Real scene point cloud completion surveying and mapping method

Abstract

The present application relates to the field of three-dimensional surveying and mapping technology, and particularly relates to a real scene point cloud completion surveying and mapping method, which comprises the following steps: obtaining a three-dimensional laser point cloud of a survey area, constructing a voxel evidence field and a boundary evidence field, and determining a target evidence object; determining an evidence deficiency object and a supplementary survey pose according to evidence completeness, performing supplementary survey scanning and updating evidence completeness; constructing a real scene entity based on the target evidence object satisfying a threshold value, and performing rigid body registration to obtain a real scene entity library; performing counter-evidence review on a building information model based on the real scene entity library, generating a dynamic deviation field, and outputting a completion topographic map, a parking space map and an external facade map. The present application can improve the completeness of key object acquisition, the explainability of model verification and the consistency of result map output.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional surveying and mapping technology, specifically to a method for as-built surveying and mapping of real-world points using cloud-based methods. Background Technology

[0002] The accuracy of spatial data during the completion phase directly impacts project delivery, asset registration, quality verification, parking management, and facade compliance review. Current surveying methods commonly involve scanning along a pre-defined path, then uniformly stitching the point cloud data and comparing it with the design model. While this approach generates 3D results, it often relies on experience to handle critical boundaries, occluded areas, and repetitive structures. This can lead to insufficient local evidence, incomplete boundary closures, unstable entity correspondences, and insufficient interpretability of model verification results, ultimately affecting the consistency and verifiability of the output maps. Summary of the Invention

[0003] This invention provides a real-world cloud-based as-built mapping method to at least address the problems of how to achieve reliable as-built mapping and output of result maps under conditions where key boundaries are easily missing, entity relationships are easily confused, and model verification interpretability is insufficient.

[0004] In a first aspect, the present invention provides a method for as-built mapping of real-world locations using cloud-based methods, the method comprising: Acquire 3D laser point cloud of the survey area, construct voxel evidence field and boundary evidence field, and determine the target evidence objects corresponding to the as-built topographic map, parking space map and facade map; The completeness of evidence is calculated based on the voxel evidence field, the boundary evidence field and the target evidence object. Based on the completeness of evidence, the objects with insufficient evidence and the supplementary poses are determined. Supplementary scans are performed and the completeness of evidence is updated. Based on the target evidence objects whose evidence completeness meets the threshold, real-world entities are constructed, and rigid body registration is performed on the real-world entities based on local topological constraints and boundary and entity consistency constraints to obtain a real-world entity library. The building information model is subjected to a counter-evidence review based on the real-world entity database. A dynamic deviation field is generated based on the counter-evidence review results, and as-built topographic maps, parking space maps, and facade maps are output based on the dynamic deviation field.

[0005] In one possible implementation, the completeness of evidence is calculated based on the voxel evidence field, the boundary evidence field, and the target evidence object, including: calculating the completeness of evidence of the target evidence object based on the geometric continuity, boundary closure, multi-view support, and semantic decidability of the target evidence object.

[0006] In one possible implementation, determining the object of insufficient evidence and the supplementary pose based on the completeness of evidence includes: identifying the target evidence object whose completeness of evidence is less than a preset threshold as the object of insufficient evidence; determining the free space based on the voxel evidence field, and generating candidate supplementary poses based on the free space; and determining the supplementary pose based on the visibility quality of each candidate supplementary pose to the object of insufficient evidence, the importance of the object of insufficient evidence, and the pose cost.

[0007] In one possible implementation, performing a supplementary scan and updating the evidence completeness includes: performing a supplementary scan based on the supplementary pose to obtain a supplementary point cloud; updating the occupied state, free state, unknown state, and conflict state of each voxel unit in the voxel evidence field according to the supplementary point cloud; updating the boundary continuity state, boundary closure state, multi-view support state, semantically decidable state, and boundary conflict state in the boundary evidence field according to the supplementary point cloud; and recalculating the evidence completeness of the target evidence object based on the updated voxel evidence field and boundary evidence field.

[0008] In one possible implementation, constructing a real-world entity based on a target evidence object whose evidence completeness meets a threshold includes: geometrically segmenting the target evidence object whose evidence completeness meets the threshold to obtain facets and boundaries; classifying the facets and boundaries to obtain ground entities, facade entities, parking space boundary entities, opening boundary entities, and column entities; and constructing a real-world entity based on the ground entities, facade entities, parking space boundary entities, opening boundary entities, and column entities.

[0009] In one possible implementation, local topological constraints are used to characterize the spatial relationships between real-world entities. Local topological constraints include at least one of the following: adjacent entity category distribution, entity spacing relationship, entity angle relationship, boundary closure relationship, and elevation layering relationship. Boundary and entity consistency constraints are used to characterize the correspondence between the boundary and real-world entities. Boundary and entity consistency constraints include at least one of the following: surface consistency constraints, boundary continuity constraints, nesting relationship constraints, spacing consistency constraints, and closure relationship constraints.

[0010] In one possible implementation, rigid body registration of real-world entities is performed based on local topological constraints and boundary and entity consistency constraints to obtain a real-world entity library. This includes: establishing registration constraint relationships between real-world entities; using patch alignment relationships, boundary coincidence relationships, spacing consistency relationships, and boundary closure relationships as constraints of the registration constraint relationships; and solving the spatial pose of each real-world entity based on the registration constraint relationships to obtain the real-world entity library.

[0011] In one possible implementation, the building information model is subjected to a counter-evidence review based on a real-world entity database, including: matching real-world entities in the real-world entity database with model objects in the building information model; calculating the geometric consistency, boundary consistency, relational consistency, and functional consistency of the model objects based on the matching results; and determining the uncertainty of the model objects and the counter-evidence review results of the model objects based on the geometric consistency, boundary consistency, relational consistency, and functional consistency.

[0012] In one possible implementation, a dynamic deviation field is generated based on the results of the counter-evidence review, including: determining the normal deviation, tangential deviation, boundary deviation, and risk level corresponding to the model object based on the counter-evidence review results and the uncertainty of the model object, and generating a dynamic deviation field; the uncertainty of the model object is used to indicate the measurement reliability of the model object, and the risk level is used to indicate the compliance risk level of the model object.

[0013] In one possible implementation, the as-built topographic map, parking space map, and facade map are output based on the dynamic deviation field, including: extracting a first real-scene entity corresponding to the as-built topographic map, a second real-scene entity corresponding to the parking space map, and a third real-scene entity corresponding to the facade map based on the real-scene entity library and the dynamic deviation field; projecting the first real-scene entity onto the horizontal reference plane to generate the as-built topographic map; projecting the second real-scene entity onto the parking area reference plane to generate the parking space map; and projecting the third real-scene entity onto the facade reference plane to generate the facade map.

[0014] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows: By employing joint modeling techniques using voxel evidence fields and boundary evidence fields, we achieved complete quantification and directional supplementary measurement of key observation objects, improving the relevance of key boundary acquisition. Through local topological constraints and boundary-entity consistency constraints, we achieved stable registration and structured representation of real-world entities. Through counter-evidence review and dynamic deviation field techniques, we achieved interpretable output of model differences and linked generation of result maps. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the observation and supplementary measurement pose in the measurement area in a specific embodiment of the present invention; Figure 3 This is a graph showing the change in the completeness of the target evidence object in a specific embodiment of the present invention; Figure 4 This is a schematic diagram of the dynamic deviation field distribution in a specific embodiment of the present invention; Figure 5 This is a schematic diagram showing the output of the as-built topographic map, parking space map, and exterior elevation map in a specific embodiment of the present invention. Detailed Implementation

[0016] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0017] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0018] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0019] Point cloud as-built mapping is a technology that uses a set of spatial sampling points to digitally represent the true geometric shape, boundary distribution, and structural relationships of a target area. This technology can directly preserve real-world information such as ground surfaces, facades, openings, boundaries, and structure outlines, and can further form a physical data foundation for as-built verification. Unlike processing methods that only focus on single-location measurement or local dimension acquisition, point cloud as-built mapping emphasizes the continuous representation of the overall spatial form, accurate identification of key boundaries, and unified support for the generation of the final map. Based on this, how to establish a stable evidence organization method around the real-world data, and then complete entity construction, model verification, and final map output, becomes the key to further improving the practicality of point cloud as-built mapping. This invention revolves around the above-mentioned content.

