Building facade three-dimensional surveying and mapping method and system based on oblique photography

Abstract

The embodiment of the invention provides a building facade three-dimensional surveying and mapping method and system based on oblique photography, and belongs to the technical field of surveying and mapping. The method comprises the following steps: acquiring a multi-angle inclined image of a building facade, and generating region labeling data for representing region characteristics of the building facade; acquiring fused image data reflecting natural texture information of the building facade and structural light stripe information of the building facade at the same time; performing spatial matching on the natural texture feature points and the structured light feature points based on a feature point matching rule to generate initial three-dimensional point cloud data of the building facade; and optimizing the initial three-dimensional point cloud data based on the priori parameters of the component structure, and outputting a three-dimensional surveying and mapping result of the building facade. According to the scheme of the invention, based on a joint reconstruction mechanism of an inclined image, structured light information and a component prior constraint, a building elevation can still obtain a three-dimensional surveying and mapping result with a stable structure and coherent details in a texture weak area, a reflection area and a shielding area.

Description

Technical Field

[0001] This invention relates to the field of surveying and mapping technology, specifically to a method and system for three-dimensional mapping of building facades based on oblique photogrammetry. Background Technology

[0002] 3D mapping of building facades has become routine in urban renewal, facade defect surveys, and digital management. However, existing technologies still revolve around multi-angle oblique photography, and their reliability remains a significant challenge under complex facade conditions. The industry standard is to rely on natural textures for feature point matching, and then calculate spatial geometric relationships between multiple views. However, when the facade surface texture is insufficient, or when there are large areas of monochromatic walls, feature points are often difficult to extract stably, and the subsequent matching process is prone to scattering or failure. This problem is particularly typical in common scenarios such as old neighborhoods and community walls; what appears "easy to photograph" on-site often turns out to be unreconstructable in large areas when the data is returned to the data center, with some point clouds even showing gaps.

[0003] Another long-standing and unresolved problem stems from reflective surfaces. Modern urban architecture extensively utilizes glass curtain walls, aluminum panels, or polished stone, materials that produce strong reflections in oblique photography. The resulting highlights and saturated areas not only distort the feature points themselves but also introduce significant drift between multiple views. Even if a match is barely achieved, the final facade often exhibits issues such as receding and geometric distortion. While the industry has attempted post-processing repairs, such as using prior architectural lines to refine the model, these remedial methods are essentially "patching holes" and rarely address the root cause of the geometric stability of the reflective areas.

[0004] The issue of occlusion was also repeatedly raised. Components such as air conditioner outdoor units, awnings, and billboards can create complex spatial relationships in localized areas. Simply relying on conventional oblique photography makes it difficult to obtain sufficient parallax during the shooting process to explain these occlusion details, ultimately resulting in localized collapses or mis-stitching in the point cloud. Although some teams have tried to directly increase the shooting density, excessively dense flight paths incur additional costs and are often unsuitable for environments with narrow streets or small building spacing.

[0005] Overall, existing oblique photogrammetry techniques still lack reliable methods when dealing with areas lacking texture and sensitive to reflection, problems that are most common in building facades. Most systems can only maintain basic reconstruction quality in these areas through algorithmic fault tolerance, but struggle to obtain truly continuous, stable, and morphologically reliable facade geometry. This is precisely the main bottleneck currently facing the field of 3D building facade mapping. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for three-dimensional mapping of building facades based on oblique photogrammetry, so as to at least solve the problems of existing building facades having difficulty forming stable feature matching in areas with poor texture and being prone to geometric deviations in areas with sensitive reflection.

[0007] To achieve the above objectives, the first aspect of the present invention provides a method for three-dimensional mapping of building facades based on oblique photogrammetry. The method includes: acquiring multi-angle oblique images of the building facade, and performing region recognition processing based on the multi-angle oblique images of the building facade to generate region annotation data for characterizing the regional features of the building facade; adjusting acquisition control parameters based on the region annotation data to acquire fused image data that simultaneously reflects the natural texture information and structure light stripe information of the building facade; extracting natural texture feature points and structure light feature points based on the fused image data, and constructing feature point matching rules based on the region annotation data, and performing spatial matching on the natural texture feature points and structure light feature points based on the feature point matching rules to generate initial three-dimensional point cloud data of the building facade; identifying the geometric features of building facade components based on the initial three-dimensional point cloud data and constructing prior structural parameters of the components, optimizing the initial three-dimensional point cloud data based on the prior structural parameters of the components, and outputting the three-dimensional mapping result of the building facade.

[0008] Optionally, multi-angle tilted images of the building facade are acquired, and region recognition processing is performed based on the multi-angle tilted images of the building facade to generate region annotation data for characterizing the regional features of the building facade. This includes: extracting regional features for characterizing texture changes, brightness distribution, and geometric occlusion based on the multi-angle tilted images of the building facade; constructing region recognition rules based on the regional features; and identifying texture level regions, reflection regions, and occlusion regions of the building facade based on the region recognition rules to generate region annotation data containing region annotation data for characterizing texture level regions, reflection regions, and occlusion regions.

[0009] Optionally, region recognition rules are constructed based on region features, and texture level regions, reflection regions, and occlusion regions of the building facade are identified based on the region recognition rules. Region annotation data containing the texture level regions, reflection regions, and occlusion regions is generated, including: calculating the region grayscale gradient value, region brightness saturation value, and region contour abruptness value based on the region features; setting region recognition thresholds based on the region grayscale gradient value, region brightness saturation value, and region contour abruptness value; determining the texture level regions, reflection regions, and occlusion regions in the building facade image based on the region recognition thresholds, and generating region annotation data containing the texture level regions, reflection regions, and occlusion regions.

[0010] Optionally, the acquisition control parameters are adjusted based on the regional annotation data to obtain fused image data that simultaneously reflects the natural texture information and the structural light stripe information of the building facade. This includes: adjusting the acquisition control parameters, including the shooting exposure time, structured light projection intensity, and shooting angle, based on the regional annotation data; and simultaneously acquiring natural texture images reflecting the natural texture information of the building facade and structured light stripe images reflecting the structured light stripe information of the building facade based on the adjusted acquisition control parameters, thereby generating fused image data that simultaneously reflects the natural texture information and the structured light stripe information of the building facade.

