BIM real scene three-dimensional fusion monitoring method and system for smart construction site

Abstract

The application discloses a smart construction site BIM real scene three-dimensional fusion monitoring method and system, and particularly relates to the field of building engineering image data processing, and comprises the following steps: acquiring a BIM as-built model, construction procedure records and field real scene three-dimensional data of a target construction area; determining a current procedure number and generating a component index table according to a component number, a procedure time period and a collection pose; converting component geometric points in the BIM as-built model into process surfaces bound with the current procedure number according to the component number, the current procedure number and the collection pose in the component index table; and generating trace lines that should be left, trace lines that should be removed and trace lines that should not be present based on the process surfaces. The application converts BIM as-built components into process surfaces bound with the current procedure, and drives image enhancement, three-trace cutting clustering and difference source determination by using the trace lines that should be left, the trace lines that should be removed and the trace lines that should not be present, so as to solve the problem that direct comparison between the as-built BIM and the process state real scene leads to monitoring misjudgment.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology in construction engineering, and more specifically, to a method and system for BIM real-scene 3D fusion monitoring of smart construction sites. Background Technology

[0002] In smart construction site progress verification and quality inspection, BIM models are mostly used as design benchmarks. On-site, images and point cloud data are acquired through tower crane cameras, drones, mobile inspection terminals or laser scanning equipment. After coordinate registration, image enhancement, image clustering and 3D reconstruction, they are spatially compared with the walls, beams, slabs, columns, pipelines, openings and equipment foundations in the BIM to generate construction status and deviation results. However, during the construction phase of the main structure and electromechanical pre-embedded components of high-rise buildings, the same component will successively undergo rebar binding, formwork erection, pouring, formwork removal, pre-embedded positioning, and sealing. On-site, there are also formwork obstructions, safety net obstructions, material stacking, limited viewing angles, and insufficient local lighting. As a result, the collected real-world 3D images are not the BIM as-built components themselves, but rather the process forms such as rebar surfaces, formwork surfaces, pouring obstructed surfaces, exposed surfaces after formwork removal, sleeve openings, and sealed covering surfaces. If the current real-world 3D patches are directly compared with the BIM as-built components, it is easy to see that the formwork surface is written as the wall completion surface, the rebar surface is written as the component missing, the sleeve opening is merged into the background plane, the material stacking is written as the newly added component, and the area invisible from the viewing angle is written as the construction delay. The root cause is that the image processing results are not first classified into the construction process state source. The difference areas have not been distinguished as reasonable process states, obstructions, temporary occupancy, invisible data collection, construction deviation, and suspected changes, and are directly converted into monitoring anomalies. Therefore, the technical problem to be solved in this application is: how to classify the differences in the three-dimensional reality of the site according to the source of the construction process in BIM real scene 3D fusion monitoring, and distinguish the causes of the differences accordingly, so as to reduce the monitoring misjudgment caused by direct comparison between the as-built BIM and the process reality. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a smart construction site BIM real-scene 3D fusion monitoring method and system. By converting BIM as-built components into process surfaces bound to the current process, and using trace lines to be retained, trace lines to be removed, and trace lines to be prohibited, image enhancement, three-trace segmentation and clustering, and difference source determination are driven to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a smart construction site BIM real-scene 3D fusion monitoring method, comprising: S1. Obtain the BIM as-built model, construction process records and on-site 3D data of the target construction area, determine the current process number according to component number, process time period and acquisition pose and generate a component index table; S2. Based on the component number, current operation number and acquisition pose in the component index table, convert the component geometric points in the BIM as-built model into process surfaces bound to the current operation number, and generate the required trace lines, required removal trace lines and prohibited trace lines based on the process surfaces, and output the process surface rule table. S3. Based on the acquisition pose, project the process surface, the lines to be retained, the lines to be removed, and the lines to be prohibited onto the field image, generate the line center pixel sequence and the original edge position, perform image enhancement through line center accumulation, adjacent gray freezing and edge position recalculation, and classify them according to the original edge position and the enhanced edge position being in the same pixel interval, and output the line block table. S4. Generate a three-trace fragment table by combining the enhanced trace blocks in the trace block table with the actual scene 3D data according to the acquisition pose. Generate a source end judgment table based on the hit value of the trace line that should be retained, the residual value of the trace line that should be removed, and the occurrence value of the trace line that should not appear. Construct a three-trace conservation cut graph. Merge the actual scene fragments into process fragment clusters, occluded fragment clusters, temporary occupied fragment clusters, and suspected offset fragment clusters through three-trace segmentation and clustering. Output the source fragment cluster table. S5. Based on the collected pose, project the BIM as-built model to generate the BIM pixel range, compare the gap and increment with the projected pixel range in the source cluster table, and distinguish the differences according to the source cluster category, the residual value of the trace line and the appearance value of the prohibited trace line, and output the BIM real scene monitoring results.

[0005] In a preferred embodiment, S1 includes: S1-1. Read the geometric points of the component under the same component number in the BIM as-built model, sort the coordinates of the component geometric points on the X-axis, Y-axis and Z-axis in ascending order, and generate the component spatial frame with the first and last coordinates of each axis, and output the component spatial frame table. S1-2. Read the process number, process start time, process end time and process sequence number under the same component number in the construction process record. Perform half-open interval assignment calculation on the acquisition time of the on-site real scene 3D data with the process start time and process end time. Determine the process number of the process with the process start time not later than the acquisition time and the acquisition time earlier than the process end time as the current process number. If the acquisition time and the process end time are the same, determine the process number with the later process sequence number as the current process number. Associate the current process number with the component space frame and output the component process frame table. S1-3. Based on the acquired pose, convert the component space frame in the component process frame table to the coordinate system of the actual scene 3D data, count the number of real points in the component space frame, and generate an index row with component number, current process number, acquired pose and number of real points, and output the component index table.

[0006] In a preferred embodiment, S2 includes: S2-1. Read the component number, current operation number and acquisition pose from the component index table. Extract the component geometric points from the BIM as-built model according to the component number, and convert the component geometric points to the on-site real-world 3D data coordinate system according to the acquisition pose. Output the component acquisition point table. S2-2. Read the process sequence number of the current process in the construction process record, extract the component collection points with the same process sequence number value, and generate process surfaces by projecting them according to the collection pose. Output the process surface table.

[0007] In a preferred embodiment, S2 further includes: S2-3. Record the process sequence number of the current process as the current sequence number. Perform coordinate difference between the process range whose process sequence number value is less than the current sequence number and the process surface. Determine the line segments that are still on the boundary of the process surface after the difference as the lines to be left with traces. Determine the line segments that are off the boundary of the process surface after the difference as the lines to be removed with traces. Output the trace removal line table. S2-4. Perform coordinate intersection calculations on the process surface for the process range whose process sequence value is greater than the current sequence value. Determine the line segment obtained from the intersection as the no-mark line, and generate a process surface rule table using the component number, the current process number, the process surface, the mark line to be retained, the mark line to be removed, and the no-mark line.