[0020] like Figure 1 As shown, a method for as-built mapping of real-world points using cloud-based methods includes: Acquire 3D laser point cloud of the survey area, construct voxel evidence field and boundary evidence field, and determine the target evidence objects corresponding to the as-built topographic map, parking space map and facade map; In one embodiment, a three-dimensional laser scan of the survey area is first performed to obtain raw point cloud data covering the ground area, parking area, and building facade area. Before scanning, multiple scanning stations can be set up according to the scale of the survey area, occlusion conditions, and the scale of the results. Overlapping observation areas are retained between adjacent scanning stations to facilitate unified processing of point clouds from different stations later. After the raw point cloud enters the processing flow, it first undergoes time synchronization, merging of repeated echoes, and removal of obvious outliers. Then, coarse registration is performed according to a unified coordinate reference to obtain the three-dimensional laser point cloud of the survey area. This three-dimensional laser point cloud is not only a set of spatial coordinates but also retains basic attributes such as echo intensity, scanning angle, and inter-point neighborhood relationships, providing a data source for subsequent evidence field construction.

[0021] After obtaining the 3D laser point cloud of the survey area, the space of the survey area is discretized into voxels to construct a voxel evidence field. The voxel evidence field is used to express the observation state at different locations in the survey area. In specific implementation, the survey area is divided into multiple regular voxel units, and the voxel side length can be set according to the average spacing of the point cloud, the accuracy requirements of the target map, and the scale of the survey area. When the terrain undulation is small and the parking space boundaries are dense, a smaller voxel side length is taken; when the exterior facade is large and the overall outline is more important, a medium value is taken. For each voxel unit, the number of point clouds falling into the voxel unit, the stability of the point cloud distribution, and the repeated observations at multiple stations are counted. Combined with the laser beam crossing situation, the corresponding occupied state, free state, unknown state, or conflict state of the voxel unit is marked. The occupied state indicates that there is a stable entity supporting the location; the free state indicates that the location has been effectively observed but there is no entity surface; the unknown state indicates that there is a lack of reliable observations at the location; and the conflict state indicates that the observation results of different scanning stations at the same location are inconsistent. Through this processing, the discrete point cloud can be converted into a spatially interpretable observation state field.

[0022] Building upon the voxel evidence field, a boundary evidence field is further constructed. This boundary evidence field is used to express linear or planar boundary evidence that requires focus during map generation, rather than simply recording the existence of points. In implementation, ground elevation change zones, near-ground linear high-contrast regions, elevation normal change regions, and opening contour change regions are first extracted from the 3D laser point cloud of the survey area. These regions are then mapped to corresponding spatial units to form a set of boundary candidates. Subsequently, for each type of boundary candidate, the boundary continuity, boundary closure, multi-view support, and boundary category determination are recorded. Boundary continuity reflects whether there are breaks in the boundary; boundary closure reflects whether the boundary forms a complete contour; multi-view support reflects whether the boundary has been repeatedly observed from two or more stations; and boundary category determination reflects whether the boundary is closer to a terrain boundary, parking space boundary, lane boundary, elevation transition boundary, or opening boundary. For unstable boundaries caused by glass reflection, occluded edges, or local noise, the boundary evidence field can mark them as conflict boundaries to prevent them from directly participating in subsequent object determination.

[0023] After establishing the voxel evidence field and boundary evidence field, the target evidence objects are determined according to the requirements of the final deliverables. Target evidence objects are the smallest observation units used to support the generation of as-built topographic maps, parking space maps, and facade maps. For as-built topographic maps, target evidence objects may include continuous ground areas, site boundaries, road edges, slope boundaries, and elevation control locations; for parking space maps, target evidence objects may include parking space boundaries, lane boundaries, column outlines, and wall intrusion boundaries; for facade maps, target evidence objects may include the main facade outline, inter-floor boundaries, door and window opening boundaries, and facade turning boundaries. When determining target evidence objects, a forced fit is not performed directly based on the design model; instead, the spatial occupancy relationships in the voxel evidence field and the boundary stability relationships in the boundary evidence field are used as the basis for judgment. This process enables the formation of evidence organization results oriented towards map generation at the survey area level, providing a unified input for subsequent evidence completeness calculations, supplementary survey decisions, and real-world entity construction.

[0024] The completeness of evidence is calculated based on the voxel evidence field, the boundary evidence field and the target evidence object. Based on the completeness of evidence, the objects with insufficient evidence and the supplementary poses are determined. Supplementary scans are performed and the completeness of evidence is updated. In one embodiment, after acquiring the 3D laser point cloud of the survey area and determining the voxel evidence field, boundary evidence field, and target evidence objects, the process enters the evidence completeness calculation and supplementary measurement control stage. The core task of this stage is not simply determining whether the point cloud is sufficient, but rather whether the current observation results can support the stable generation of as-built topographic maps, parking space maps, and facade maps. Specifically, for each type of target evidence object, the spatial continuity, boundary closure, repeated observations, and category clarity are statistically analyzed to form an evidence completeness score. Then, based on the evidence completeness score, objects with insufficient evidence are filtered out, and supplementary measurement poses are generated in the free space corresponding to the voxel evidence field. After the supplementary measurement is completed, the supplementary point cloud is re-integrated into the voxel evidence field and boundary evidence field, updating the evidence completeness score of each target evidence object. If the evidence completeness meets the preset requirements, the current results are maintained and the process proceeds to the subsequent real-world entity construction stage; if the evidence completeness is still below the requirements, a new round of supplementary measurement is performed until the key target evidence objects meet the mapping requirements.

[0025] The completeness of evidence is calculated based on the voxel evidence field, the boundary evidence field, and the target evidence object, including: calculating the completeness of evidence based on the geometric continuity, boundary closure, multi-view support, and semantic decidability of the target evidence object.

[0026] In one embodiment, the calculation of evidence completeness is based on the correlation between the voxel evidence field, the boundary evidence field, and the target evidence object. The newly added limitation is that it does not use the number of points, point density, or coverage as the basis for judgment, but instead incorporates whether the target evidence object possesses the geometric and semantic conditions required for mapping into the calculation. Introducing this limitation can avoid misjudging "collection complete" when there are many local point clouds but the boundaries are not closed, there is insufficient repeated observation, or the category is unclear.

[0027] In practical implementation, the target evidence objects are first classified. When the target evidence object corresponds to an as-built topographic map, the focus is on extracting continuous ground areas, road edges, slope breaks, and site boundaries. When the target evidence object corresponds to a parking space map, the focus is on extracting parking space boundaries, lane boundaries, column outlines, and wall intrusion boundaries. When the target evidence object corresponds to an exterior facade map, the focus is on extracting the main facade outline, inter-floor boundaries, and opening boundaries. Subsequently, geometric continuity, boundary closure, multi-view support, and semantic decidability are calculated. Geometric continuity reflects the continuous spatial distribution of the target evidence object and can be determined by statistically analyzing the connectivity of effective point clouds in adjacent voxel units, the break length of boundary segments, and the proportion of interrupted patches. Boundary closure reflects whether the target evidence object forms a complete boundary and can be determined by judging the closure distance between boundary endpoints, the presence of obvious gaps in the outline, and the continuity of boundary turning points. Multi-view support reflects whether the target evidence object was repeatedly observed from multiple scanning stations and can be determined by statistically analyzing the number of point clouds corresponding to the target evidence object from different stations and their overlap consistency. Semantic decidability is used to reflect whether the category of the target evidence object is clear. It can be determined by the category stability, adjacency structure relationship and morphological characteristics of the target evidence object in the boundary evidence field.