[0011] Optionally, natural texture feature points and structured light feature points are extracted based on fused image data, and feature point matching rules are constructed based on region annotation data. This includes: identifying edge pixel regions used to characterize local texture changes on building facades based on fused image data, and extracting natural texture feature points in the edge pixel regions; identifying stripe coding regions used to characterize structured light stripe sequences on building facades based on fused image data, and extracting structured light feature points in the stripe coding regions; setting feature point matching rule parameters to limit the matching order and matching range of natural texture feature points and structured light feature points, and constructing feature point matching rules based on the feature point matching rule parameters to perform spatial correspondence determination of natural texture feature points and structured light feature points.

[0012] Optionally, the verification rules for the feature point matching rule parameters used to limit the matching order and matching range of natural texture feature points and structured light feature points are as follows: based on the region annotation data, the region priorities of texture level region, reflection region and occlusion region are determined respectively; based on the region priority, the matching order of natural texture feature points is determined to prioritize matching in texture level region and restrict the matching range in reflection region and occlusion region; based on the region priority, the matching order of structured light feature points is determined to prioritize matching in reflection region and occlusion region and restrict the matching range in texture level region.

[0013] Optionally, spatial matching is performed on natural texture feature points and structured light feature points based on feature point matching rules to generate initial 3D point cloud data of the building facade. This includes: selecting natural texture feature points and structured light feature points for spatial correspondence determination based on feature point matching rules; performing matching candidate pair filtering based on the spatial distribution characteristics of the selected natural texture feature points and structured light feature points to obtain multiple matching candidate pairs; calculating the spatial intersection result of natural texture feature points and structured light feature points based on each matching candidate pair and generating corresponding spatial coordinate values; and generating initial 3D point cloud data of the building facade based on the spatial coordinate values ​​of all matching candidate pairs.

[0014] Optionally, the geometric features of building facade components are identified based on the initial 3D point cloud data, and prior structural parameters of the components are constructed. This includes: extracting point cloud normal vectors and point cloud position distributions to characterize local geometric changes of the building facade based on the initial 3D point cloud data; performing identification of vertical linear components, horizontal linear components, and planar components of the building facade based on the point cloud normal vectors and point cloud position distribution; extracting component size parameters and component spatial constraint parameters to characterize geometric relationships based on the vertical linear components, horizontal linear components, and planar components of the building facade, and constructing prior structural parameters of the components based on the component size parameters and component spatial constraint parameters.

[0015] Optionally, the initial three-dimensional point cloud data is optimized based on the prior parameters of the component structure to output the three-dimensional mapping results of the building facade, including: determining the point cloud constraint conditions used to constrain the geometric relationship of the initial three-dimensional point cloud data based on the prior parameters of the component structure; performing point cloud position adjustment processing on the initial three-dimensional point cloud data based on the point cloud constraint conditions to obtain adjusted point cloud data used to characterize the spatial relationship of the building facade components, and generating the three-dimensional mapping results of the building facade based on the adjusted point cloud data.

[0016] A second aspect of the present invention provides a three-dimensional mapping system for building facades based on oblique photogrammetry. The system includes: an acquisition unit for acquiring multi-angle oblique images of the building facade and performing region recognition processing based on the multi-angle oblique images to generate region annotation data characterizing the regional features of the building facade; a control unit for adjusting acquisition control parameters based on the region annotation data to acquire fused image data that simultaneously reflects the natural texture information and structured light stripe information of the building facade; a processing unit for extracting natural texture feature points and structured light feature points based on the fused image data, constructing feature point matching rules based on the region annotation data, and performing spatial matching on the natural texture feature points and structured light feature points based on the feature point matching rules to generate initial three-dimensional point cloud data of the building facade; and an output unit for identifying the geometric features of building facade components based on the initial three-dimensional point cloud data and constructing prior structural parameters of the components, optimizing the initial three-dimensional point cloud data based on the prior structural parameters of the components, and outputting the three-dimensional mapping results of the building facade.

[0017] Through the above technical solution, this invention identifies regions and generates region annotation data from multi-angle tilted images. This allows for an understanding of the texture, reflectivity, and occlusion distribution of building facades during the acquisition phase, enabling differentiated acquisition control strategies for different regions. The fused images simultaneously contain natural textures and structured light stripes, ensuring stable feature points usable for spatial intersection even in areas with insufficient texture or excessive reflection. Feature point matching rules established based on region annotation data allow natural texture and structured light feature points to function within their respective advantageous regions, reducing feature drift and improving geometric consistency. After initial 3D point cloud generation, constraints on local geometric relationships are applied using prior structural parameters, further correcting morphological deviations in texture-poor areas and making the final facade model closer to the actual building form in terms of continuity and structural integrity. Overall, this invention effectively improves the stability and geometric reliability of 3D mapping results under complex facade conditions.

[0018] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the steps of a three-dimensional mapping method for building facades based on oblique photography provided in one embodiment of the present invention; Figure 2 This is a system structure diagram of a three-dimensional mapping system for building facades based on oblique photography, provided in one embodiment of the present invention. Detailed Implementation

[0020] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] like Figure 1 As shown, this invention provides a method for three-dimensional mapping of building facades based on oblique photogrammetry, the method comprising: Step S10: Acquire multi-angle tilted images of the building facade, and perform region recognition processing based on the multi-angle tilted images of the building facade to generate region labeling data to characterize the regional features of the building facade.

[0022] Specifically, regional features are extracted from multi-angle tilted images of building facades to characterize texture changes, brightness distribution, and geometric occlusion. Based on these regional features, regional recognition rules are constructed, and texture level regions, reflection regions, and occlusion regions of the building facades are identified based on these rules. This generates regional annotation data containing these regions to characterize texture level regions, reflection regions, and occlusion regions.

[0023] Furthermore, region recognition rules are constructed based on regional features, and texture level regions, reflection regions, and occlusion regions of the building facade are identified based on these rules. Region annotation data containing characterizations of texture level regions, reflection regions, and occlusion regions is generated, including: calculating regional grayscale gradient values, regional brightness saturation values, and regional contour abruptness values ​​based on regional features; setting region recognition thresholds based on these values; and determining texture level regions, reflection regions, and occlusion regions in the building facade image based on these thresholds, generating region annotation data containing characterizations of texture level regions, reflection regions, and occlusion regions.