[0008] In a preferred embodiment, S3 includes: S3-1. Read the process surface, trace lines to be retained, trace lines to be removed and trace lines to be prohibited from appearing in the process surface rule table. Based on the acquisition pose, project the process surface and the three types of trace lines onto the field image. Generate the line center pixel sequence along the pixel order of the trace line projection line. Generate the trace line projection table with pixel number, trace line type and line direction number. S3-2. Read the line center pixel sequence from the trace projection table, extract the left and right neighbor pixels of each line center pixel along the perpendicular direction of the trace projection line, calculate the absolute value of the gray level difference between the line center pixel and the left neighbor pixel as the left gray level difference, calculate the absolute value of the gray level difference between the line center pixel and the right neighbor pixel as the right gray level difference, when the left gray level difference is not less than the right gray level difference, determine the pixel interval between the line center pixel and the left neighbor pixel as the original edge position, when the left gray level difference is less than the right gray level difference, determine the pixel interval between the line center pixel and the right neighbor pixel as the original edge position, and output the trace edge position table.

[0009] In a preferred embodiment, S3 further includes: S3-3. Read the line direction number, pixel number and original edge position from the edge position table, execute the edge position locking enhancement algorithm consisting of line center accumulation, neighbor gray freezing and edge position recalculation, accumulate the gray values ​​of each line center pixel under the same line direction number along the pixel number to form the line center enhancement value, and keep the gray values ​​of the left neighbor pixel and the right neighbor pixel from participating in the accumulation, and output the enhanced edge block. S3-4. Read the enhanced etcher blocks, calculate the absolute difference between the enhanced value of the center pixel and the gray value of the left neighboring pixel as the enhanced left gray difference, and calculate the absolute difference between the enhanced value of the center pixel and the gray value of the right neighboring pixel as the enhanced right gray difference. When the enhanced left gray difference is not less than the enhanced right gray difference, the pixel interval between the center pixel and the left neighboring pixel is determined as the enhanced edge position. When the enhanced left gray difference is less than the enhanced right gray difference, the pixel interval between the center pixel and the right neighboring pixel is determined as the enhanced edge position. Then, the enhanced etcher blocks with enhanced edge positions and original edge positions in the same pixel interval are classified according to the etcher type, and the etcher block table is output.

[0010] In a preferred embodiment, S4 includes: S4-1. Read the enhanced trace blocks and the 3D data of the actual scene from the trace block table. Based on the acquisition pose, project the actual points onto the scene image. Merge the actual points whose projections fall into the same enhanced trace block and whose depth order is adjacent into traced actual scene fragments. Merge the actual points whose projections do not fall into the enhanced trace block and whose projection pixel ranges are connected into traceless actual scene fragments. Record the number of actual points in the actual scene fragments that belong to trace lines to be retained, trace lines to be removed, and trace lines to be prohibited as the trace line hit value, trace line residual value, and trace line appearance value, respectively. Output the three-trace fragment table. S4-2. Read the three-trace fragment table and generate a source determination table based on the three types of trace values. The process source is established when the prohibited trace value is zero, the required trace hit value is positive, and the required trace residue value is zero; otherwise, the process source is not established. The occlusion source is established when the prohibited trace value is zero, the required trace hit value is zero, and the required trace residue value is positive; otherwise, the occlusion source is not established. The temporary occupation source is established when the prohibited trace value is zero and both the required trace hit value and the required trace residue value are positive, or when all three types of trace values ​​are zero; otherwise, the temporary occupation source is not established. The suspected offset source is established when the prohibited trace value is positive; otherwise, the suspected offset source is not established.

[0011] In a preferred embodiment, S4 further includes: S4-3. Construct a three-trace cut graph based on the three-trace fragment table and the source end determination table. Use real scene fragments as cut graph nodes and process source end, occlusion source end, temporary occupation source end, and offset suspected source end as label ends. When the source end is established, the edge cost of the process source end is the sum of the residual value of the trace line that should be removed and the occurrence value of the trace line that should appear. The edge cost of the occlusion source end is the sum of the hit value of the trace line that should be retained and the occurrence value of the trace line that should appear. The edge cost of the temporary occupation source end is the sum of the occurrence value of the trace line that should appear plus the absolute number of the difference between the hit value of the trace line that should be retained and the residual value of the trace line that should be removed. The edge cost of the offset suspected source end is the sum of the hit value of the trace line that should be retained and the residual value of the trace line that should be removed. When the source end is not established, the edge cost is the sum of the number of real scene points, the hit value of the trace line that should be retained, the residual value of the trace line that should be removed, and the occurrence value of the trace line that should be removed. S4-4. Generate adjacent edges between real scene fragments whose projected pixel ranges are connected. Subtract the absolute number of the difference between the same type of trace values ​​from the sum of the trace values ​​of two real scene fragments, and divide by two to obtain the common trace value. Use the sum of the three types of common trace values ​​as the adjacent edge cutting cost. Perform three-trace segmentation clustering in the order of process source end, occlusion source end, temporary occupation source end, and offset suspected source end. Use the sum of the label end edge cost and the cut-off cost of the adjacent edge to be used as the total cost. If the total cost after expansion is less than the total cost before expansion, use the expanded label. Otherwise, retain the label before expansion. If the label end names are the same in both rounds, stop. Otherwise, enter the next round of expansion. Merge real scene fragments with the same label end name and whose projected pixel ranges are connected into process fragment clusters, occlusion fragment clusters, temporary occupation fragment clusters, and offset suspected fragment clusters. Output the source fragment cluster table.

[0012] In a preferred embodiment, S5 includes: S5-1. Read the projected pixel range and BIM as-built model from the source patch table, generate the BIM pixel range based on the collected pose, determine the area in the BIM pixel range that is not occupied by the projected pixel range as the gap difference area, determine the area in the projected pixel range that does not fall into the BIM pixel range as the incremental difference area, and output the difference area table. S5-2. Read the difference area table and the source patch table, associate the notch difference area with the source patch that is connected to the projected pixel range, associate the incremental difference area with the source patch that overlaps with the projected pixel range, and extract the source patch category, the residual value of the trace line that should be removed and the appearance value of the trace line that should not appear, and output the difference judgment table. S5-3. Read the difference judgment table. If the process area cluster has a zero value for the forbidden trace line, it is judged as a reasonable process state difference. If it does not meet the requirements, it is judged as an occlusion difference, a temporary occupation difference, a collection invisible difference, and a construction deviation difference in the following order: occlusion area cluster, temporary occupation area cluster, unassociated source area cluster, and suspected offset area cluster. If it still does not meet the requirements, it is judged as a suspected change difference. Output the BIM real scene monitoring results.