[0028] For ease of standardized processing, the completeness of evidence can be calculated using the following expression:

[0029] in, For the completeness of evidence, For geometric continuity, For boundary closure, To provide support from multiple perspectives, For semantic decidability, , , and These are geometric continuity weight, boundary closure weight, multi-view support weight, and semantic decidability weight, respectively.

[0030] The principles for setting each weight are related to the map type. When the target evidence object is used for parking space map generation, the weights for boundary closure and multi-view support can be set to larger values; when the target evidence object is used for facade map generation, the weights for geometric continuity and semantic decidability can be appropriately increased; when the target evidence object is used for as-built topographic map generation, the weights for geometric continuity and boundary closure can be kept basically balanced. The value range of evidence completeness can be uniformly normalized to between zero and one. The closer the evidence completeness is to one, the closer the target evidence object is to the complete observation state required for map generation. The threshold for evidence completeness can be set according to the scale of the results, the importance of the boundary, and the complexity of the survey area. For example, for target evidence objects that directly affect the accuracy of the resulting map, such as site boundaries, parking space boundaries, and facade opening boundaries, the evidence completeness threshold can be set to a higher value; for local decorative boundaries or non-critical auxiliary boundaries, the evidence completeness threshold can be set to a relatively low value. Through the above processing, the evidence completeness can have a clear calculation basis and application boundaries.

[0031] The method for determining objects with insufficient evidence and supplementary poses based on evidence completeness includes: identifying target evidence objects with evidence completeness less than a preset threshold as objects with insufficient evidence; determining free space based on voxel evidence field and generating candidate supplementary poses based on free space; and determining supplementary poses based on the visibility quality of each candidate supplementary pose to the object with insufficient evidence, the importance of the object with insufficient evidence, and pose cost.

[0032] In one embodiment, the determination of objects with insufficient evidence and the supplementary measurement pose is a further development based on the evidence completeness calculation results. The added constraint is that instead of performing a comprehensive and repeated scan of the entire measurement area, targeted supplementary measurement is only performed on objects with insufficient evidence, while the supplementary measurement pose is constrained by visual and cost conditions. Introducing this constraint allows the supplementary measurement process to directly correspond to the preceding evidence judgment, reducing invalid data collection.

[0033] In practice, firstly, target evidence objects with a completeness level below a preset threshold are identified as insufficient evidence objects. Insufficient evidence objects can be further subdivided according to map type and risk level. If an insufficient evidence object belongs to a critical boundary such as a site boundary, parking space boundary, column intrusion boundary, or opening boundary, it is prioritized for inclusion in the supplementary measurement sequence; if the insufficient evidence object belongs to a non-critical auxiliary boundary, it can be processed after the supplementary measurement of critical objects is completed. Subsequently, the free space is determined based on the voxel evidence field. Free space refers to the spatial region that has been effectively traversed by the laser beam and is not occupied by any stable entity. The determination of free space can be accomplished through the connectivity of free-state voxel units in the voxel evidence field, eliminating regions that are too close to fixed obstacles, pose a risk of collision with existing entities, or are located in inaccessible positions. After obtaining the free space, candidate supplementary measurement poses are generated within the free space. Candidate supplementary measurement poses can be discretely arranged in the free space according to a preset step size, or can be preferentially generated along existing scanning channels, vehicle lanes, floor corridors, or facade observation zones. To avoid overly dense candidate supplementary poses, a minimum spacing can be set between adjacent poses to ensure that different candidate poses have effective differences in viewpoint. Then, for each candidate pose, its effectiveness in supplementing objects with insufficient evidence is evaluated. It can be seen that the quality is mainly determined by whether the target direction is obstructed, whether the observation distance is appropriate, whether the laser incident angle is conducive to boundary extraction, and whether the pose can form a supplementary viewpoint with the existing position.

[0034] The importance of objects with insufficient evidence can be categorized according to their impact on the resulting maps, with higher-impact objects assigned higher priority. Pose cost is used to reflect equipment movement distance, deployment difficulty, and on-site accessibility. For underground parking garage scenarios, areas where vehicle passageways intersect, the middle of the column grid, and near the ends of parking spaces can be considered priority candidate areas; for facade scenarios, ground locations with a suitable angle to the facade normal can be considered priority candidate areas; for terrain scenarios, areas outside the slope boundary and parallel to the road edge can be considered priority candidate areas. Finally, the supplementary measurement pose that balances visibility quality, importance, and pose cost is selected from the candidate supplementary measurement poses as the execution location for the next round of supplementary measurement scanning. This approach ensures a one-to-one correspondence between supplementary measurement poses and objects with insufficient evidence, guaranteeing that supplementary measurement actions have a clear objective rather than blindly increasing the amount of scan data.

[0035] Perform supplementary scanning and update the evidence completeness, including: performing supplementary scanning based on the supplementary pose to obtain supplementary point cloud; updating the occupied state, free state, unknown state, and conflict state of each voxel unit in the voxel evidence field according to the supplementary point cloud; updating the boundary continuity state, boundary closure state, multi-view support state, semantically decidable state, and boundary conflict state in the boundary evidence field according to the supplementary point cloud; and recalculating the evidence completeness of the target evidence object according to the updated voxel evidence field and boundary evidence field.

[0036] In one embodiment, the supplementary scanning and evidence integrity update are performed after the supplementary pose is determined. The added constraint is that the supplementary point cloud is not simply merged into the original point cloud, but is used to update the voxel evidence field and the boundary evidence field respectively, and then the evidence integrity of the target evidence object is recalculated based on the update results. Introducing this constraint allows the supplementary results to be directly reflected in subsequent judgments, forming a closed-loop process.

[0037] In practice, the device performs supplementary scanning according to the supplementary pose to obtain a supplementary point cloud. After entering the processing chain, the supplementary point cloud undergoes time synchronization with the initial point cloud, outlier removal, and duplicate echo merging. Then, it is locally aligned with the existing point cloud to avoid direct superposition of supplementary data causing local ghosting. After alignment, the voxel evidence field is updated using the supplementary point cloud. For voxel cells into which the supplementary point cloud falls, if the point cloud distribution is stable after repeated observations, the confidence level of the occupied state is increased; if the supplementary laser beam passes through a voxel cell but does not form a stable echo, the free state is strengthened; if no effective information is obtained after supplementary measurement, the unknown state is retained; if the supplementary measurement result is significantly inconsistent with the previous observation result within the same voxel cell, it is marked as a conflict state. Subsequently, the boundary evidence field is updated using the supplementary point cloud. If the supplementary point cloud extends the existing boundary endpoints, the boundary continuity state is updated; if the supplementary point cloud fills in the original boundary gaps, the boundary closure state is updated; if the supplementary point cloud comes from a new valid scanning station, the multi-view support state is updated; if the supplementary point cloud makes the boundary category clearer, for example, distinguishing the originally unstable boundary into a parking space boundary or a facade opening boundary, the semantically determinable state is updated; if there are still obvious inconsistencies in the boundary position after supplementary measurement, the boundary conflict state is maintained, awaiting further judgment.