[0024] In this embodiment of the invention, multi-angle tilted images are always a set of data sources with high information density. The geometric relationships of building facades usually exhibit significant parallax changes in these images, so region identification often begins with this batch of images. A quick quality check is first required to confirm that the exposure range, sharpness, and occlusion level are all within acceptable ranges. Subsequently, all images are placed under the same coordinate frame to examine texture undulations, brightness distribution, and the location of local geometric breakpoints. This step is similar to adding a preliminary semantic layer to the images, quickly determining which areas have stable information and which areas have potential risks.

[0025] Texture variation is often the primary focus of observation because building facades frequently exhibit large areas of repetitive or weakly textured textures. To quantify texture distribution, grayscale gradients are calculated within local windows to determine the persistence of texture direction and intensity. The magnitude of gradient changes often reflects material differences on the building surface; for example, paint, stone, and glass exhibit different gradient statistical characteristics. Brightness distribution is typically related to illumination and reflection. Significantly bright areas often indicate specular reflection or partial saturation, which frequently interferes with subsequent feature matching. Determining geometric occlusion is slightly more complex, usually requiring consideration of abrupt contour changes to identify structural edges or protruding objects, such as window frames, billboards, and pipelines, all of which can cause localized occlusion.

[0026] After obtaining a set of quantifiable indicators for the regional features, the next step is to construct region recognition rules. A set of thresholds is used to classify different types of regions, but these thresholds should not be fixed values. Instead, they should be adaptively adjusted based on the overall brightness level and texture distribution of different images. Specifically, the distributions of grayscale gradient, brightness saturation, and contour abruptness are statistically analyzed, and then the intervals used to classify texture level regions, reflection regions, and occlusion regions are determined based on these distributions. The thresholds here are not simply high or low, but are dynamically set based on the variance, kurtosis, or quantiles of the image content, making the rules more closely reflect the actual situation of the current acquisition.

[0027] The division of texture level regions primarily relies on grayscale gradient values. High gradient regions typically indicate rich texture, and these regions tend to be more stable in subsequent feature point extraction and spatial matching. Low gradient regions, on the other hand, may exhibit large areas of flat, uniform texture, requiring careful handling. Reflective regions are identified based on brightness saturation values; once the brightness exceeds the sensor's linear response range, pixels become saturated, and the area is classified as a reflective region. Occlusion regions are mainly determined by contour abruptness values. Locations with high abruptness and discontinuous distribution usually indicate overlapping structures, and these regions require separate processing.

[0028] Once all thresholds are set, region classification can be performed on the entire image. This involves sliding a fixed window across the image, row by row and column, comparing the statistical characteristics of each window with the thresholds, and determining the region category based on the matching results. To avoid abrupt changes caused by window boundaries, local smoothing techniques are incorporated, such as using neighborhood consistency to determine whether isolated blocks need to be reclassified. The resulting region annotation data often contains multiple layers of information, not only marking the locations of the three regions but also reflecting the distribution proportion of each region within the image.

[0029] The resulting region annotation data will serve as the foundation for subsequent processes. The annotation results provide a relatively clear region map, allowing subsequent data collection and control, feature extraction, and spatial matching to operate within a more defined context. Overall, this processing rule is quite stable in architectural facade surveying scenarios, especially when dealing with areas with insufficient texture, prominent reflections, or complex occlusion. Region recognition can expose potential problems early, reducing many unnecessary matching errors in subsequent steps.

[0030] Step S20: Adjust the acquisition control parameters based on the regional annotation data to obtain fused image data that simultaneously reflects the natural texture information and structural light stripe information of the building facade.

[0031] Specifically, based on the regional annotation data, the acquisition control parameters, including shooting exposure time, structured light projection intensity, and shooting angle, are adjusted; based on the adjusted acquisition control parameters, natural texture images reflecting the natural texture information of the building facade and structured light stripe images reflecting the structured light stripe information of the building facade are acquired simultaneously, generating fused image data that simultaneously reflects the natural texture information and the structured light stripe information of the building facade.

[0032] In this embodiment of the invention, the sensitivity differences between the light source and the shooting angle are significant, so a set of rules is needed to make local adjustments based on region classification. Texture level regions are considered relatively stable areas, and the exposure time for these regions is not significantly changed to maintain the true brightness and gradation of the natural texture. The processing of reflective areas requires caution; generally, the exposure time is appropriately reduced to avoid saturation banding. Occluded areas rely more on changes in the shooting angle; adjusting the angle reduces the overlap of foreground and background structures, making local geometric relationships easier to capture.

[0033] Structured light projection intensity is also a key parameter. Texture-level areas possess stable textures, so the structured light intensity should not be too high to avoid damaging the original texture details. Reflective areas, on the other hand, typically require increased projection intensity to ensure the stripes remain recognizable against a bright background. Structured light strategies for occluded areas are usually combined with viewing angle adjustments, moderately increasing the intensity to ensure the stripes fall as close to the visible surface as possible, rather than being completely obscured by the foreground.

[0034] After parameter adjustments are completed, image acquisition enters the synchronization phase. The purpose of synchronous acquisition is to ensure that natural texture images and structured light stripe images correspond to the same geometric instant, avoiding slight displacement caused by time differences. Specifically, this is usually achieved by binding shutter trigger signals to ensure that the two types of images record the optical information of the building facade almost simultaneously. Natural texture images record surface material and details, while structured light stripe images express local geometric relationships through the deformation of projected stripes. The combination of the two can form a relatively complete observation view under complex facade conditions.

[0035] The fused image data does not require depth processing; instead, it uses a combination of natural texture images and structured light stripe images as input for subsequent feature extraction. The significance of fusion lies in ensuring that the same spatial location simultaneously possesses both texture and stripe cues, providing two complementary information channels for subsequent feature point extraction. This way, areas lacking texture no longer rely entirely on natural texture, and reflectivity-sensitive areas do not need to be judged solely by brightness changes. The fused image achieves better information coverage across the overall structure, laying a more solid foundation for subsequent feature matching and 3D reconstruction.