[0013] In a preferred embodiment, the smart construction site BIM real-scene 3D fusion monitoring system includes: The index generation module is used to acquire the BIM as-built model, construction process records and on-site 3D data of the target construction area, determine the current process number according to the component number, process time period and acquisition pose and generate a component index table. The process conversion module converts the geometric points of components in the BIM as-built model into process surfaces bound to the current process number based on the component number, the current process number, and the acquisition pose in the component index table. It also generates lines to be left, lines to be removed, and lines to be prohibited from being left based on the process surfaces and outputs a process surface rule table. The trace enhancement module projects the process surface, traces to be retained, traces to be removed, and traces to be prohibited onto the field image based on the acquired pose, generates a line center pixel sequence and original edge position, performs image enhancement through line center accumulation, adjacent gray freezing, and edge position recalculation, and classifies the original edge position and enhanced edge position into the same pixel interval, and outputs a trace block table. The fragment segmentation module is used to generate a three-trace fragment table by combining the enhanced trace blocks in the trace block table with the actual scene 3D data according to the acquisition pose. It generates a source end judgment table based on the hit value of the trace line that should be retained, the residual value of the trace line that should be removed, and the occurrence value of the trace line that should not appear. It also constructs a three-trace conservation cut map and merges the actual scene fragments into process fragment clusters, occluded fragment clusters, temporary occupied fragment clusters, and suspected offset fragment clusters through three-trace segmentation clustering, and outputs a source fragment cluster table. The difference determination module projects the BIM as-built model based on the collected pose to generate the BIM pixel range, compares the gaps and increments with the projected pixel range in the source cluster table, and distinguishes the differences according to the source cluster category, the residual value of the trace line, and the appearance value of the prohibited trace line, and outputs the BIM real-scene monitoring results.

[0014] The technical effects and advantages of this invention are as follows: 1. By first connecting the construction stages such as reinforcement, formwork, pouring, and pre-embedding through the process surface, and then explaining the gaps and incremental differences by the source cluster, the formwork surface, reinforcement surface, and material stacking surface are no longer directly treated as as-built BIM anomalies, which relatively reduces the probability of the process-state reality being misjudged as construction deviation. 2. By placing the component number, current process number, acquisition pose, and number of real points in the same index row through the component index table, the subsequent process surfaces are generated only for the current process, reducing the comparison deviation caused by incorrect component shapes across processes; 3. By distinguishing between "should be retained," "should disappear," and "should not appear" on-site traces through "should be retained lines," "should be removed lines," and "should not appear lines," image enhancement only processes construction trace areas, reducing the situation where background textures and material edges are treated as component boundaries during overall image enhancement; 4. By accumulating the grayscale of the line core, freezing adjacent grays to retain references on both sides, and recalculating the edge position to remove displaced lines, the enhanced line blocks retain their edge positions after brightness enhancement, thus relatively reducing the risk of introducing false boundaries during the enhancement process. 5. The number of real spots is converted into the hit value of the trace line that should be left, the residual value of the trace line that should be removed, and the appearance value of the trace line that should not appear through the three-trace fragment table. Then, the three-trace segmentation and clustering process, occlusion, temporary occupation, and suspected offset patches are formed, so that the image clustering results directly carry the meaning of the construction source. 6. By comparing the gaps and increments between the BIM pixel range and the projected pixel range of the source cluster, and judging the differences according to the cluster type and trace value, the output of occlusion, temporary occupation, invisible data collection and construction deviation is separated, reducing false alarms caused by a single difference map. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method steps of the present invention.

[0016] Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Refer to the instruction manual appendix Figure 1-2 The intelligent construction site BIM real-scene 3D fusion monitoring method of the present invention includes: S1. Obtain the BIM as-built model, construction process records and on-site 3D data of the target construction area, determine the current process number according to component number, process time period and acquisition pose and generate a component index table; This implementation method is used to form a component index table, so that the component geometric range in the BIM as-built model, the process time range in the construction process record, and the acquisition status in the on-site 3D data are entered into the same index row; the on-site 3D data includes the real-world point, on-site image, acquisition time, and acquisition pose, and the acquisition pose includes the rotation matrix, translation vector, and imaging parameters; the component geometric points in the BIM as-built model are composed of boundary vertices and face boundary sampling points under the same component number; the implementation process includes the following steps: In S1-1, all geometric points of components belonging to the same component number within the BIM as-built model are first retrieved using the component number as the search key. Each component geometric point contains X-axis coordinates, Y-axis coordinates, and Z-axis coordinates. Then, the X-axis coordinate column, Y-axis coordinate column, and Z-axis coordinate column are sorted in ascending order. The first and last coordinates of the X-axis coordinate column are taken as the lower and upper bounds of the X-axis, the first and last coordinates of the Y-axis coordinate column are taken as the lower and upper bounds of the Y-axis, and the first and last coordinates of the Z-axis coordinate column are taken as the lower and upper bounds of the Z-axis. Then, a six-sided enclosed component space frame is generated using the three lower and upper bounds. The component number, X-axis lower bound, X-axis upper bound, Y-axis lower bound, Y-axis upper bound, Z-axis lower bound, and Z-axis upper bound are combined to form a component space frame row, and the component space frame table is output. In S1-2, the process number, process start time, process end time, and process sequence number under the same component number in the construction process record are read, and the acquisition time of the on-site real-scene 3D data is also read. For each process record under the same component number, the process start time is not later than the acquisition time and the acquisition time is earlier than the process end time as the half-open interval assignment condition, and the process number that meets the half-open interval assignment condition is taken as the current process number. When the acquisition time is the same as the end time of a certain process, and the same acquisition time falls into the start boundary of the next process, the process number with the larger value of the two process sequence numbers is taken as the current process number. When the acquisition time does not fall into the start or end time of any process under the same component number, no component process frame row is generated for that component number. Finally, the component number that generates the current process number is associated with the component spatial frame to form a component process frame table containing the component number, the current process number, the process sequence number, and the component spatial frame. In S1-3, the component spatial frame in the component process frame table and the acquisition pose in the on-site 3D data are read. The rotation matrix and translation vector in the acquisition pose are used to perform coordinate transformation on the eight corner points of the component spatial frame to obtain the acquisition frame in the coordinate system of the on-site 3D data. For each real point in the on-site 3D data, the coordinates of the real point are read. The coordinates of the real point are compared with the coordinate range of the acquisition frame on the X-axis, Y-axis and Z-axis respectively. Real points whose coordinates on all three axes fall within the range of the acquisition frame are counted as real points. Real points whose coordinates on any axis do not fall within the range of the acquisition frame are not counted as real points. Then, an index row is generated with component number, current process number, acquisition pose and real point count, and the component index table is output. Through the above processing, the component index table compresses the component spatial range, process time attribution, and on-site acquisition status into the same index row. This allows the component number, current process number, and acquisition pose to be directly called when generating subsequent process surfaces. The number of real points reflects the on-site occupancy within the component spatial frame, thereby reducing retrieval breakpoints between BIM as-built components and on-site acquisition data. In practical applications: For example, the shear wall component number of a certain floor is Q-03. The construction process record has the process start time, process end time, and process sequence number for rebar binding, formwork erection, and pouring, respectively. After the fixed camera generates the on-site real-scene 3D data at 10:30, the system first generates the component spatial frame from the component geometry points of Q-03, then obtains the current process number based on the half-open interval that falls within at 10:30, and then converts the component spatial frame to the on-site real-scene 3D data coordinate system according to the acquisition pose, counts the number of real points within the frame, and finally generates the component index row of Q-03 in the acquisition state at 10:30.

