long as the data dependency and flow relationship between them is not disrupted.

[0113] Example 8: Please see Figure 4 As shown, this embodiment provides a document automatic generation device based on multi-source spatial data fusion. This device serves as a product fallback, providing a complete hardware operating platform for the methods in the aforementioned embodiments, ensuring the efficient execution and deployment of the feedback closed-loop mechanism in actual computing devices, and preventing the avoidance of end-application scenarios.

[0114] The device includes a processor and a memory storing a computer program, wherein the processor executes the computer program to implement the steps of the document automatic generation method based on multi-source spatial data fusion as described in any one of the foregoing embodiments 1 to 7.

[0115] Specifically, the processor is the core computing and control unit in the device. It can be a general-purpose central processing unit (CPU), graphics processing unit (GPU), digital signal processor (DSP), or a dedicated artificial intelligence acceleration chip, as long as it can provide sufficient computing power to support the rendering and iterative calculation of multi-view sequence. Memory includes volatile memory (such as RAM) and non-volatile memory (such as ROM, hard disk, solid-state drive, or flash memory). Non-volatile memory is used to persistently store computer program instructions, initial fusion datasets, and the final generated target document, while volatile memory is used to temporarily store dynamic data such as multi-view sequence, intermediate projection parameters, spatial consistency quantification indicators, and correction offsets during iteration during program execution.

[0116] The key point is that the processor does not merely passively execute code line by line, but rather constructs and drives the core mechanism of "quantization index-driven feedback closed-loop iterative calculation and document rendering generation" at the hardware level by executing specific computer program instructions. During device operation, the processor first reads a multi-view sequence containing the spatial projection of the target object and the reference background from memory via the bus and loads it into the cache or register set. Subsequently, the processor calls the vector operation instruction set to extract and match the feature information in the view sequence in parallel, quickly generating a spatial consistency quantization index. When the quantization index does not meet the preset conditions, the logic control unit inside the processor automatically jumps to the closed-loop iterative program segment, triggering the feedback closed-loop mechanism. In each iteration, the processor calculates the correction offset of the projection transformation matrix based on the quantization index and adds this offset to the current projection parameters through the arithmetic logic unit (ALU), and then reschedules the rendering pipeline to generate a new view sequence. This process forms a compact "evaluation-offset calculation-parameter update-re-rendering" data loop in the processor's instruction pipeline, so that the correction of spatial deviation does not require manual intervention, but relies on the processor's high-speed instruction loop to automatically approach the optimal solution. Only when the processor detects that the consistency quantification metric meets the preset conditions will it exit the loop program segment and enter the document rendering and generation process, encapsulating the optimized view sequence into the target document.

[0117] It should be understood that although this embodiment describes the basic architecture of the processor and memory, in actual engineering deployments, the device can also include a transceiver, which includes peripheral hardware such as communication interfaces and input / output devices. The processor can also be a multi-core processor or a distributed computing cluster to meet the concurrent processing needs of massive multi-source spatial data. As long as its hardware architecture can support the processor to implement quantitative index-driven feedback closed-loop iterative calculation and document rendering generation at the physical level through instruction execution, it should be covered within the protection scope of this invention. This embodiment, by clearly defining the hardware operating carrier and instruction execution logic at the device level, ensures that the aforementioned method closed loop is not only valid in algorithmic logic, but also has the feasibility of efficient execution and commercial application in physical computing devices.

[0118] Example 9: To more clearly illustrate the technical solution of this invention and its operational effects in practical business, a detailed explanation is provided below using the scenario of urban planning and architectural geographic information fusion as an example. It should be understood that this application scenario is merely illustrative and not restrictive, intended to help those skilled in the art better understand the core closed-loop mechanism and commercial value of this invention. This invention is also applicable to other applications requiring multi-source spatial data fusion and visual... Figure 1 Areas requiring consistency optimization include, for example, the integration of underground pipe networks with urban base maps, the integration of interior furniture with building structural layouts, and the integration of mechanical parts with assembly frames. Any such integration can be achieved as long as it involves optimizing the spatial fit between the target object and the reference background under multi-view projection.

[0119] In this urban planning scenario, the "target object" in the aforementioned embodiments is specified as "details of high-rise residential buildings," the "reference background" is specified as "geographical environment background (such as surrounding roads, green spaces, and topography)," and the "target document" is specified as "urban planning spatial consistency report." The specific processing flow executed by the system is as follows: Step S9001: Obtain a multi-view sequence containing details of high-rise residential buildings and spatial projections of the geographical environment.