[0038] After updating the voxel evidence field and boundary evidence field, the evidence completeness of the target evidence object is recalculated according to geometric continuity, boundary closure, multi-view support, and semantic decidability. If the evidence completeness of a target evidence object has reached a preset threshold, it is removed from the set of objects with insufficient evidence; if it has not yet reached the threshold, it is retained in the set of objects with insufficient evidence and enters the next round of supplementary pose selection. For target evidence objects where conflict persists, their priority in directly entering the real-world entity construction stage can be reduced in subsequent processing to avoid unstable boundaries participating in subsequent calculations prematurely. Through this "supplementary scan—dual-field update—completeness recalculation" approach, each supplementary scan can clearly change the subsequent processing state, thus forming an executable and verifiable closed-loop implementation path.

[0039] Based on the target evidence objects whose evidence completeness meets the threshold, real-world entities are constructed, and rigid body registration is performed on the real-world entities based on local topological constraints and boundary and entity consistency constraints to obtain a real-world entity library. In one embodiment, after completing the evidence completeness calculation and supplementary measurement update, the target evidence object whose evidence completeness reaches a set threshold is used as the input basis for entity construction, entering the real-scene entity generation and unified registration stage. This stage first extracts basic objects such as ground, facades, parking space boundaries, opening boundaries, and columns based on stable evidence objects, and then combines these basic objects into real-scene entities with clear spatial meaning. Subsequently, registration constraints are established based on the spatial adjacency, boundary closure, and correspondence relationships between real-scene entities. Rigid body registration is performed on real-scene entities formed from different scanning sources, unifying them within the same spatial coordinate framework, ultimately forming a real-scene entity library that can directly support the generation of as-built topographic maps, parking space maps, and facade maps. Through this processing, the preceding evidence objects no longer remain at the level of local boundaries or local point clouds, but are transformed into structured objects that can directly participate in mapping and review.

[0040] Constructing real-world entities based on target evidence objects whose evidence completeness meets a threshold includes: geometrically segmenting the target evidence objects whose evidence completeness meets the threshold to obtain facets and boundaries; classifying the facets and boundaries to obtain ground entities, facade entities, parking space boundary entities, opening boundary entities, and column entities; and constructing real-world entities based on the ground entities, facade entities, parking space boundary entities, opening boundary entities, and column entities.

[0041] In one embodiment, the construction of real-world entities is carried out under the premise that the completeness of the evidence reaches a threshold. The added limitation is that instead of directly classifying all point clouds indiscriminately, geometric segmentation and category determination are performed only on target evidence objects that have formed a stable observation basis. Then, real-world entities are formed by combining the base objects with clearly defined categories. This approach can, on the one hand, avoid local noise, occlusion residues, and boundary conflict areas from prematurely entering the entity construction process, and on the other hand, it also ensures that the real-world entities have a clear source and a traceable generation path.

[0042] In practice, the corresponding point cloud subsets are first extracted from target evidence objects whose evidence completeness meets the threshold, and then processed according to object type. For continuous ground areas and site boundaries, ground point cloud subsets are extracted primarily based on elevation continuity, local flatness, and boundary transitions. For parking space boundaries, lane boundaries, and column outlines, target point cloud subsets are extracted primarily based on near-ground distribution characteristics, linear distribution characteristics, and local contour abrupt changes. For facade main outlines, inter-floor boundaries, and opening boundaries, target point cloud subsets are extracted primarily based on normal distribution, facade transition positions, and contour changes. After obtaining the various target point cloud subsets, geometric segmentation is performed. Geometric segmentation can employ a patch growing method based on neighborhood normal consistency or a boundary extraction method based on curvature changes. When the point cloud distribution in the surface area is relatively flat, points whose normal difference and elevation difference between adjacent points are both within a set range are aggregated into the same patch. When the parking space boundary or facade boundary exhibits a clear linear banded distribution, points that satisfy the conditions of consistent direction and continuous spacing within a local neighborhood are aggregated into the same boundary. When the column outline exhibits a relatively independent vertical envelope, points with continuous vertical extension and stable lateral outline are aggregated into candidate column regions. After geometric segmentation, multiple patches and boundaries are obtained. Patches are mainly used to represent ground areas and facade areas, while boundaries are mainly used to represent parking space outlines, opening outlines, facade turning outlines, and site boundaries.

[0043] Subsequently, category determination is performed on the face pieces and boundaries. Ground entities are determined based on the characteristics of nearly horizontal face pieces, low elevation zones, and large continuous areas; facade entities are determined based on the characteristics of nearly vertical face pieces, continuous expansion in the facade direction, and connection to ground entities; parking space boundary entities are determined based on the characteristics of boundaries located near the ground level in the parking area, exhibiting a regular line segment or broken line distribution, and forming a matching relationship with lane boundaries; opening boundary entities are determined based on the characteristics of boundaries embedded in the facade area, having a closed or semi-closed outline, and matching the inter-floor location; column entities are determined based on the characteristics of point clouds being continuous in the vertical direction, having stable cross-sectional outlines, and forming a contact relationship with ground entities. After category determination, entities are combined according to spatial adjacency and boundary affiliation relationships. For example, continuous ground areas and adjacent boundaries are combined into site ground entities; the main facade outline, inter-floor boundaries, and opening boundaries are combined into facade real-world entities; and parking space boundaries, lane boundaries, and column outlines are combined into parking area real-world entities. For boundaries with localized gaps but stable overall structure, small-scale connections can be made based on the continuous distribution relationship of adjacent entities, provided that the threshold for evidence completeness is not exceeded. For faces or boundaries with obvious conflicts in category determination, they are not included in the current round of real-world entity construction but are reserved for subsequent verification. Through the above processing, real-world entities with clear origins, clear categories, and interpretable spatial structures can be formed, providing stable input for subsequent rigid body registration.

[0044] Local topological constraints are used to characterize the spatial relationships between real-world entities. Local topological constraints include at least one of the following: adjacent entity category distribution, entity spacing relationship, entity angle relationship, boundary closure relationship, and elevation layering relationship. Boundary and entity consistency constraints are used to characterize the correspondence between the boundary and real-world entities. Boundary and entity consistency constraints include at least one of the following: surface consistency constraint, boundary continuity constraint, nesting relationship constraint, spacing consistency constraint, and closure relationship constraint.

[0045] In one embodiment, local topological constraints and boundary-entity consistency constraints are reinforcing limitations introduced after the construction of real-world entities. The added constraint is that instead of using a single geometric distance relationship as the basis for determining whether entities correspond, both spatial relationships between entities and correspondences between boundaries and entities are introduced as the fundamental constraints for subsequent rigid body registration. The reason for introducing this constraint is that common repetitive structures in the survey area, such as underground parking garage column grids, standard parking space boundaries, regular facade openings, and continuous inter-story boundaries, can easily lead to mismatches if relying solely on local point proximity relationships. By supplementing with local topological constraints and boundary-entity consistency constraints, registration judgments can be based on structural relationships, thereby limiting the matching boundaries.

[0046] Specifically, local topological constraints are used to characterize the spatial relationships between real-world entities. The distribution of adjacent entity categories reflects the category composition of common entities surrounding a given real-world entity. For example, ground entities are typically surrounded by parking space boundary entities, column entities, or facade entities; facade entities are typically surrounded by opening boundary entities, inter-floor boundaries, and ground entities. Entity spacing relationships characterize the distance distribution between different real-world entities, such as the distance from a column entity to a parking space boundary entity, the distance from a facade entity to the site boundary, and the distance from a lane boundary to an adjacent parking space boundary. Entity angle relationships characterize the directional relationships between different boundaries or surfaces. For example, parking space boundaries and lane boundaries typically form approximately parallel or perpendicular relationships; facade entities on both sides of a facade turning boundary form a clear angle; and specific angular relationships can also form between site boundaries and road boundaries. Boundary closure relationships characterize whether several boundary entities can collectively enclose a complete area. For example, whether a parking space boundary can be closed into a single parking space outline, and whether an opening boundary can be closed into a door or window outline. Elevation layering is used to characterize the height level of an entity. For example, ground entities are at the lowest level, parking space boundary entities are distributed in the near-ground level, inter-level boundaries are located in a specific height zone in the middle of the facade, and opening boundaries are also within the corresponding floor range.