[0036] In one specific implementation, the acquisition process utilizes a lightweight oblique photography setup. The combination consists of a camera with a zoom lens and a structured light module based on line stripe projection, both held relatively stable on a fixed support. The camera's tilt angle is set between 30° and 45°, an angle advantageous for capturing depth differences in the facade while minimizing information loss due to overhead obstruction. The structured light module's projection direction is essentially aligned with the camera's line of sight, allowing the stripes to form a clear, coded pattern on the facade. The entire setup is fixed to a sliding rail or tripod, and can be moved or raised to cover different heights depending on the building's dimensions.

[0037] Control strategies typically begin with area annotation data. The shooting exposure time and structured light projection intensity are set in segments by the control unit based on the annotation results; for example, texture-level areas maintain their original exposure, while reflective areas have their exposure appropriately shortened and stripe brightness increased. The shooting angle is adjusted by a tilt servo mechanism, which reduces geometric blind spots caused by occlusion by fine-tuning the elevation angle. The trigger signals of the camera and structured light module are linked through a set of synchronization circuits to ensure that both types of images are recorded on the building surface simultaneously.

[0038] During the data acquisition process, the device moves segment by segment along the trajectory. Upon reaching a preset shooting point, it completes a synchronous acquisition according to the current area's acquisition control parameters. The data is typically written directly to the storage module, along with metadata such as shooting angle, projection intensity, and location number, facilitating further analysis by associating it with labeled data in subsequent processing stages.

[0039] In another possible implementation, the acquisition device adds a controllable brightness auxiliary illumination unit to provide directional supplementary lighting in areas of unstable brightness. The auxiliary illumination unit is installed side-by-side with the structured light module, its illumination direction aligned with the camera's line of sight, ensuring that the supplementary lighting coverage corresponds to the imaging area. The illumination intensity is adjusted by an independent drive module, which automatically sets the supplementary lighting level for different areas based on the area labeling data.

[0040] Identifying low-brightness areas relies on a previously generated brightness saturation distribution map. The control process activates supplemental lighting upon entering these areas and stabilizes surface brightness with a fixed illuminance. The supplemental lighting level is not globally applied but triggered locally, providing illumination only to dark areas within the current shooting window. This is done to avoid large-scale changes in the light field and maintain the overall brightness structure of the building facade.

[0041] Once the supplemental lighting is applied, the process of simultaneously acquiring natural texture images and structured light stripe images remains unchanged, but the image quality is more stable. Deep window openings, the edges of recesses, and facade areas that have been obscured for extended periods reveal resolvable texture information under the supplemental lighting. The deformation of structured light stripes in dark areas is also easier to discern, and the stripe outlines are no longer obscured by noise.

[0042] This implementation is suitable for building surfaces with significant lighting variations. Many facades are in shadow for extended periods of the day, and the brightness in these areas is often below the camera's effective response range. The supplemental lighting unit provides additional light to these areas, keeping texture and striation information within a stable range. Images illuminated with this supplemental lighting exhibit less error diffusion during feature extraction and spatial matching, and are more conducive to subsequent 3D reconstruction.

[0043] Step S30: Extract natural texture feature points and structured light feature points based on fused image data, and construct feature point matching rules based on regional annotation data. Perform spatial matching on natural texture feature points and structured light feature points based on feature point matching rules to generate initial 3D point cloud data of building facade.

[0044] Specifically, natural texture feature points and structured light feature points are extracted based on fused image data, and feature point matching rules are constructed based on region annotation data. This includes: identifying edge pixel regions used to characterize local texture changes on building facades based on fused image data, and extracting natural texture feature points from these edge pixel regions; identifying stripe coding regions used to characterize structured light stripe sequences on building facades based on fused image data, and extracting structured light feature points from these stripe coding regions; setting feature point matching rule parameters to limit the matching order and range of natural texture feature points and structured light feature points, and constructing feature point matching rules based on these parameters to determine the spatial correspondence between natural texture feature points and structured light feature points.

[0045] Furthermore, the verification rules for the feature point matching rule parameters used to limit the matching order and matching range of natural texture feature points and structured light feature points are as follows: based on the region annotation data, the region priorities of texture level region, reflection region and occlusion region are determined respectively; based on the region priority, the matching order of natural texture feature points is determined to prioritize matching in texture level region and restrict the matching range in reflection region and occlusion region; based on the region priority, the matching order of structured light feature points is determined to prioritize matching in reflection region and occlusion region and restrict the matching range in texture level region.

[0046] Specifically, the initial 3D point cloud data of the building facade is generated by performing spatial matching of natural texture feature points and structured light feature points based on feature point matching rules. This includes: selecting natural texture feature points and structured light feature points for spatial correspondence determination based on feature point matching rules; performing matching candidate pair screening based on the spatial distribution characteristics of the selected natural texture feature points and structured light feature points to obtain multiple matching candidate pairs; calculating the spatial intersection result of natural texture feature points and structured light feature points based on each matching candidate pair and generating the corresponding spatial coordinate values; and generating the initial 3D point cloud data of the building facade based on the spatial coordinate values ​​of all matching candidate pairs.

[0047] In this embodiment of the invention, the processing objective at this stage is to transform the fused image data into stable feature correspondences that can be used for 3D reconstruction. Since the front-end has already provided region-annotated data and fused image data, the acquisition results will not be changed here. Instead, a standardized workflow is designed around feature point extraction and spatial matching. The entire process ultimately outputs initial 3D point cloud data of the building facade, providing a foundation for subsequent component prior constraints.

[0048] First, the structure of the fused image data needs to be clearly defined. The fused image consists of two parts: a grayscale or color channel recording the natural texture information of the building facade, and a stripe-coded channel recording the structured light stripe information. Each pixel location simultaneously carries features such as texture brightness, color, and stripe brightness variations. The region annotation data provides a region type label for each pixel or pixel block, including texture level regions, reflective regions, and occluded regions. Subsequent feature point extraction and matching rules rely on this annotation result for zoning control.