[0120] Specifically, the system extracts local 3D model data of a high-rise residential building from the urban planning database and obtains macro-global data of the area, including roads, green spaces, and terrain elevation, from the geographic information system interface. After conventional static coordinate alignment preprocessing, the system, based on business requirements, generates projection views from multiple specific angles, such as a 0-degree overhead view, a 90-degree frontal view, and a 45-degree side view. According to the pre-screening logic in Example 5, the system determines that views with large skew angles, such as a 45-degree side view, are prone to perspective distortion and marks them as view sequences that need optimization, thereby obtaining a multi-view set of views containing the spatial projection of high-rise residential building details and geographical background from different perspectives.

[0121] Step S9002: Based on the multi-view sequence, determine the spatial consistency index between the details of high-rise residential buildings and the geographical environment.

[0122] Specifically, the system extracts the edge contours and corner features of high-rise residential building details from a 45-degree side-view projection view, as well as feature points of road edges and terrain contour lines in the geographical background. Then, the system uses the first optional calculation dimension in Example 2 to calculate the variance of the feature point offset between the high-rise residential building details and the geographical background. For example, the system calculates the Euclidean distance deviation between the residential building corner point and the corresponding road positioning reference point in the projection plane, and calculates the statistical variance based on the deviation values ​​of all feature point pairs. If the calculated variance value is large, exceeding a preset variance threshold, the system generates a lower consistency score through a mapping function. This score directly represents the degree of spatial relative position deviation of the high-rise residential building details relative to the geographical background from the current side-view perspective; that is, the residential building appears detached from the terrain surface or has an unreasonable intersection with the road layout.

[0123] Step S9003: In response to the spatial consistency quantification index not meeting the preset conditions, the projection parameters of the multi-view view sequence are iteratively adjusted by establishing a feedback closed-loop mechanism driven by the spatial consistency quantification index.

[0124] Specifically, because the consistency score of the 45-degree side view is lower than the preset reliability standard, the system automatically triggers a feedback loop. Based on the underlying mechanisms of Examples 3 and 4, the system first extracts viewpoint depth information to construct an initial projection transformation matrix based on the spatial coordinates of the high-rise residential building details and the texture features of the geographical environment background, and determines the degree of geometric distortion under this side view. For this degree of geometric distortion, the system calculates the lighting rendering intensity deviation, quantifies the perspective distortion value, and maps it to scene blending weights. Subsequently, the system uses pixel mapping relationships to convert the scene blending weights corresponding to high-distortion areas into correction offsets of the projection transformation matrix. The system uses a gradient descent-based optimization algorithm to iteratively update this correction offset with an initial step size of 0.1, gradually adjusting the rotation and translation components in the projection parameters to correct the perspective shift of the residential building in the geographical background. After each iteration, the system re-renders and generates the side view and recalculates the consistency score until the blending error converges to a preset error threshold of less than 0.1, at which point the iteration terminates.

[0125] Step S9004: Based on the multi-view sequence that meets the preset conditions, generate an urban planning spatial consistency report.

[0126] Specifically, once the feedback loop mechanism successfully optimizes the spatial consistency of the 45-degree side view to meet the requirements, the system organizes and renders the sequence of top-down, front-view, and optimized side-view views, along with parameter adjustment records and consistency score change curves from the iteration process, according to a pre-defined planning report structure framework. The final output is an urban planning spatial consistency report containing multi-angle consistency analysis charts and spatial alignment reliability verification records. This report provides accurate and consistent visual and data support for urban planning decisions, completely eliminating the risk of spatial misleading caused by perspective shifts.

[0127] This embodiment demonstrates the end-to-end operational effectiveness of the present invention through the aforementioned complete industry application scenario, from view acquisition, quantitative evaluation, closed-loop iterative correction to document output. It particularly emphasizes the practical technical value of gradient descent iteration based on feature point offset variance score and perspective distortion correction offset in resolving multi-view projection deviations between high-rise residential buildings and geographical backgrounds. This proves that the core closed-loop mechanism of the present invention is not an abstract mathematical game, but a key technical means that can effectively solve industry pain points and improve the quality and reliability of document generation.