[0047] Boundary and entity consistency constraints characterize the correspondence between boundaries and real-world entities. Surface consistency constraints reflect whether boundaries are attached to the correct entity surfaces; for example, opening boundaries should be embedded within facade entities, and site boundaries should be located near the outer edge of ground entities. Boundary continuity constraints reflect whether boundaries of the same type extend continuously in space; for example, parking space boundaries should not abruptly break without cause, and inter-floor boundaries should remain continuous between adjacent facade panels. Nesting constraints reflect whether local boundaries are contained within the outline of larger entities; for example, opening boundaries are within the envelope of facade entities. Spacing consistency constraints reflect whether boundaries and entities maintain reasonable spacing; for example, the distance from a column outline to a parking space boundary should not significantly contradict the confirmed parking space dimensions. Closure constraints reflect whether several boundaries and adjacent entities together constitute a complete object; for example, whether the parking space outline formed by the parking space boundary, column outline, and wall boundary is complete. Local topological constraints and boundary and entity consistency constraints can be set according to the survey area type, common building structures, and the rules for representing results. When the underground parking garage is the main survey area, the relationships between entity spacing and boundary closure can be given greater importance; when the exterior facade is the main survey area, the relationships between entity angles, nesting relationships, and elevation layering relationships can be given greater importance. With the above constraints, subsequent rigid body registration no longer relies solely on the approximate overlap of point clouds, but is based on structural correspondences that are more in line with the actual scene.

[0048] Rigid body registration of real-world entities is performed based on local topological constraints and boundary and entity consistency constraints to obtain a real-world entity library. This includes: establishing registration constraint relationships between real-world entities; using patch alignment, boundary coincidence, spacing consistency, and boundary closure as constraints of the registration constraint relationships; and solving the spatial pose of each real-world entity based on the registration constraint relationships to obtain the real-world entity library.

[0049] In one embodiment, rigid body registration and the generation of the real-world entity library are performed after the real-world entities have been constructed and the registration constraints are clear. The added limitation is that the original point cloud is not directly registered as a whole. Instead, registration constraint relationships between real-world entities are established first, and then the spatial pose of each real-world entity is solved based on the registration constraint relationships. Using this approach, the registration process can be elevated from the point level to the entity level, reducing the interference of local noise and repetitive structures on the overall result.

[0050] In practice, candidate correspondences are first screened for real-world entities extracted from different scanning stations or local areas. During screening, priority is given to real-world entity pairs that are consistent in category, spatially similar, have compatible local topological relationships, and whose boundaries correspond to the entities. For example, the same column entity extracted from different stations should have similar vertical extent, similar horizontal contours, and consistent distribution of surrounding parking space boundaries; facade entities extracted from different stations should have similar normal directions, continuous inter-layer boundary positions, and matching opening boundary structures. After obtaining candidate correspondences, registration constraints are established. Registration constraints consist of multiple constraints. The patch alignment constraint is used to constrain similar patches to overlap as much as possible after registration; for example, identical ground entities or facade entities should maintain consistent patch orientation and continuous position after registration. The boundary coincidence constraint is used to constrain identical boundaries to overlap as much as possible after registration; for example, identical parking space boundaries, opening boundaries, or site boundaries should be consistent along their main directions after registration.

[0051] The spacing consistency constraint is used to ensure that the feature distances between entities remain stable before and after registration. For example, the distance between a cylindrical entity and the boundary entity of an adjacent parking space should not show significant abnormal changes after registration. The boundary closure constraint is used to ensure that the contours formed by multiple boundaries and adjacent entities remain closed after registration. For example, the contours of parking spaces, openings, and the site should not show significant gaps or overlaps after registration. After establishing the registration constraints, the spatial pose of each real-world entity is solved. The spatial pose includes translational position and rotational attitude, and the entire registration process only allows rigid body transformations, without introducing scale scaling or deformation correction. This is because as-built mapping scenes require maintaining the true size and relative relationships of entities; introducing scale changes can easily affect the accuracy of subsequent mapping.

[0052] During the solution process, stable, clearly defined, and frequently observed real-world entities can be used as initial references, such as large-scale ground entities, continuous facade entities, or stable column entities. Other real-world entities are then gradually incorporated into a unified coordinate frame. For a candidate correspondence, if the main conditions of patch alignment, boundary coincidence, spacing consistency, and boundary closure cannot be simultaneously met after registration, the candidate correspondence is discarded to avoid the propagation of incorrect registration. After completing all spatial pose solutions, the real-world entities unified within the same coordinate frame are merged and organized to generate a real-world entity library. This library not only stores the spatial location and category information of each real-world entity but also its connection relationships, boundary relationships, and source station information with other entities. This process allows for direct access to structured, coordinate-unified entity results during subsequent building information model (BIM) verification and drawing output stages, eliminating the need to revert to the original point cloud for repeated analysis.

[0053] The building information model is subjected to a counter-evidence review based on the real-world entity database. A dynamic deviation field is generated based on the counter-evidence review results, and as-built topographic maps, parking space maps, and facade maps are output based on the dynamic deviation field.

[0054] In one embodiment, after the real-world entity library is constructed, the process enters the results review and map output stage. This stage first establishes associations between various real-world entities in the real-world entity library and their corresponding model objects in the Building Information Model (BIM). Then, a counter-evidence review is conducted focusing on geometric location, boundary location, spatial relationships, and functional usage. This counter-evidence review verifies the BIM based on the real-world entities, rather than using the BIM to constrain the real-world entities. After verification, the results are further organized into a dynamic deviation field. The dynamic deviation field is used to uniformly express the deviation states of different model objects at the location, boundary, and risk levels. Subsequently, combining the real-world entity library and the dynamic deviation field, entity sets for as-built topographic maps, parking space maps, and facade maps are extracted and projected onto the corresponding reference planes to generate three types of output maps. Through this process, the previous point cloud processing results can be transformed into verifiable, interpretable, and directly output as-built surveying results.

[0055] The process of conducting a counter-evidence review of a Building Information Model (BIM) based on a real-world entity database includes: matching real-world entities in the database with model objects in the BIM; calculating the geometric consistency, boundary consistency, relational consistency, and functional consistency of the model objects based on the matching results; and determining the uncertainty of the model objects and the counter-evidence review results based on the geometric consistency, boundary consistency, relational consistency, and functional consistency.

[0056] In one embodiment, the rebuttal review is completed based on the established correspondence between the real-world entity database and the Building Information Model (BIM). The added limitation is that instead of directly judging consistency by overlaying the entire model, the review verifies each model object individually across four categories: geometry, boundaries, relationships, and functions. This process is then used to synthesize the model object uncertainty and the rebuttal review result. This approach breaks down the differences between different objects in the model to a level that can be judged independently, preventing local errors from masking overall problems.

[0057] In practice, the Building Information Model (BIM) is first objectified and decomposed. The decomposed objects can include site boundary objects, road objects, ground objects, parking space objects, driveway objects, column objects, wall objects, facade objects, opening objects, and inter-floor boundary objects. After decomposition, matching relationships are established between real-world entities in the real-world entity library and model objects in the BIM, based on object category and spatial location. Matching prioritizes category consistency, proximity, and contour compatibility as fundamental conditions. For example, ground entities are prioritized for candidate relationships with ground objects, road objects, and site boundary objects; parking space boundary entities, column entities, and wall boundaries are prioritized for candidate relationships with parking space objects and driveway objects; facade entities, opening boundary entities, and inter-floor boundaries are prioritized for candidate relationships with facade objects, opening objects, and inter-floor boundary objects. For cases where multiple candidate objects exist within the same category, further comparisons of circumscribed contours, relative elevations, and adjacency relationships can be made to select the model object that best matches the actual correspondence.