[0019] S2. Based on the component number, current operation number and acquisition pose in the component index table, convert the component geometric points in the BIM as-built model into process surfaces bound to the current operation number, and generate the required trace lines, required removal trace lines and prohibited trace lines based on the process surfaces, and output the process surface rule table. This implementation method converts the component number, current process number, and acquisition pose from the component index table into a process surface rule table. This prevents the BIM as-built model from being directly used for on-site comparison. Instead, it first generates process surfaces based on the current process number, and then generates lines for leaving traces, retractions, and prohibitions based on the process sequence. The construction process record sets the process range, which includes the component number, process number, process sequence, range boundary points, and range line segments. The range line segments are formed by connecting the range boundary points according to the arrangement order in the process record. The process surface is a three-dimensional surface in the on-site 3D data coordinate system, which is subsequently projected onto the on-site image. This implementation process includes the following steps: In S2-1, the component number, current process number, and acquisition pose are read from the component index table, and the component geometric points under the same component number are extracted from the BIM as-built model using the component number as the search key. The component geometric points include the boundary vertices and surface boundary sampling points of the BIM component, and each component geometric point has three-dimensional coordinates. Then, the rotation matrix and translation vector in the acquisition pose are used to perform coordinate transformation on the component geometric points, transforming the component geometric points in the BIM as-built model coordinate system to the actual scene three-dimensional data coordinate system, forming a component acquisition point table containing the component number, current process number, acquisition pose, and the transformed component geometric points. In S2-2, the sequence number of the current process is read from the construction process record, and the process range is read using the component number and the current process number. When the sequence number of the process range is the same as the sequence number of the current process number, and the component acquisition point falls within the area enclosed by the boundary points of the process range, the component acquisition point is taken as the process point of the current process. Then, the process point is projected onto a three-dimensional surface area at the same level as the three-dimensional data coordinate system of the actual scene according to the acquisition pose. The process surface is generated according to the outer edge connection order of the process points, and the process surface table containing the component number, the current process number, the process point, and the process surface is output. In S2-3, the process sequence number of the current process is recorded as the current sequence number. The process range with a process sequence number value less than the current sequence number under the same component number is read, and the coordinate difference is performed between the range line segment in the process range and the process surface. When performing coordinate difference, it is first determined whether the two endpoints of the range line segment are located on the boundary of the process surface. The line segment with both endpoints located on the boundary of the process surface is retained as the line to be left, and the line segment with either endpoint detached from the boundary of the process surface is taken as the line to be removed. If there is no process range with a process sequence number value less than the current sequence number, the line to be left and the line to be removed are empty, and the table of left and removed lines containing component number, current process number, line to be left and the line to be removed is output. In S2-4, the process range with process sequence values ​​greater than the current sequence value under the same component number is read, and the coordinate intersection calculation is performed between the range line segments in the process range and the process surface. During coordinate intersection, the part where the endpoints of the range line segments coincide or the line segments coincide with the boundary of the process surface is taken as the intersection line segment, and the intersection line segment is determined as the no-marking line. If there is no process range with process sequence values ​​greater than the current sequence value, the no-marking line is empty. Then, the process surface rule row is composed of component number, current process number, process surface, line to be marked, line to be removed, and no-marking line, and the process surface rule table is output. Through the above processing, the process surface rule table fixes the traces that should be retained, should be removed, and should not appear prematurely under this process to the process surface, so that subsequent image enhancement no longer focuses on the entire site image, but performs edge processing on the projection position of the three types of traces; at the same time, empty traces participate in subsequent calculations as empty sets, take zero values ​​in S4, and do not interrupt the generation of subsequent three-trace fragments and source fragments; in practical applications: for example, shear wall component Q-03 is in the formwork process under this process number. The system first converts the geometric points of component Q-03 to the three-dimensional data coordinate system of the actual site scene, and then reads the process range of the formwork process to generate the template process surface; the line segments in the range of the rebar binding process that still fall on the boundary of the template process surface form the traces that should be retained, the line segments that are off the boundary of the template process surface form the traces that should be removed, and the intersection line segments of the demolding edge line that should appear after pouring and the template process surface form the traces that should not appear, thereby generating the process surface rule table under the formwork process of Q-03.

[0020] S3. Based on the acquisition pose, project the process surface, the lines to be retained, the lines to be removed, and the lines to be prohibited onto the field image, generate the line center pixel sequence and the original edge position, perform image enhancement through line center accumulation, adjacent gray freezing and edge position recalculation, and classify them according to the original edge position and the enhanced edge position being in the same pixel interval, and output the line block table. This implementation method converts the three types of trace lines in the process surface rule table into enhanced trace line blocks in the field image, so that image enhancement only applies to the slender regions where trace lines that should be retained, trace lines that should be removed, and trace lines that should not be displayed are located. It also limits the enhancement process from changing the trace line edge position by verifying the original edge position and the enhanced edge position at the same interval. The field image is first converted into a luminance channel, and the pixel grayscale values ​​in the luminance channel are used for subsequent calculations of left gray difference, right gray difference, line center enhancement value, enhanced left gray difference, and enhanced right gray difference. Empty trace lines do not generate trace line projection lines and are not included in the trace line projection table. This implementation process includes the following steps: In S3-1, the process surface, trace lines to be retained, trace lines to be removed, and trace lines to be prohibited from appearing are read from the process surface rule table. Based on the rotation matrix, translation vector, and imaging parameters in the acquisition pose, the process surface is converted into the process surface projection area in the field image, and the two endpoints of the three types of trace lines are converted into pixel endpoints in the field image. For each trace line projection line, the direction with the larger absolute value of the row coordinate difference and the column coordinate difference between the pixel endpoints is taken as the progression axis. A line center pixel sequence is generated pixel by pixel along the progression axis. The line center pixel sequence is assigned a pixel number in the order from the first endpoint to the last endpoint. The same trace line projection line is assigned a line direction number, and the trace lines to be retained, trace lines to be removed, or trace lines to be prohibited from appearing are taken as the trace line type, and the pixel number and line direction number are used to form the trace line projection table. In S3-2, the sequence of center pixels in the trace projection table is read. The direction of pixel number increment is taken as the line direction. The left neighbor pixel is taken in the direction of counterclockwise rotation of 90 degrees, and the right neighbor pixel is taken in the direction of clockwise rotation of 90 degrees. For each center pixel, the gray value of the center pixel, the gray value of the left neighbor pixel, and the gray value of the right neighbor pixel are read. The absolute difference between the gray value of the center pixel and the gray value of the left neighbor pixel is calculated as the left gray difference, and the absolute difference between the gray value of the center pixel and the gray value of the right neighbor pixel is calculated as the right gray difference. When the left gray difference is not less than the right gray difference, the pixel interval between the center pixel and the left neighbor pixel is determined as the original edge position. When the left gray difference is less than the right gray difference, the pixel interval between the center pixel and the right neighbor pixel is determined as the original edge position. The trace edge position table is composed of line direction number, pixel number, left gray difference, right gray difference, and original edge position. In S3-3, the line direction number, pixel number, and original edge position are read from the edge position table, and the gray value of each center pixel under the same line direction number is read. The center pixel is accumulated in the direction of increasing pixel number. The center pixel enhancement value of the first pixel number is taken as the gray value of the first center pixel. Starting from the second pixel number, the gray value of the current center pixel is added to the center pixel enhancement value of the previous pixel number to obtain the center pixel enhancement value of the current pixel number. The gray values ​​of the left and right neighboring pixels are kept in their original values ​​during the accumulation process and do not participate in the calculation of the center pixel enhancement value. Thus, an enhanced edge block is formed, which includes the line direction number, pixel number, center pixel enhancement value, gray value of the left neighboring pixel, gray value of the right neighboring pixel, and original edge position. In S3-4, the enhancement value of the center pixel, the gray value of the left neighbor pixel, the gray value of the right neighbor pixel, and the original edge position in the enhancement trace block are read. The absolute difference between the enhancement value of the center pixel and the gray value of the left neighbor pixel is calculated as the enhancement left gray difference, and the absolute difference between the enhancement value of the center pixel and the gray value of the right neighbor pixel is calculated as the enhancement right gray difference. When the enhancement left gray difference is not less than the enhancement right gray difference, the pixel interval between the center pixel and the left neighbor pixel is determined as the enhancement edge position. When the enhancement left gray difference is less than the enhancement right gray difference, the pixel interval between the center pixel and the right neighbor pixel is determined as the enhancement edge position. When the enhancement edge positions of each pixel number in the same enhancement trace block are all in the same pixel interval as the original edge position under the same pixel number, the enhancement trace block is classified into the trace block table according to the trace type. When there is an enhancement edge position and the original edge position of any pixel number that are not in the same pixel interval, the enhancement trace block is recorded in the displacement trace record and is not entered into the trace block table. Through the above processing, image enhancement is restricted to the core pixel sequence. The left and right neighboring pixels are used as edge references and do not participate in the accumulation, thus enhancing the brightness of the enhanced trace lines. At the same time, the edge position movement is restricted by the result of the enhanced edge position recalculation. The trace line block table therefore retains the enhanced trace line blocks, trace line types, original edge positions, and enhanced edge positions, providing pixel-level input for the subsequent generation of the three-trace fragment table. In practical applications: for example, in the wall process surface during the formwork stage, there are template seams as trace lines to be left, exposed steel bar edges as trace lines to be removed, and demolding hole edges as trace lines to be prohibited. After the system projects the three types of trace lines onto the fixed camera image, it generates a core pixel sequence along the template seam direction and takes the left and right neighboring pixels on both sides of each core pixel. The core pixels are accumulated along the seam direction to form an enhanced trace line block. If the enhanced edge position is still within the original pixel interval, the template seam trace line block enters the trace line block table. If the edge position changes, it is not used as input for the subsequent three-trace segmentation and clustering.