[0128] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention, such as adjusting the specific calculation dimension of the spatial consistency quantification index, equivalent replacement of the mechanism for determining the offset correction of the projection transformation matrix, or reasonable changes to the feedback closed-loop iterative optimization algorithm and termination conditions, as long as they essentially still belong to the technical solution of iteratively correcting the relative position deviation of multi-view projection space based on the feedback closed-loop mechanism driven by the spatial consistency quantification index, should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for automatic document generation based on multi-source spatial data fusion, characterized in that, include: Obtain a multi-view sequence containing spatial projections of a target object and a reference background; the multi-view sequence includes a set of views generated by spatially projecting the target object and the reference background from different viewpoints. Based on the multi-view sequence, a spatial consistency quantification index between the target object and the reference background is determined; The spatial consistency quantification index is used to characterize the degree of spatial relative position deviation between the target object and the reference background under different viewpoint projections; In response to the failure of the spatial consistency quantification index to meet the preset conditions, a feedback closed-loop mechanism driven by the spatial consistency quantification index is established to iteratively adjust the projection parameters of the multi-view view sequence to correct the spatial relative position deviation, and the multi-view view sequence is regenerated based on the adjusted projection parameters until the spatial consistency quantification index meets the preset conditions. A target document is generated based on the multi-view sequence that meets the preset conditions.

2. The document automatic generation method based on multi-source spatial data fusion according to claim 1, characterized in that, The determination of the spatial consistency quantification index between the target object and the reference background includes: Based on the multi-view sequence, feature information of the target object and the reference background is extracted; Based on the degree of matching of the feature information, the spatial consistency quantification index is determined.

3. The document automatic generation method based on multi-source spatial data fusion according to claim 1, characterized in that, The iterative adjustment of the projection parameters of the multi-view sequence includes: Based on the spatial consistency quantification index, determine the correction offset of the projection transformation matrix of the multi-view view sequence; The projection parameters are iteratively adjusted based on the corrected offset.

4. The document automatic generation method based on multi-source spatial data fusion according to claim 2, characterized in that, The determination of the spatial consistency quantification index based on the matching degree of the feature information includes: Calculate the variance of the feature point offset or the percentage of the overlapping area between the target object and the reference background; Based on the feature point offset variance or the overlap area ratio, a consistency score is generated and marked as a spatial consistency quantification index.

5. The document automatic generation method based on multi-source spatial data fusion according to claim 3, characterized in that, The step of determining the correction offset of the projection transformation matrix of the multi-view view sequence based on the spatial consistency quantification index includes: Based on the spatial coordinates of the target object and the texture features of the reference background, viewpoint depth information is extracted to construct an initial projection transformation matrix and determine the degree of geometric deformation of the multi-view sequence. Based on the degree of geometric deformation, the perspective distortion value is calculated by the lighting rendering intensity to obtain the scene fusion weight of the multi-view sequence; Based on the scene fusion weights, the correction offset of the projection transformation matrix is ​​determined using the pixel mapping relationship.

6. The document automatic generation method based on multi-source spatial data fusion according to claim 3, characterized in that, The iterative adjustment of the projection parameters includes: The projection parameters are iteratively updated using a gradient descent-based optimization algorithm. The iterative update is terminated when the fusion error of the multi-view sequence converges to less than a preset error threshold.

7. The document automatic generation method based on multi-source spatial data fusion according to claim 1, characterized in that, Before obtaining the multi-view sequence containing the target object and the reference background space projection, the method further includes: Requirements for obtaining multi-angle views from a merged dataset; Determine whether the view angles in the multi-angle view requirements exceed a preset threshold, and obtain the view sequence that needs to be optimized as the multi-view view sequence.

8. A document automatic generation system based on multi-source spatial data fusion, applied to the method described in any one of claims 1-7, characterized in that, include: The acquisition module is configured to acquire a multi-view sequence that includes the target object and the spatial projection of the reference background space; The indicator determination module is configured to determine the spatial consistency quantification index between the target object and the reference background based on the multi-view sequence. The parameter adjustment module is configured to iteratively adjust the projection parameters of the multi-view view sequence in response to the spatial consistency quantification index not meeting the preset conditions, so as to regenerate the multi-view view sequence until the spatial consistency quantification index meets the preset conditions. The document generation module is configured to generate a target document based on the multi-view sequence that meets the preset conditions.

9. The document automatic generation system based on multi-source spatial data fusion according to claim 8, characterized in that, The parameter adjustment module includes: The offset determination unit is configured to determine the correction offset of the projection transformation matrix of the multi-view view sequence based on the spatial consistency quantification index. The parameter update unit is configured to iteratively adjust the projection parameters based on the correction offset.

10. A document automatic generation device based on multi-source spatial data fusion, applied to the method described in any one of claims 1-7, characterized in that, The device includes a processor and a memory storing a computer program, wherein the processor executes the computer program to implement the steps of the document automatic generation method based on multi-source spatial data fusion as described in any one of claims 1 to 7.

Images