[0058] After establishing the matching relationship, geometric consistency is calculated first. Geometric consistency mainly reflects the proximity of real-world entities and model objects in terms of position, orientation, and size. Geometric consistency is high when ground entities and ground objects are similar in elevation distribution and range; it is also high when the center position and outline dimensions of column entities and column objects differ little; and it is high when the main direction and projection position of facade entities and facade objects are similar. Next, boundary consistency is calculated. Boundary consistency mainly reflects the correspondence between real-world boundaries and model boundaries, such as whether the parking space boundary matches the outline of the parking space object, whether the opening boundary matches the outline of the opening object, and whether the site boundary matches the site boundary object. If the boundary has undergone overall translation but the shape remains consistent, it can be determined that the boundary consistency has decreased rather than being completely inconsistent. Then, relational consistency is calculated. Relational consistency is used to determine whether the structural relationship between model objects and surrounding objects is supported by real-world entities, such as whether the spacing between columns and parking space boundaries is reasonable, whether the nesting relationship between facade objects and opening objects is valid, and whether the adjacency relationship between site boundary objects and road objects is valid. Finally, functional consistency is calculated. Functional consistency emphasizes how well model objects conform to actual usage, such as whether parking space objects still meet the boundary conditions for parking, whether lane objects still meet the traffic width requirements, and whether opening objects are located in reasonable elevation positions. After completing the four types of consistency calculations, the uncertainty of model objects is further determined.

[0059] Model object uncertainty reflects the reliability of the current review conclusion. Its determination is primarily based on factors including the coverage of surrounding real-world entities, repeated observations from multiple stations, boundary closure, and the presence of conflicting areas. If a model object is surrounded by complete real-world entities, has continuous boundaries, and has undergone sufficient repeated observations, its uncertainty is low; conversely, if a model object is surrounded by occlusions, boundary gaps, or conflicting boundaries, its uncertainty is high. Finally, by combining geometric consistency, boundary consistency, relational consistency, functional consistency, and model object uncertainty, a counter-evidence review result is formed. This counter-evidence review result can be categorized as consistent, divergent, pending verification, or inconsistent. This itemized verification process ensures that each model object in the Building Information Modeling (BIM) has corresponding real-world evidence and verification conclusions, facilitating subsequent deviation expression and output.

[0060] The dynamic deviation field is generated based on the results of the counter-evidence review, including: determining the normal deviation, tangential deviation, boundary deviation and risk level corresponding to the model object based on the counter-evidence review results and the uncertainty of the model object, and generating the dynamic deviation field; the uncertainty of the model object is used to indicate the measurement reliability of the model object, and the risk level is used to indicate the compliance risk level of the model object.

[0061] In one embodiment, the dynamic deviation field is generated after the evidence of dissent review is completed. The added limitation is that it does not directly output a single distance difference value, but instead establishes a joint expression of normal deviation, tangential deviation, boundary deviation, and risk level around the model object. This allows the deviation results to reflect both geometric differences and the impact at the usage and review levels. After this processing, the dynamic deviation field is no longer just a static comparison result, but a unified description result for subsequent map generation and risk warning.

[0062] In practice, the sources of deviation are first determined based on the results of the counter-evidence review and the uncertainty of each model object. For ground objects, road objects, and site boundary objects, deviations mainly originate from elevation changes, boundary offsets, and local contour deformations; for parking space objects and lane objects, deviations mainly originate from boundary position changes, local intrusions, and changes in usable space; for facade objects and opening objects, deviations mainly originate from facade position changes, opening contour offsets, and inter-story boundary misalignments. Subsequently, normal deviation, tangential deviation, and boundary deviation are determined for each model object. Normal deviation reflects the deviation of the real-world entity from the model object in the principal normal direction. For example, when a facade entity bulges outward or contracts inward, it can be represented by normal deviation; when a ground entity is higher or lower than the design elevation, it can also be reflected by normal deviation.

[0063] Tangential deviation is used to reflect the translational or misaligned changes of real-world entities within the main surface. For example, the overall translation of an opening boundary within the elevation plane, or the positional shift of a parking space boundary within the parking area plane, can both be represented by tangential deviation. Boundary deviation is used to reflect the difference between the outline boundary itself and the model boundary. For example, local deviations or narrowing / widening of the site boundary, parking space boundary, lane boundary, and opening boundary can be represented by boundary deviation. After determining the deviation values, the risk level is determined based on the magnitude of the deviation, the purpose of the object, and the uncertainty of the model object. The principle for setting the risk level is based on whether it affects the expression of the as-built results, whether it affects the judgment of spatial boundaries, and whether it affects the functionality.

[0064] Model objects with only slight positional changes that do not affect map representation and functional judgment can be classified as low-risk. For models with significant boundary deviations but high uncertainty, they can be classified as medium-risk, with room for subsequent manual verification. Model objects that have already affected parking space usage boundaries, site boundary judgments, or facade opening location judgments can be classified as high-risk. Model object uncertainty is used here to adjust the risk level, avoiding premature conclusions based on insufficient observation data. For example, if a facade opening boundary has a significant deviation, but the area is also affected by occlusion and insufficient repeated observations, the risk level can be maintained at an intermediate level and marked on the map. After the dynamic deviation field is generated, it can be organized by spatial location or by object category. When organized by spatial location, it is suitable for direct overlay onto topographic maps, parking space maps, and facade maps; when organized by object category, it is suitable for outputting site object deviation results, parking object deviation results, and facade object deviation results separately. Through this processing, the dynamic deviation field retains the specific differences at the model object level while possessing the ability for unified output and unified invocation.

[0065] The system outputs as-built topographic map, parking space map, and facade map based on the dynamic deviation field. This includes: extracting the first real-scene entity corresponding to the as-built topographic map, the second real-scene entity corresponding to the parking space map, and the third real-scene entity corresponding to the facade map from the real-scene entity library and the dynamic deviation field; projecting the first real-scene entity onto the horizontal reference plane to generate the as-built topographic map; projecting the second real-scene entity onto the parking area reference plane to generate the parking space map; and projecting the third real-scene entity onto the facade reference plane to generate the facade map.

[0066] In one embodiment, the output of the as-built topographic map, parking space map, and facade map is completed through the combined action of a reality entity database and a dynamic deviation field. The added limitation is that the graphics are not directly extracted from the original point cloud. Instead, reality entities corresponding to different output maps are first extracted from the reality entity database, and then combined with the dynamic deviation field for projection representation and deviation marking. This approach ensures that the object boundaries and deviation information in the output maps all originate from entities that have undergone verification and unified positioning.

[0067] In practice, the following steps are taken: First, extract the real-world entities corresponding to the as-built topographic map. These entities typically include ground entities, site boundaries, road boundaries, slope boundaries, and elevation-related boundaries. After extraction, these entities are projected onto a horizontal datum plane. The horizontal datum plane can be determined based on the as-built survey plane datum used in the survey area. During projection, the planar positional relationships of the site boundaries are preserved, and slope boundaries and road boundaries are distinguished and expressed. For deviation results related to site boundaries and road objects in the dynamic deviation field, they can be expressed in the as-built topographic map as additional markers, deviation lines, or attribute annotations. Next, extract the real-world entities corresponding to the parking space map. These entities typically include parking space boundary entities, lane boundary entities, column entities, and wall boundaries related to the parking area.