[0021] S4. Generate a three-trace fragment table by combining the enhanced trace blocks in the trace block table with the actual scene 3D data according to the acquisition pose. Generate a source end judgment table based on the hit value of the trace line that should be retained, the residual value of the trace line that should be removed, and the occurrence value of the trace line that should not appear. Construct a three-trace conservation cut graph. Merge the actual scene fragments into process fragment clusters, occluded fragment clusters, temporary occupied fragment clusters, and suspected offset fragment clusters through three-trace segmentation and clustering. Output the source fragment cluster table. This implementation method binds enhanced trace blocks in the trace block table to real points in the on-site 3D real-world data, and uses traces that should be retained, traces that should be removed, and traces that should not be present to form a three-dimensional segmentation cluster. This allows the on-site 3D real-world data to be grouped according to the construction process source, rather than by ordinary appearance similarity, into process clusters, occupied clusters, temporary occupied clusters, and suspected offset clusters. The trace block table includes enhanced trace blocks, trace types, original edge positions, enhanced edge positions, and projected pixel ranges. The on-site 3D real-world data includes real points, on-site images, acquisition poses, and acquisition times. The trace values ​​for all three types are obtained from the number of real points and are non-negative integers. This implementation process includes the following steps: In S4-1, the enhanced trace blocks in the trace block table and the real points in the 3D real-world data are read. Each real point is projected onto the real-world image based on its acquisition pose, obtaining the projected pixel coordinates and depth value of each real point. When a real point's projection falls within the projected pixel range of an enhanced trace block, it is first grouped by the enhanced trace block number, then by the projected pixel coordinates, and then sorted in ascending order of depth value under the same projected pixel coordinates. Real points with adjacent depths form depth columns, and depth columns of adjacent projected pixel coordinates are merged into traced real-world fragments based on their shared edge connection. When the projection of a real point does not fall into any enhancement line block, it is merged into a seamless real-scene fragment according to the common edge connection relationship of the projection pixel range; when the same real point falls into multiple enhancement line blocks, a set of point numbers is formed according to the line type. The same point number under the same line type is counted only once. The number of real points in the real-scene fragment belonging to the lines to be retained, the lines to be removed, and the lines to be prohibited are recorded as the hit value of the lines to be retained, the residual value of the lines to be removed, and the appearance value of the lines to be prohibited, respectively. The output includes a three-line fragment table including fragment number, fragment type, projection pixel range, number of real points and three types of line values. In S4-2, the hit value of the trace line to be retained, the residual value of the trace line to be removed, and the occurrence value of the trace line to be prohibited from appearing are read from the three-trace fragment table, and a source end determination table is generated according to the three types of trace line values; when the occurrence value of the trace line to be prohibited from appearing is zero, the hit value of the trace line to be retained is positive, and the residual value of the trace line to be removed is zero, the process source end is established; otherwise, the process source end is not established; when the occurrence value of the trace line to be prohibited from appearing is zero, the hit value of the trace line to be retained is zero, and the residual value of the trace line to be removed is positive, the occlusion source end is established; otherwise, the occlusion source end is not established; the occurrence value of the trace line to be prohibited from appearing... When the value is zero and both the hit value of the required trace line and the residual value of the required trace line are positive, the temporary occupation of the source end is established; when all three types of trace line values ​​are zero, the temporary occupation of the source end is established; except for the above two cases, the temporary occupation of the source end is not established; when the value of the prohibited trace line is positive, the suspected deviation of the source end is established, otherwise the suspected deviation of the source end is not established; the source end determination table includes the fragment number, the process source end establishment bit, the occlusion source end establishment bit, the temporary occupation of the source end establishment bit, and the deviation suspected source end establishment bit; In S4-3, a three-trace cut graph is constructed based on the three-trace fragment table and the source end determination table. The three-trace cut graph uses real scene fragments as cut graph nodes and process source end, occlusion source end, temporary occupation source end, and offset suspected source end as label ends. For each cut graph node, if the process source end is valid, the edge cost of the process source end is the sum of the residual value of the retrace line and the occurrence value of the forbidden retrace line. If the process source end is invalid, the edge cost of the process source end is the sum of the number of real spots, the hit value of the retrace line, the residual value of the retrace line, and the occurrence value of the forbidden retrace line. If the occlusion source end is valid, the edge cost of the occlusion source end is the sum of the hit value of the retrace line and the occurrence value of the forbidden retrace line. If the occlusion source end is invalid, the edge cost of the occlusion source end is the sum of the number of real spots and the hit value of the retrace line. The cost of temporary occupation of the source end is calculated as follows: if the source end is temporarily occupied, the cost is calculated as the sum of the value of the forbidden trace and the absolute difference between the hit value of the trace to be retained and the value of the trace to be removed; if the source end is temporarily occupied, the cost is calculated as the sum of the number of real spots, the hit value of the trace to be retained, the residual value of the trace to be removed, and the value of the trace to be removed. If the source end is suspected of being off-center, the cost is calculated as the sum of the hit value of the trace to be retained and the residual value of the trace to be removed; if the source end is suspected of being off-center, the cost is calculated as the sum of the hit value of the trace to be retained and the residual value of the trace to be removed; if the source end is suspected of being off-center, the cost is calculated as the sum of the number of real spots, the hit value of the trace to be retained, the residual value of the trace to be removed, and the value of the trace to be removed. This forms the cost of each real scene element to the four tag ends. In S4-4, adjacent edges are generated between real-scene fragments that share a common edge within the projected pixel range; corner contacts do not generate adjacent edges. For two adjacent real-scene fragments, the sum of their similar edge values ​​is subtracted from the absolute difference of their similar edge values, and then divided by two to obtain the common edge value. The common edge value represents the number of similar edges shared by the two real-scene fragments. The sum of the three types of common edge values ​​is used as the adjacent edge cutting cost. The initial labels for the three-edge cutting clustering are determined in ascending order of the edge costs of the four label ends. When the edge costs are the same, the label ends that appear first are selected in the order of process source end, occlusion source end, temporary occupation source end, and offset suspected source end. Subsequently, the labels are selected according to the same process... The expansion calculation is performed sequentially, with the sum of the label edge cost and the cut cost of the adjacent cut edge as the total cost. If the total cost after expansion is less than the total cost before expansion, the expanded label is used; otherwise, the label before expansion is retained. The expansion stops when the label edge names of the two rounds are identical; otherwise, the expansion proceeds to the next round. After stopping, real scene fragments with the same label edge name and whose projected pixel ranges are connected by the same edge are merged to obtain process fragment clusters, occluded fragment clusters, temporarily occupied fragment clusters, and suspected offset fragment clusters, and the source fragment cluster table is output. The source fragment cluster table includes fragment cluster number, source fragment cluster category, projected pixel range, hit value of the line to be left, residual value of the line to be removed, and occurrence value of the line to be prohibited. Through the above processing, the three-line segmentation clustering transforms the construction process information in the enhanced line block into source labels for real-world elements. The element affixing is then restricted by the source end establishment location, edge cost, adjacent edge cutting cost, and expansion stopping rules, avoiding grouping of the actual 3D data based solely on color, texture, or spatial distance. The projected pixel range and three types of line values ​​in the source element cluster table directly enter subsequent gap and incremental comparisons. In practical applications: for example, during formwork construction, formwork seams form lines that should be retained, exposed rebar edges form lines that should be removed, and demolding hole edges form lines that should not be present. When a real-world element contains only lines that should be retained but no lines that should be removed or lines that should not be present, the process source end is established and obtains a small edge cost; this element is assigned to a process element cluster in the three-line segmentation clustering. When a real-world element contains lines that should not be present, a suspected offset source end is established, and after segmentation, it forms a suspected offset element cluster with adjacent elements, for subsequent BIM real-world monitoring results to differentiate between them.