[0049] The extraction of natural texture feature points begins with local edge structures. In the fused image, gradient operations are first performed on the texture channels to calculate the grayscale gradient components of each pixel. A common approach is to calculate the difference in the horizontal and vertical directions to obtain the gradient components. and Based on this, the gradient magnitude is defined as: ; Where x and y represent the image plane coordinates. , This indicates the amount of grayscale change in the corresponding direction. This is used to measure the intensity of local texture changes. Gradient magnitude is thresholded and combined with non-maximum suppression to obtain a set of connected edge pixel regions, which are used to characterize the local texture contours of the building facade.

[0050] After identifying edge pixel regions, the extraction of natural texture feature points focuses on these regions. A common approach is to calculate corner response values ​​within fixed windows over the edge regions. For example, within each small window, the gradient direction distribution and gray-level covariance matrix are analyzed, and locations with larger corner response values ​​are selected as candidate feature points. Subsequently, non-maximum suppression and minimum spacing constraints are applied to the candidate feature points to avoid excessively dense feature clustering in local high-frequency texture regions. This round of selection results in a set of evenly distributed, stable natural texture feature points for subsequent matching.

[0051] The extraction of structured light feature points primarily focuses on stripe-coded regions. In the fused image, stripe-coded regions can be identified based on the alternating brightness characteristics of the stripe channels. For example, the brightness variation period along the scanning direction can be statistically analyzed to determine pixel bands that meet preset stripe period and contrast thresholds. For the identified stripe-coded regions, the brightness centroid is calculated on the cross-section of each stripe to obtain the sub-pixel position of the stripe centerline. Simultaneously, the projection order of each stripe is determined based on the stripe sequence number encoding information, enabling the differentiation of spatial positions corresponding to different stripes during subsequent 3D intersection calculations.

[0052] To obtain the set of structured light feature points, the stripe centerline needs to be discretized. Several points are sampled along the stripe direction at a fixed step size, and the image coordinates of each sampled point are recorded together with the corresponding stripe number to form the structured light feature points. This set is particularly important in regions with weak original texture, because in these regions the number of natural texture feature points is limited, while stripe variations are often clearer and more stable. The natural texture feature point set and the structured light feature point set overlap in spatial distribution and each has its own advantageous regions, providing a foundation for subsequent matching rules.

[0053] Feature point matching rules depend on the region type provided by the region annotation data. When constructing the rules, different matching priorities are assigned to each region type, and feature point matching rule parameters are formed accordingly. Region priority coefficients can be defined separately for texture level regions, reflection regions, and occlusion regions. , , The three coefficients are set based on practical experience. For example, a higher priority is given to natural textures in texture level regions, and a higher priority is given to structured light in reflection and occlusion regions. Through these coefficients, the feature point matching rule can automatically favor more reliable feature sources when selecting candidate matching pairs.

[0054] The feature point matching rule parameters, which define the matching order and range between natural texture feature points and structured light feature points, are calculated using a set of validation rules. First, based on the area proportions and distribution characteristics of various regions using region annotation data, the region priorities for texture level regions, reflection regions, and occlusion regions are determined. Then, the matching order for natural texture feature points is validated as prioritizing matching within texture level regions, while narrowing the matching search range within reflection and occlusion regions, for example, searching for candidate points only in smaller neighborhoods. For structured light feature points, a higher matching priority is set within reflection and occlusion regions, while they are used only as auxiliary information to limit the search space within texture level regions. In this way, different feature source combinations are used for the same spatial location under different region types, and the matching strategy exhibits significant region-adaptive characteristics.

[0055] When it is necessary to quantify the matching cost, a distance metric based on descriptor differences can be introduced. Let the descriptor of natural texture feature points be... The descriptor for structured light feature points is Then the basic matching distance between the two can be written as: ; in, and These are feature vectors obtained statistically within a local neighborhood, typically containing information such as brightness, gradient direction, or stripe shape. This represents the Euclidean norm. This distance measures the similarity in local appearance between natural texture feature points and structured light feature points, providing a numerical basis for matching candidate selection.

[0056] In the candidate pair selection stage, a target subset for spatial correspondence determination is first selected from the natural texture feature point set and the structured light feature point set according to the feature point matching rules. The selection process comprehensively considers region type, region priority, and local point density to avoid generating too many candidate pairs in low-reliability regions. Subsequently, within the neighborhood of each natural texture feature point, the descriptor distance is used to determine the target subset for spatial correspondence determination. Multiple candidate matching pairs are constructed based on geometric constraints (such as parallax direction or approximate epipolar location). Matching scores are calculated for these candidate pairs, and combinations that do not meet preset conditions are eliminated based on the scores and region priority, retaining a set of several reasonable candidate matching pairs.

[0057] The core of spatial matching lies in transforming image coordinates into three-dimensional spatial coordinates. For each retained matching candidate pair, the image coordinates of natural texture feature points and structured light feature points jointly participate in the spatial intersection calculation. Taking a two-view scenario as an example, the camera projection matrix of the two images can be denoted as... and The homogeneous image coordinates of the matching points are denoted as follows: and The homogeneous coordinates X of a point in space can be solved using the least squares method, that is, by finding the X with the smallest error in the following formula: ; in, , This is the projection matrix determined by the exterior and interior orientation elements. , Let X represent the observed location of the matching point in the two images, and let X represent the 3D coordinates of the point to be solved. Through this spatial intersection, corresponding spatial coordinate values ​​can be generated for each candidate matching pair.

[0058] In practice, spatial intersection results are calculated for all matching candidate pairs, and the resulting spatial point set undergoes a quality screening. For example, 3D points with excessively large reprojection errors can be removed to ensure the geometric accuracy of the initial 3D point cloud data. The filtered spatial coordinate set is then unified according to the building facade coordinate system or the external control point coordinate system to form the initial 3D point cloud data for the building facade. This point cloud exhibits a certain point density and geometric stability in texture level regions, reflection regions, and occlusion regions, providing a relatively complete input for the subsequent point cloud optimization process based on the prior parameters of the component structure.

[0059] This set of feature point extraction and spatial matching rules based on fused image data significantly changes the traditional approach's complete reliance on natural texture feature points. While natural texture feature points still play a major role in texture-level regions, structured light feature points have higher weight in reflective and occluded areas, forming a complementary relationship between the two types of feature points in spatial matching. The introduction of region annotation data and region priority allows the matching process to clearly perceive the reliability differences between different regions, making the matching strategy no longer globally uniform but targeted. The resulting initial 3D point cloud data is more continuous within the building facade area, and its geometry more closely resembles the actual building structure, laying a more solid foundation for the overall quality of 3D mapping.