[0068] After extraction, these entities are projected onto the parking area reference plane. The parking area reference plane can be the horizontal reference plane of the floor where the parking area is located. During projection, the relative relationships of the parking space outline, driveway outline, and column positions are maintained first. For deviation results related to parking space objects, driveway objects, and column objects in the dynamic deviation field, the locations where boundary intrusion, boundary offset, or functional impact occurs can be simultaneously marked on the parking space map, so that the parking space map not only shows the boundary outline but also the actual deviation state. Then, the real-world entities corresponding to the facade drawing are extracted. These real-world entities typically include facade entities, inter-floor boundaries, opening boundaries, and facade turning boundaries. After extraction, these entities are projected onto the facade reference plane. The facade reference plane can be set according to the overall orientation of the main facade of the building.

[0069] For transitional facades, corresponding facade reference planes can be established segment by segment. During projection, the relative positional relationship between opening boundaries and inter-story boundaries is maintained, and deviation marks are made based on the deviation results related to facade objects, opening objects, and inter-story objects in the dynamic deviation field. Deviation marks can be placed near the corresponding contours to facilitate direct identification of deviation locations and risk levels in the exterior facade drawings. After completing the projection and deviation representation of the three types of maps, the maps are then uniformly organized, including cleaning up duplicate boundaries, merging collinear boundaries, unifying object category identifiers, and outputting corresponding map attribute information. The resulting as-built topographic maps, parking space maps, and exterior facade drawings not only contain the spatial representation results of real-world entities but also the deviation representation results after verification by the building information model, thus forming the final deliverables for as-built surveying applications.

[0070] In a specific embodiment, the execution process of this invention is illustrated using a completed surveying project of a commercial complex as an example. The survey area includes an outdoor site area, an underground parking area, and an east facade area. The outdoor site area measures 86 meters by 54 meters, the underground parking area measures 42 meters by 28 meters, and the east facade area has a width of 36 meters and a height of 18 meters. The Building Information Model (BIM) pre-defines 8 site boundary objects, 4 road boundary objects, 12 parking space objects, 2 lane objects, 2 column objects, 3 facade objects, 12 opening objects, and 2 inter-floor boundary objects. Fourteen initial scanning poses are initially set up on-site, with 6 in the underground parking area, 5 in the outdoor site area, and 3 in the east facade area. The number of original sampling points per station ranges from 5.4 million to 6.7 million, for a total of 84.6 million original points. After time synchronization, duplicate echo merging, and outlier removal, 79.3 million effective point clouds are obtained. The voxel side length is set according to the region: 0.20 meters for the outdoor area, 0.10 meters for the underground parking area, and 0.08 meters for the east facade area. The principle for setting the side length is to ensure that the terrain boundary, parking space boundary, and opening boundary can all fall within at least two consecutive cells, and to avoid the boundary information being flattened due to excessively large voxels.

[0071] Figure 2 This is a schematic diagram of the observation and supplementary measurement poses in the survey area, generated based on the initial scanning pose, the supplementary measurement pose, and the spatial position of the key target evidence object. Figure 2 The left side shows the initial scanning pose, the supplementary measurement pose, and the positional relationship of the site boundary inflection point, parking space boundary P12, column intrusion boundary C3, and lane boundary L2 in the outdoor site area and the underground parking area. Figure 2 The right side shows the positional relationship between the supplementary survey pose of the east facade area and the opening boundary W7. Based on the voxel evidence field and the boundary evidence field, a total of 34 target evidence objects were identified, including 10 corresponding to the as-built topographic map, 14 corresponding to the parking space map, and 10 corresponding to the facade map. The evidence completeness of each target evidence object was calculated using the expression:

[0072] in, For the completeness of evidence, For geometric continuity, For boundary closure, To provide support from multiple perspectives, Semantic decidability is defined as follows: Geometric continuity is determined by the connectivity ratio of the target evidence object in adjacent voxel units; boundary closure is determined by the ratio of the boundary gap length to the total length of the target boundary; multi-view support is determined by the number of effective scan poses and the consistency of repeated observations; and semantic decidability is determined by the stability of boundary categories and the stability of adjacency relationships. The evidence completeness threshold is set to 0.85, the principle of which is to ensure that the target evidence object can directly support subsequent entity construction without the need for manual edge patching.

[0073] Taking four representative target evidence objects as examples, after the initial scan, the geometric continuity of the site boundary inflection points is 0.82, the boundary closure is 0.76, the multi-view support is 0.67, and the semantic decidability is 0.85, resulting in an evidence completeness of 0.7765. The four indicators for the parking space boundary P12 are 0.71, 0.58, 0.52, and 0.80, with an evidence completeness of 0.656. The four indicators for the column intrusion boundary C3 are 0.69, 0.51, 0.48, and 0.76, with an evidence completeness of 0.615. The four indicators for the opening boundary W7 are 0.74, 0.55, 0.61, and 0.72, with an evidence completeness of 0.655. Thus, seven objects with insufficient evidence were identified. Nine candidate supplementary measurement poses were then generated based on free space. Considering visibility quality, object importance, and pose cost, five supplementary measurement poses were selected, including four for the underground parking area and one for the east facade area. The first round of supplementary testing added 16.2 million valid point clouds. After updating the voxel evidence field and boundary evidence field, the evidence completeness of the site boundary vertices improved to 0.892, parking space boundary P12 to 0.822, column intrusion boundary C3 to 0.778, and opening boundary W7 to 0.824. Since the latter three objects were still below the threshold, a second round of supplementary testing was performed, adding 5.1 million valid point clouds. Finally, the evidence completeness of parking space boundary P12, column intrusion boundary C3, and opening boundary W7 improved to 0.924, 0.895, and 0.916, respectively. Figure 3 The graph shows the change in the completeness of the target evidence objects, which is drawn based on the evidence completeness data of the four representative target evidence objects after the initial scan, the first round of supplementary testing, and the second round of supplementary testing. Figure 3The dashed line represents the integrity threshold of 0.85. It can be seen that none of the four objects reached the threshold after the initial scan. After the first round of supplementary testing, only the site boundary inflection point reached the threshold. After the second round of supplementary testing, all four objects reached the threshold. This indicates that the supplementary testing process did indeed play a role in addressing objects with insufficient evidence, rather than simply increasing the number of scan points.

[0074] After the evidence completeness reached the threshold, the construction of real-world entities began. Geometric segmentation and category determination were performed on target evidence objects that met the threshold, resulting in 18 ground entities, 11 facade entities, 12 parking space boundary entities, 12 opening boundary entities, and 2 column entities. Registration constraints were then established based on the distribution of adjacent entity categories, entity spacing relationships, entity angle relationships, boundary closure relationships, and elevation layering relationships, and rigid body registration was performed on real-world entities from different scanning sources. Before registration, the average positional residual between adjacent entities in the parking area was 0.071 meters, and the average positional residual between adjacent entities in the east facade area was 0.064 meters; after registration, these values ​​decreased to 0.019 meters and 0.021 meters respectively, forming a unified real-world entity library of 55 entities. Subsequently, the real-world entity library was matched and reviewed against the building information model. The results showed 48 consistent entities, 5 discrepancies, and 2 entities requiring further verification. Among the deviations, the eastern boundary of the site shifted outward by 0.042 meters; the net width of parking space P12 decreased from the design value of 2.50 meters to 2.37 meters due to column intrusion; the net width of parking space P11 decreased from 2.50 meters to 2.43 meters; the normal deviation of the opening W7 on the east facade was 0.062 meters, and the opening width decreased from the design value of 1.80 meters to 1.74 meters; the third-floor inter-floor boundary shifted upward by 0.034 meters. The objects to be checked are mainly located in the reflective glass area of ​​the east facade because although repeated observations have been conducted twice, the boundary closure is still insufficient.