[0022] S5. Based on the collected pose, project the BIM as-built model to generate the BIM pixel range, compare the gap and increment with the projected pixel range in the source cluster table, and distinguish the differences according to the source cluster category, the residual value of the trace line and the appearance value of the prohibited trace line, and output the BIM real scene monitoring results. This implementation method compares the pixel range of the source cluster table with the BIM as-built model, and converts the difference areas into BIM real-scene monitoring results according to the source cluster category and three types of trace values. The source cluster table includes cluster number, source cluster category, projected pixel range, hit value of traces that should be retained, residual value of traces that should be removed, and occurrence value of traces that should not appear. The BIM pixel range is obtained by projecting the pose of the BIM as-built model components associated with the component number in the component index table. BIM components outside the target construction area are not included in this step. The implementation process includes the following steps: In S5-1, the projected pixel range in the source cluster table is read, and the component geometric points associated with the component number in the component index table in the BIM as-built model are read. Based on the rotation matrix, translation vector, and imaging parameters in the acquired pose, the component geometric points are projected onto the site image to obtain the BIM pixel range. The union of the projected pixel ranges of each source cluster in the source cluster table is taken to obtain the cluster pixel range. The pixel set after subtracting the cluster pixel range from the BIM pixel range is determined as the gap difference area, and the pixel set after subtracting the BIM pixel range from the cluster pixel range is determined as the incremental difference area. Each gap difference area and incremental difference area is configured with a difference area number, difference type, and pixel range, and a difference area table is output. In S5-2, the difference area table and the source patch table are read. The difference area with the notch is associated with the source patch that shares an edge with the projected pixel range; corner contact is not considered an edge. The incremental difference area is associated with the source patch that has overlapping pixels in the projected pixel range. When a difference area with a notch shares an edge with multiple source patches, the first source patch is selected in descending order of the number of pixels sharing an edge. When multiple source patches have the same number of pixels sharing an edge, the first source patch is selected in the order of process patch, occluded patch, temporarily occupied patch, and suspected offset patch. When an incremental difference area overlaps with multiple source patches, the first source patch is selected in descending order of the number of overlapping pixels. When multiple source patches have the same number of overlapping pixels, the first source patch is selected in the order of process patch, occluded patch, temporarily occupied patch, and suspected offset patch. Then, the source patch category, residual value of the trace line to be removed, and occurrence value of the prohibited trace line are extracted from the selected source patch. The difference judgment table is output, and difference areas without associated source patches are retained as empty association markers. In S5-3, the difference judgment table is read, and difference type judgment is performed for each difference area. When the difference area is associated with a process area and the forbidden trace line has a value of zero, the difference area is judged as a reasonable process difference; otherwise, it enters the occlusion judgment. When the difference area is associated with an occluded area and the forbidden trace line has a value of zero, the difference area is judged as an occlusion difference; otherwise, it enters the temporary occupation judgment. When the difference area is associated with a temporary occupied area and the residual value of the trace line and the forbidden trace line have a value of zero, the difference area is judged as a temporary occupation difference; otherwise, it enters the collection invisible judgment. When the difference area has an empty associated marker, the difference area is judged as a collection invisible difference; otherwise, it enters the construction deviation judgment. When the difference area is associated with a suspected deviation area and the forbidden trace line has a positive value, the difference area is judged as a construction deviation difference; otherwise, the difference area is judged as a suspected change difference. Suspected change differences are only used as input for design version review or on-site review and are not directly used as design change conclusions. The final output is the BIM real-scene monitoring result. Through the above processing, the BIM pixel range and the projected pixel range in the source cluster table first form gap difference areas and incremental difference areas. Then, the source cluster category, residual value of the trace line, and occurrence value of the prohibited trace line complete the differentiation of the difference causes. This makes the monitoring results no longer given by the direct difference between the BIM as-built model and the actual site clusters, but determined by the relationship between the source clusters. In practical applications: for example, when a gap difference area is formed in the wall of a certain floor during the formwork stage, if the source cluster connected to the gap difference area is a process cluster and the occurrence value of the prohibited trace line is zero, then a reasonable process difference is output; if the connected source cluster is an occlusion cluster and the occurrence value of the prohibited trace line is zero, then an occlusion difference is output; if the difference area is not associated with any source cluster, then an invisible difference is output; if the source cluster is a suspected offset cluster and the occurrence value of the prohibited trace line is positive, then a construction offset difference is output. This forms a BIM real-world monitoring result that includes the difference area number, difference type, source cluster category, and three types of trace line values.