[0060] Step S40: Identify the geometric features of building facade components based on the initial 3D point cloud data and construct the prior parameters of the component structure. Optimize the initial 3D point cloud data based on the prior parameters of the component structure and output the 3D mapping results of the building facade.

[0061] Specifically, the process involves identifying the geometric features of building facade components and constructing prior structural parameters based on initial 3D point cloud data. This includes: extracting point cloud normal vectors and point cloud position distributions to characterize local geometric changes in the building facade based on the initial 3D point cloud data; identifying vertical linear components, horizontal linear components, and planar components of the building facade based on the point cloud normal vectors and point cloud position distribution; extracting component size parameters and component spatial constraint parameters to characterize geometric relationships based on the vertical linear components, horizontal linear components, and planar components of the building facade; and constructing prior structural parameters for the components based on the component size parameters and component spatial constraint parameters.

[0062] Furthermore, the initial 3D point cloud data is optimized based on the prior parameters of the component structure to output the 3D mapping results of the building facade. This includes: determining the point cloud constraint conditions used to constrain the geometric relationship of the initial 3D point cloud data based on the prior parameters of the component structure; performing point cloud position adjustment processing on the initial 3D point cloud data based on the point cloud constraint conditions to obtain adjusted point cloud data used to characterize the spatial relationship of the building facade components; and generating the 3D mapping results of the building facade based on the adjusted point cloud data.

[0063] In this embodiment of the invention, the core idea of ​​this step is to combine the initial 3D point cloud with the geometric priors of the building components. Relying solely on the point cloud itself often results in significant geometric deviations due to noise, occlusion, and defects. By introducing structural priors, the point cloud is no longer treated as an isolated set of points, but rather constrained to a set of geometric relationships that more closely approximate the actual form of the building. The resulting 3D mapping is more stable in terms of structural integrity and scale consistency.

[0064] In the initial 3D point cloud stage, each point in the point cloud can be viewed as a sample on the building facade. To perceive local geometric changes, it is necessary to estimate the normal vector for each point. Specifically, a set of neighboring points is selected in the neighborhood of each point, and the coordinates of these neighboring points are used as samples to fit the local plane. Assume the coordinates of the neighboring points of a certain point are... By fitting the plane normal vector using least squares This yields the local normal directions. The local normal directions, along with the point distribution, constitute the fundamental data describing the local geometric features of the building facade. The point cloud distribution can be directly given by the three-dimensional coordinates of the points, or it can be organized using two-dimensional coordinates projected onto the facade reference plane.

[0065] Based on the point cloud normal vectors and their location distribution, we can begin to identify the main component types of a building facade. Vertical linear components generally correspond to a set of linear points close to the vertical direction, such as columns, sidewalls, and balcony edges. During identification, the normal vectors can be compared with the global vertical direction vectors to filter out point sets with small changes in normal vectors and an approximately straight-line distribution in the planar projection. The identification approach for horizontal linear components is similar, except that the reference direction is changed to the horizontal direction, such as floor dividing lines and window sill edges. Planar components usually correspond to large-area wall surfaces or curtain wall panels. The point cloud points roughly fall near the same plane in three-dimensional space, the normal vector directions are concentrated, and the distance between the points and the fitted plane is small.

[0066] In the process of geometric recognition, it is necessary to introduce some quantitative thresholds. For example, a threshold for the angle between the normal vector and the reference direction can be defined. When the angle between the normal vector of a candidate point set for a component and the vertical direction is within a preset range, the point set is marked as a candidate for a vertical linear component. Similarly, planar components can be filtered using a plane fitting residual threshold. Assuming the distance from a point to the plane after fitting is... When most points satisfy hour( (To the maximum allowable error), this area can be marked as a planar component. Through these simple yet reliable threshold controls, the component geometry category gradually stabilizes.

[0067] After identifying vertical linear components, horizontal linear components, and planar components, the component dimensional parameters and spatial constraint parameters can be extracted. Dimensional parameters primarily focus on length, height, thickness, and the spacing between components. For example, for vertical linear components, the three-dimensional coordinates of the upper and lower endpoints can be extracted, and the component height can be calculated using the coordinate difference; for horizontal linear components, the span on the wall surface can be calculated. Spatial constraint parameters focus more on the geometric relationships between components, such as whether the vertical spacing between adjacent floor lines is basically consistent, whether columns and floor slabs maintain an approximately orthogonal relationship, and whether planar components remain coplanar or have a fixed angle.

[0068] These dimensional parameters and spatial constraint parameters constitute the set of prior structural parameters for the components. These prior structural parameters can be statistically analyzed within a single building or generalized across multiple buildings within the same project. Within the same building, floor spacing, window opening pitch, and other parameters often exhibit significant repetition; such repetitive patterns are well-suited for constraining discrete points in the point cloud. For multiple similar buildings, a set of typical component size ranges can be extracted at a higher level, providing a unified reference for different facades. Here, the prior structural parameters act more like a set of rule templates, serving as a reminder in subsequent processing of which point cloud geometric relationships are reasonable and which deviations have exceeded acceptable limits.

[0069] After obtaining the prior parameters of the component structure, the point cloud optimization stage can begin. The goal of point cloud optimization is not to completely reshape the geometry, but rather to moderately adjust the positional distribution based on the original point cloud through a set of constraints. First, based on the prior parameters of the component structure, point cloud constraints are determined to constrain the geometric relationships of the initial 3D point cloud. Constraints may include distance constraints from points to the target plane, distance constraints from points to the target line, and angle constraints between components. For example, for a set of points belonging to a certain wall plane component, the distance between these points and the fitted plane can be limited to not exceeding a given threshold; for a set of points belonging to a certain column component, the lateral deviation of the points from the column axis can be limited to a certain range.