[0075] Figure 4 This is a schematic diagram of the dynamic deviation field distribution, generated based on the spatial location of the deviation object, the object's uncertainty, and the risk level. Figure 4 The left side shows the risk level distribution of the parking area, with P12 having the highest risk level because the column intrusion caused a reduction of 0.13 meters in the net width, which exceeds the 0.10-meter parking boundary tolerance set in this embodiment; P11 and P7 have the next lowest risk levels, and the remaining parking spaces are in a low-risk state. Figure 4 The right side shows the distribution of normal deviations in the east facade area, with the largest normal deviation (0.062 meters) at location W7. Risk levels are set in three tiers: low risk indicates deviations that do not affect map representation or functional assessment; medium risk indicates deviations that are identifiable but still require verification; and high risk indicates deviations that affect spatial boundaries or functional use. P12 is classified as high risk, W7 as medium risk, and the eastern boundary of the site is classified as medium risk.

[0076] Finally, three types of output maps are generated based on the real-world entity database and the dynamic deviation field. The as-built topographic map extracts ground entities, site boundaries, and slope boundaries, and projects them onto the horizontal reference plane to form site boundaries, road boundaries, and elevation contour lines. The parking space map extracts parking space boundary entities, lane boundary entities, and column entities, and projects them onto the parking area reference plane to form parking space outlines, lane outlines, and column intrusion locations. The facade map extracts facade entities, inter-floor boundaries, and opening boundaries, and projects them onto the facade reference plane to form facade outlines, opening outlines, and deviation markers. Figure 5 Output schematic diagrams for as-built topographic maps, parking space maps, and facade maps, generated based on the final real-world entity library and dynamic deviation field. Figure 5 The left side shows the site boundaries and elevation contour lines in the as-built topographic map; the middle shows the parking space boundaries, column positions, and the result of a net width of 2.37 meters for P12 in the parking space map; and the right side shows the opening positions and the W7 deviation of 0.062 meters mark in the facade map. As can be seen from this embodiment, this invention does not rely on a building information model for overall fitting, but rather first constructs an evidence field based on real-world point cloud data, filters objects with insufficient evidence, completes directional supplementary surveys, and then generates a real-world entity database and conducts counter-evidence review. Therefore, it can provide data results with clear sources for key objects such as site boundaries, parking boundaries, and facade openings. Compared with processing methods that rely solely on initial scanning, in this embodiment, all key target evidence objects reach the evidence completeness threshold, and the average registration residuals of the parking area and facade area are reduced to within 0.021 meters. The final output map simultaneously possesses spatial representation results and deviation marking results, which can be directly used for as-built surveying and discrepancy verification.

[0077] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0078] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for as-built mapping of real-world locations using cloud-based methods, characterized in that, The method includes: Acquire 3D laser point cloud of the survey area, construct voxel evidence field and boundary evidence field, and determine the target evidence objects corresponding to the as-built topographic map, parking space map and facade map; The completeness of evidence is calculated based on the voxel evidence field, the boundary evidence field, and the target evidence object. Based on the completeness of evidence, objects with insufficient evidence and supplementary poses are determined. Supplementary scans are performed and the completeness of evidence is updated. Based on the target evidence objects whose evidence completeness meets the threshold, real-world entities are constructed, and rigid body registration is performed on the real-world entities based on local topological constraints and boundary and entity consistency constraints to obtain a real-world entity library. The building information model is subjected to a counter-evidence review based on the real-world entity database. A dynamic deviation field is generated based on the counter-evidence review results, and as-built topographic map, parking space map, and facade map are output according to the dynamic deviation field.

2. The method according to claim 1, characterized in that, The step of calculating the completeness of evidence based on the voxel evidence field, the boundary evidence field, and the target evidence object includes: The evidentiary completeness of the target evidence object is calculated based on its geometric continuity, boundary closure, multi-perspective support, and semantic decidability.

3. The method according to claim 2, characterized in that, The process of determining objects with insufficient evidence and supplementary poses based on the completeness of the evidence includes: Target evidence objects with a completeness of less than a preset threshold are identified as evidence-insufficient objects. The free space is determined based on the voxel evidence field, and candidate supplementary poses are generated based on the free space; The supplementary pose is determined based on the visibility quality of each candidate supplementary pose to the object with insufficient evidence, the importance of the object with insufficient evidence, and the pose cost.

4. The method according to claim 3, characterized in that, The process of performing supplementary scanning and updating the completeness of the evidence includes: Perform a supplementary scan based on the supplementary pose to obtain the supplementary point cloud; The occupancy, free, unknown, and conflict states of each voxel unit in the voxel evidence field are updated based on the supplementary point cloud. Update the boundary continuity state, boundary closure state, multi-view support state, semantically decidable state, and boundary conflict state in the boundary evidence field based on the supplementary point cloud. The completeness of the evidence of the target evidence object is recalculated based on the updated voxel evidence field and the boundary evidence field.

5. The method according to claim 1, characterized in that, The construction of real-world entities based on target evidence objects whose evidence completeness meets a threshold includes: Geometric segmentation is performed on target evidence objects whose evidence integrity meets the threshold to obtain patches and boundaries; The surface and the boundary are classified to obtain ground entity, facade entity, parking space boundary entity, opening boundary entity and column entity; The real-world entity is constructed based on the ground entity, the facade entity, the parking space boundary entity, the opening boundary entity, and the column entity.

6. The method according to claim 5, characterized in that, The local topological constraints are used to characterize the spatial relationships between real-world entities. The local topological constraints include at least one of the following: adjacent entity category distribution, entity spacing relationship, entity angle relationship, boundary closure relationship, and elevation layering relationship. The boundary and entity consistency constraint is used to characterize the correspondence between the boundary and the real-world entity. The boundary and entity consistency constraint includes at least one of the following: surface consistency constraint, boundary continuity constraint, nesting relationship constraint, spacing consistency constraint, and closure relationship constraint.

7. The method according to claim 6, characterized in that, The rigid body registration of the real-world entities based on local topological constraints and boundary-entity consistency constraints yields a real-world entity library, including: Establish registration constraint relationships between the real-world entities; The alignment relationship of the facets, the coincidence relationship of the boundaries, the consistency of the spacing, and the closure relationship of the boundaries are used as the constraints of the registration constraint relationship; The spatial poses of each real-world entity are solved based on the registration constraint relationship to obtain the real-world entity library.

8. The method according to claim 1, characterized in that, The verification of the building information model based on the real-world entity database includes: Match the real-world entities in the real-world entity library with the model objects in the building information model; Calculate the geometric consistency, boundary consistency, relational consistency, and functional consistency of the model object based on the matching results; Based on the geometric consistency, boundary consistency, relational consistency, and functional consistency, the uncertainty of the model object and the result of the rebuttal review of the model object are determined.

9. The method according to claim 8, characterized in that, The generation of the dynamic bias field based on the results of the counter-evidence review includes: Based on the counter-evidence review results of the model object and the uncertainty of the model object, the normal deviation, tangential deviation, boundary deviation and risk level corresponding to the model object are determined, and the dynamic deviation field is generated. The model object uncertainty is used to indicate the reliability of the model object's measurement, and the risk level is used to indicate the compliance risk level of the model object.

10. The method according to claim 9, characterized in that, The step of outputting the as-built topographic map, parking space map, and exterior elevation map based on the dynamic deviation field includes: Based on the real-scene entity library and the dynamic deviation field, extract the first real-scene entity corresponding to the as-built topographic map, the second real-scene entity corresponding to the parking space map, and the third real-scene entity corresponding to the facade map; The first real-world entity is projected onto a horizontal reference plane to generate the as-built topographic map; The second real-world entity is projected onto the parking area reference plane to generate the parking space map; The third real-world entity is projected onto the facade reference plane to generate the facade drawing.

Images