[0023] It should be further explained that during the on-site data acquisition phase, the target construction area is divided into monitoring units according to floor number, grid number, and component number. The monitoring unit records the acquisition terminal number, acquisition time, acquisition pose, and data type. The acquisition terminals include tower crane cameras, floor-fixed cameras, drones, mobile inspection terminals, and laser scanning equipment. On-site images are processed through trace projection, line center pixel sequence generation, and edge locking image enhancement processes. Real-world points are processed through component space frame point counting, trace-based real-world patch generation, and traceless real-world patch generation processes. The acquisition pose is used for coordinate transformation between the BIM as-built model, process surfaces, three types of traces, and real-world points, so that both image processing results and real-world 3D results fall into the processing chain of the component index table. At the same acquisition time, if multiple component numbers fall into the same field image, the system reads the component number, the current process number, and the acquisition pose item by item according to the component index table, and generates a process surface rule table, a trace block table, and a three-trace fragment table respectively; when reinforced trace blocks of different component numbers have overlapping projections, the component number, trace type, original edge position, and reinforced edge position are retained respectively; when the same real point falls into multiple reinforced trace blocks, point number sets are formed according to the trace type, and the same point number under the same trace type is only counted once; when walls, beams, slabs, openings, embedded sleeves, and equipment foundations are in the same field of view, they are still entered into three-trace segmentation and clustering according to the component number and the current process number respectively; When formwork, safety nets, material stacks, or construction machinery obstruct the construction site, the real-world points formed by the obstructions are first merged into either traceable or traceless real-world fragments according to their projected pixel range. Then, the source end is determined based on the hit value of the required trace line, the residual value of the required trace line, and the appearance value of the prohibited trace line. When the appearance value of the prohibited trace line is zero, the hit value of the required trace line is zero, and the residual value of the required trace line is positive, the obstruction source end is established. When all three types of trace line values ​​are zero, the temporary occupation source end is established. When the appearance value of the prohibited trace line is positive, the deviation suspected source end is established. The three-trace segmentation and clustering uses the edge cost and the adjacent edge cutting cost to jointly constrain the fragment ownership. When the total cost after expansion is not less than the total cost before expansion, the label before expansion is retained, so that the obstructed fragments, the temporarily occupied fragments, and the deviation suspected fragments are respectively entered into the corresponding source fragment clusters. BIM real-scene monitoring results include the difference zone number, component number, current process number, data collection time, source cluster category, difference type, residual value of the removal line, and occurrence value of the prohibition line. When the difference type is a reasonable process-state difference, a construction stage record is generated; when the difference type is an obstruction difference or a temporary occupation difference, a site review record is generated; when the difference type is a data collection invisible difference, a supplementary data collection record is generated; when the difference type is a construction deviation difference, a quality inspection record is generated; when the difference type is a suspected change difference, a design version review record is generated. Each record is linked to the difference zone number and component number, and enters the subsequent processing flow of the construction management system according to the difference type.

[0024] Furthermore, the smart construction site BIM real-scene 3D fusion monitoring system includes: The index generation module is used to acquire the BIM as-built model, construction process records and on-site 3D data of the target construction area, determine the current process number according to the component number, process time period and acquisition pose and generate a component index table. The process conversion module converts the geometric points of components in the BIM as-built model into process surfaces bound to the current process number based on the component number, the current process number, and the acquisition pose in the component index table. It also generates lines to be left, lines to be removed, and lines to be prohibited from being left based on the process surfaces and outputs a process surface rule table. The trace enhancement module projects the process surface, traces to be retained, traces to be removed, and traces to be prohibited onto the field image based on the acquired pose, generates a line center pixel sequence and original edge position, performs image enhancement through line center accumulation, adjacent gray freezing, and edge position recalculation, and classifies the original edge position and enhanced edge position into the same pixel interval, and outputs a trace block table. The fragment segmentation module is used to generate a three-trace fragment table by combining the enhanced trace blocks in the trace block table with the actual scene 3D data according to the acquisition pose. It generates a source end judgment table based on the hit value of the trace line that should be retained, the residual value of the trace line that should be removed, and the occurrence value of the trace line that should not appear. It also constructs a three-trace conservation cut map and merges the actual scene fragments into process fragment clusters, occluded fragment clusters, temporary occupied fragment clusters, and suspected offset fragment clusters through three-trace segmentation clustering, and outputs a source fragment cluster table. The difference determination module projects the BIM as-built model based on the collected pose to generate the BIM pixel range, compares the gaps and increments with the projected pixel range in the source cluster table, and distinguishes the differences according to the source cluster category, the residual value of the trace line, and the appearance value of the prohibited trace line, and outputs the BIM real-scene monitoring results.

[0025] Working principle: This solution first reads the BIM as-built model, construction process records, and on-site 3D data. It determines the current process number based on component number, process time period, and acquisition pose, generating a component index table. Then, it converts the geometric points of the BIM components into process surfaces for the current process, generating required trace lines, removable trace lines, and prohibited trace lines based on the process sequence. Subsequently, it projects the process surfaces and the three types of trace lines onto the on-site image, performing edge-locked image enhancement on the trace line areas to enhance trace line features without affecting edge positioning. The enhanced trace line blocks are then bound to the actual points to generate a three-trace fragment table. Through three-trace segmentation image clustering, the actual scene fragments are categorized into process clusters, occluded clusters, temporarily occupied clusters, and suspected misaligned clusters. Finally, it compares the projection range of the source clusters with the BIM pixel range using gap and increment comparisons, distinguishing the causes of differences based on cluster origin and trace line value, and outputting the BIM real-scene monitoring results. In practical applications, for example, when a shear wall on a certain floor is in the formwork stage, the BIM as-built model shows the completed wall surface, but the on-site images mainly show formwork seams, rebar edges, obstructed edges, and local material occupation. The system first determines that the shear wall is in the formwork process based on the acquisition time, then projects the formwork process surface onto the on-site image, extracting the formwork seams as lines to be left, the exposed edges of the rebar as lines to be removed, and the edges of holes that appear only after formwork removal as lines to be prohibited from being displayed. After image enhancement and image clustering, if a certain area of ​​the actual site mainly hits the lines to be left and does not have any lines to be prohibited from being displayed, it is classified into a process cluster, and subsequent differences are judged as reasonable process state differences. If lines to be prohibited from being displayed appear and form a suspected misaligned cluster, it enters the construction misalignment difference judgment, thereby avoiding misjudging the normal formwork state as wall defects or construction abnormalities.

[0026] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart construction site BIM real-scene 3D fusion monitoring method, characterized in that, include: S1. Obtain the BIM as-built model, construction process records and on-site 3D data of the target construction area, determine the current process number according to component number, process time period and acquisition pose and generate a component index table; S2. Based on the component number, current operation number and acquisition pose in the component index table, convert the component geometric points in the BIM as-built model into process surfaces bound to the current operation number, and generate the required trace lines, required removal trace lines and prohibited trace lines based on the process surfaces, and output the process surface rule table. S3. Based on the acquisition pose, project the process surface, the lines to be retained, the lines to be removed, and the lines to be prohibited onto the field image, generate the line center pixel sequence and the original edge position, perform image enhancement through line center accumulation, adjacent gray freezing and edge position recalculation, and classify them according to the original edge position and the enhanced edge position being in the same pixel interval, and output the line block table. S4. Generate a three-trace fragment table by combining the enhanced trace blocks in the trace block table with the actual scene 3D data according to the acquisition pose. Generate a source end judgment table based on the hit value of the trace line that should be retained, the residual value of the trace line that should be removed, and the occurrence value of the trace line that should not appear. Construct a three-trace conservation cut graph. Merge the actual scene fragments into process fragment clusters, occluded fragment clusters, temporary occupied fragment clusters, and suspected offset fragment clusters through three-trace segmentation and clustering. Output the source fragment cluster table. S5. Based on the collected pose, project the BIM as-built model to generate the BIM pixel range, compare the gap and increment with the projected pixel range in the source cluster table, and distinguish the differences according to the source cluster category, the residual value of the trace line and the appearance value of the prohibited trace line, and output the BIM real scene monitoring results.