[0070] When constructing the optimization objective, a relatively direct approach can be adopted. Assume the 3D coordinates of the i-th point in the initial point cloud are... The optimized coordinates are Let set C represent the prior geometric constraints given for the component structure. An objective function can be constructed that comprehensively considers both the degree to which the point cloud deviates from the prior geometry and the degree to which the point cloud deviates from its original position. For example: ; in, These are balancing parameters used to control the adjustment range of the point cloud. Let these be the initial point coordinates. To optimize the point coordinates; in the second item Indicates the first The residual function of geometric constraints, such as the distance residual from a point to a plane, the distance residual from a point to a line, or the angular deviation. The weights correspond to the constraints. By minimizing this objective function, the point cloud can be made to better conform to the prior knowledge of the component structure while maintaining its original shape.

[0071] In the actual point cloud position adjustment process, an iterative approach is used to solve the aforementioned optimization problem. Each iteration estimates the constraint residuals based on the current point cloud, and then updates the point coordinates based on the residual gradient. For structural regions with strong constraints, such as large planar walls, greater weights can be assigned to make the point cloud more closely resemble the plane; for local areas with average data quality, the constraint weights can be reduced to avoid excessive distortion. The iteration termination condition can be set when the residual change falls below a certain threshold, or when the maximum number of iterations is reached. The final point cloud obtained is the adjusted point cloud data.

[0072] The adjusted point cloud data better reflects the overall component relationships of the building facade. Wall areas tend to be coplanar, and column and floor slab lines exhibit a clearer linear structure in space. Local noise points and outliers are weakened or absorbed into the overall structure. Based on this, different forms of 3D mapping results can be generated according to project requirements. For example, the point cloud can be directly output as a facade point set with coordinate attributes, or it can be further fitted into a polygonal mesh or a simplified component model. Regardless of the representation method used, the source data comes from a point cloud that has undergone structural prior constraints, resulting in relatively better geometric consistency.

[0073] This point cloud recognition and optimization process, based on prior knowledge of component structures, is well-suited for objects with high structural redundancy, such as building facades. Local deviations introduced by occlusion, reflection, or unfavorable viewing angles in the initial point cloud are corrected to a certain extent under the constraints of global component relationships. Key geometric elements on the facade, such as floor lines, column arrays, and curtain wall panel boundaries, present more coherent spatial relationships in the 3D results. The overall mapping results are more reliable in representing the building's form, providing a more solid geometric foundation for subsequent defect detection, facade renovation design, or digital asset management.

[0074] In another possible implementation, a multi-scale geometric consistency screening method is introduced to enhance the reliability of the prior parameters of the component structure. This method relies on a hierarchical scale framework to perform multi-level resampling on the initial 3D point cloud and independently estimate the geometric relationships of the component at different scales, thereby suppressing local geometric drift caused by viewpoint occlusion.

[0075] This implementation first constructs three resolution levels: fine-scale point sets, meso-scale point sets, and coarse-scale point sets. The fine-scale set maintains the original point density to capture local changes in the facade; the meso-scale set aggregates local areas through grid downsampling to stabilize the trend structure; and the coarse-scale set approximates the overall facade skeleton through strong downsampling. The normal vector distribution, the point-to-plane deviation sequence, and the local line segment trend are calculated for each of the three point sets, forming three independent sets of geometric candidate parameters.

[0076] Within this hierarchical framework, consistency scores are applied to vertical linear components, horizontal linear components, and planar components. The consistency score is obtained by comparing the component orientation deviations across the three scales. If the orientation at a certain scale deviates significantly from the overall trend, the data at that scale is set to low confidence and restricted from being included in the final component's prior structural parameters. The rationale for this approach is straightforward: occlusion areas fluctuate significantly at fine scales but often maintain structural stability at coarse scales; therefore, coarse scales can inversely correct for noise at fine scales.

[0077] In practice, a simple scoring expression can be constructed. For example, for a candidate vertical component, it can be defined as: ; in, This represents the maximum difference in the angles between the principal directions across the three scales. This is an adjustment factor. The score S is used to determine whether to accept the component candidate into the prior set.

[0078] After completing multi-scale evaluations of all component categories, the stability of the obtained prior structural parameters was significantly enhanced, especially in areas with weak texture such as glass curtain walls and louvered structures, where geometric drift was controlled. The resulting geometric relationships used to constrain the point cloud were more reliable, providing a more stable reference for subsequent position adjustments, and consequently improving the structural coherence of the overall 3D mapping results.

[0079] like Figure 2As shown, this invention provides a 3D mapping system for building facades based on oblique photogrammetry. The system includes: an acquisition unit for acquiring multi-angle oblique images of the building facade and performing region recognition processing based on the multi-angle oblique images to generate region annotation data characterizing the regional features of the building facade; a control unit for adjusting acquisition control parameters based on the region annotation data to acquire fused image data that simultaneously reflects the natural texture information and structured light stripe information of the building facade; a processing unit for extracting natural texture feature points and structured light feature points based on the fused image data, constructing feature point matching rules based on the region annotation data, and performing spatial matching on the natural texture feature points and structured light feature points based on the feature point matching rules to generate initial 3D point cloud data of the building facade; and an output unit for identifying the geometric features of building facade components based on the initial 3D point cloud data and constructing prior structural parameters of the components, optimizing the initial 3D point cloud data based on the prior structural parameters of the components, and outputting the 3D mapping results of the building facade.

[0080] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0081] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details described above. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe the various possible combinations.

[0082] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the embodiments of the present invention, they should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A method for three-dimensional mapping of building facades based on oblique photogrammetry, characterized in that, The method includes: Acquire multi-angle tilted images of the building facade, and perform region recognition processing based on the multi-angle tilted images of the building facade to generate region annotation data to characterize the regional features of the building facade. Based on the regional labeled data, the acquisition control parameters are adjusted to obtain fused image data that simultaneously reflects the natural texture information and structural light stripe information of the building facade. Natural texture feature points and structured light feature points are extracted based on fused image data, and feature point matching rules are constructed based on regional annotation data. Spatial matching of natural texture feature points and structured light feature points is performed based on feature point matching rules to generate initial 3D point cloud data of building facade. Based on the initial 3D point cloud data, the geometric features of the building facade components are identified and the prior parameters of the component structure are constructed. Based on the prior parameters of the component structure, the initial 3D point cloud data is optimized, and the 3D mapping results of the building facade are output.

2. The method for three-dimensional mapping of building facades based on oblique photogrammetry according to claim 1, characterized in that, Acquire multi-angle oblique images of the building facade, and perform region recognition processing based on these images to generate region annotation data characterizing the features of the building facade, including: Multi-angle tilted image extraction based on building facades is used to characterize regional features such as texture variation, brightness distribution and geometric occlusion. Based on regional features, region identification rules are constructed, and based on these rules, texture level regions, reflection regions, and occlusion regions of building facades are identified, generating region annotation data containing characterizations of texture level regions, reflection regions, and occlusion regions.

3. The method for three-dimensional mapping of building facades based on oblique photogrammetry according to claim 2, characterized in that, Region identification rules are constructed based on regional features, and texture level regions, reflection regions, and occlusion regions of building facades are identified based on these rules. Region annotation data containing characterizations of texture level regions, reflection regions, and occlusion regions is generated, including: Calculate the region grayscale gradient value, region brightness saturation value, and region contour abruptness value based on region features; The region recognition threshold is set based on the region grayscale gradient value, the region brightness saturation value, and the region contour abruptness value; Based on the region recognition threshold, the texture level region, reflection region, and occlusion region in the building facade image are determined respectively, and region annotation data containing the texture level region, reflection region, and occlusion region are generated.

4. The method for three-dimensional mapping of building facades based on oblique photogrammetry according to claim 1, characterized in that, Based on the regional annotation data, the acquisition control parameters are adjusted to obtain fused image data that simultaneously reflects the natural texture information and structural light stripe information of the building facade, including: Adjustment of acquisition control parameters, including shooting exposure time, structured light projection intensity, and shooting angle, based on region-labeled data; Based on the adjusted acquisition control parameters, natural texture images reflecting the natural texture information of the building facade and structured light stripe images reflecting the structured light stripe information of the building facade are acquired simultaneously to generate fused image data that simultaneously reflects the natural texture information and structured light stripe information of the building facade.

5. The method for three-dimensional mapping of building facades based on oblique photogrammetry according to claim 1, characterized in that, Natural texture feature points and structured light feature points are extracted from fused image data, and feature point matching rules are constructed based on region annotation data, including: Based on the fusion image data, edge pixel regions used to characterize local texture changes on building facades are identified, and natural texture feature points are extracted from the edge pixel regions. Based on the fusion image data, the stripe coding region used to characterize the structured light stripe sequence of the building facade is identified, and the structured light feature points are extracted from the stripe coding region. Set feature point matching rule parameters to limit the matching order and range of natural texture feature points and structured light feature points, and construct feature point matching rules based on feature point matching rule parameters to perform spatial correspondence determination of natural texture feature points and structured light feature points.

6. The method for three-dimensional mapping of building facades based on oblique photogrammetry according to claim 5, characterized in that, The validation rules for the feature point matching rule parameters used to limit the matching order and range between natural texture feature points and structured light feature points are as follows: Based on the region annotation data, the region priorities of texture level regions, reflection regions, and occlusion regions are determined respectively. Based on region priority, the matching order of natural texture feature points is determined to prioritize matching within the texture level region and limit the matching range within the reflection and occlusion regions. Based on region priority, the matching order of structured light feature points is determined to prioritize matching in reflection and occlusion regions and limit the matching range in texture level regions.

7. The method for three-dimensional mapping of building facades based on oblique photogrammetry according to claim 5, characterized in that, Based on feature point matching rules, spatial matching is performed between natural texture feature points and structured light feature points to generate initial 3D point cloud data for the building facade, including: Natural texture feature points and structured light feature points are selected based on feature point matching rules to perform spatial correspondence determination; Based on the spatial distribution features of the selected natural texture feature points and structured light feature points, a matching candidate pair screening process is performed to obtain multiple matching candidate pairs; Based on each matching candidate pair, the spatial intersection results of natural texture feature points and structured light feature points are calculated and the corresponding spatial coordinate values ​​are generated. Based on the spatial coordinate values ​​of all matching candidate pairs, the initial three-dimensional point cloud data of the building facade is generated.

8. The method for three-dimensional mapping of building facades based on oblique photogrammetry according to claim 1, characterized in that, Based on initial 3D point cloud data, geometric features of building facade components are identified and prior structural parameters of the components are constructed, including: Extract point cloud normal vectors and point cloud position distributions based on initial 3D point cloud data to characterize local geometric changes of building facades; Based on point cloud normal vectors and point cloud location distribution, perform vertical linear components, horizontal linear components and planar components of building facades; Based on the vertical linear components, horizontal linear components, and planar components of the building facade, component size parameters and component spatial constraint parameters are extracted to characterize the geometric relationships, and component structural prior parameters are constructed based on the component size parameters and component spatial constraint parameters.

9. The method for three-dimensional mapping of building facades based on oblique photogrammetry according to claim 8, characterized in that, Based on the optimization of the initial 3D point cloud data using the prior parameters of the structural components, the 3D mapping results of the building facade are output, including: The point cloud constraint conditions used to constrain the geometric relationships of the initial 3D point cloud data are determined based on the prior parameters of the component structure. Based on point cloud constraints, point cloud position adjustment processing is performed on the initial 3D point cloud data to obtain adjusted point cloud data for characterizing the spatial relationship of building facade components, and 3D mapping results of the building facade are generated based on the adjusted point cloud data.

10. A three-dimensional mapping system for building facades based on oblique photogrammetry, characterized in that, The system includes: The acquisition unit is used to acquire multi-angle tilted images of the building facade and perform region recognition processing based on the multi-angle tilted images of the building facade to generate region annotation data to characterize the regional features of the building facade. The control unit is used to adjust the acquisition control parameters based on the regional annotation data to obtain fused image data that simultaneously reflects the natural texture information and structural light stripe information of the building facade. The processing unit is used to extract natural texture feature points and structured light feature points based on fused image data, and to construct feature point matching rules based on regional annotation data. Based on the feature point matching rules, spatial matching is performed on the natural texture feature points and structured light feature points to generate the initial three-dimensional point cloud data of the building facade. The output unit is used to identify the geometric features of building facade components based on the initial 3D point cloud data and construct the prior parameters of the component structure. Based on the prior parameters of the component structure, the initial 3D point cloud data is optimized, and the 3D mapping results of the building facade are output.

Images