2. The smart construction site BIM real-scene 3D fusion monitoring method according to claim 1, characterized in that: S1 includes: S1-1. Read the geometric points of the component under the same component number in the BIM as-built model, sort the coordinates of the component geometric points on the X-axis, Y-axis and Z-axis in ascending order, and generate the component spatial frame with the first and last coordinates of each axis, and output the component spatial frame table. S1-2. Read the process number, process start time, process end time and process sequence number under the same component number in the construction process record. Perform half-open interval assignment calculation on the acquisition time of the on-site real scene 3D data with the process start time and process end time. Determine the process number of the process with the process start time not later than the acquisition time and the acquisition time earlier than the process end time as the current process number. If the acquisition time and the process end time are the same, determine the process number with the later process sequence number as the current process number. Associate the current process number with the component space frame and output the component process frame table. S1-3. Based on the acquired pose, convert the component space frame in the component process frame table to the coordinate system of the actual scene 3D data, count the number of real points in the component space frame, and generate an index row with component number, current process number, acquired pose and number of real points, and output the component index table.

3. The smart construction site BIM real-scene 3D fusion monitoring method according to claim 2, characterized in that: S2 includes: S2-1. Read the component number, current operation number and acquisition pose from the component index table. Extract the component geometric points from the BIM as-built model according to the component number, and convert the component geometric points to the on-site real-world 3D data coordinate system according to the acquisition pose. Output the component acquisition point table. S2-2. Read the process sequence number of the current process in the construction process record, extract the component collection points with the same process sequence number value, and generate process surfaces by projecting them according to the collection pose. Output the process surface table.

4. The smart construction site BIM real-scene 3D fusion monitoring method according to claim 3, characterized in that: S2 further includes: S2-3. Record the process sequence number of the current process as the current sequence number. Perform coordinate difference between the process range whose process sequence number value is less than the current sequence number and the process surface. Determine the line segments that are still on the boundary of the process surface after the difference as the lines to be left with traces. Determine the line segments that are off the boundary of the process surface after the difference as the lines to be removed with traces. Output the trace removal line table. S2-4. Perform coordinate intersection calculations on the process surface for the process range whose process sequence value is greater than the current sequence value. Determine the line segment obtained from the intersection as the no-mark line, and generate a process surface rule table using the component number, the current process number, the process surface, the mark line to be retained, the mark line to be removed, and the no-mark line.

5. The smart construction site BIM real-scene 3D fusion monitoring method according to claim 4, characterized in that: S3 includes: S3-1. Read the process surface, trace lines to be retained, trace lines to be removed and trace lines to be prohibited from appearing in the process surface rule table. Based on the acquisition pose, project the process surface and the three types of trace lines onto the field image. Generate the line center pixel sequence along the pixel order of the trace line projection line. Generate the trace line projection table with pixel number, trace line type and line direction number. S3-2. Read the line center pixel sequence from the trace projection table, extract the left and right neighbor pixels of each line center pixel along the perpendicular direction of the trace projection line, calculate the absolute value of the gray level difference between the line center pixel and the left neighbor pixel as the left gray level difference, calculate the absolute value of the gray level difference between the line center pixel and the right neighbor pixel as the right gray level difference, when the left gray level difference is not less than the right gray level difference, determine the pixel interval between the line center pixel and the left neighbor pixel as the original edge position, when the left gray level difference is less than the right gray level difference, determine the pixel interval between the line center pixel and the right neighbor pixel as the original edge position, and output the trace edge position table.

6. The smart construction site BIM real-scene 3D fusion monitoring method according to claim 5, characterized in that: S3 further includes: S3-3. Read the line direction number, pixel number and original edge position from the edge position table, execute the edge position locking enhancement algorithm consisting of line center accumulation, neighbor gray freezing and edge position recalculation, accumulate the gray values ​​of each line center pixel under the same line direction number along the pixel number to form the line center enhancement value, and keep the gray values ​​of the left neighbor pixel and the right neighbor pixel from participating in the accumulation, and output the enhanced edge block. S3-4. Read the enhanced etcher blocks, calculate the absolute difference between the enhanced value of the center pixel and the gray value of the left neighboring pixel as the enhanced left gray difference, and calculate the absolute difference between the enhanced value of the center pixel and the gray value of the right neighboring pixel as the enhanced right gray difference. When the enhanced left gray difference is not less than the enhanced right gray difference, the pixel interval between the center pixel and the left neighboring pixel is determined as the enhanced edge position. When the enhanced left gray difference is less than the enhanced right gray difference, the pixel interval between the center pixel and the right neighboring pixel is determined as the enhanced edge position. Then, the enhanced etcher blocks with enhanced edge positions and original edge positions in the same pixel interval are classified according to the etcher type, and the etcher block table is output.

7. The smart construction site BIM real-scene 3D fusion monitoring method according to claim 6, characterized in that: S4 includes: S4-1. Read the enhanced trace blocks and the 3D data of the actual scene from the trace block table. Based on the acquisition pose, project the actual points onto the scene image. Merge the actual points whose projections fall into the same enhanced trace block and whose depth order is adjacent into traced actual scene fragments. Merge the actual points whose projections do not fall into the enhanced trace block and whose projection pixel ranges are connected into traceless actual scene fragments. Record the number of actual points in the actual scene fragments that belong to trace lines to be retained, trace lines to be removed, and trace lines to be prohibited as the trace line hit value, trace line residual value, and trace line appearance value, respectively. Output the three-trace fragment table. S4-2. Read the three-trace fragment table and generate a source determination table based on the three types of trace values. The process source is established when the prohibited trace value is zero, the required trace hit value is positive, and the required trace residue value is zero; otherwise, the process source is not established. The occlusion source is established when the prohibited trace value is zero, the required trace hit value is zero, and the required trace residue value is positive; otherwise, the occlusion source is not established. The temporary occupation source is established when the prohibited trace value is zero and both the required trace hit value and the required trace residue value are positive, or when all three types of trace values ​​are zero; otherwise, the temporary occupation source is not established. The suspected offset source is established when the prohibited trace value is positive; otherwise, the suspected offset source is not established.

8. The smart construction site BIM real-scene 3D fusion monitoring method according to claim 7, characterized in that: S4 further includes: S4-3. Construct a three-trace cut graph based on the three-trace fragment table and the source end determination table. Use real scene fragments as cut graph nodes and process source end, occlusion source end, temporary occupation source end, and offset suspected source end as label ends. When the source end is established, the edge cost of the process source end is the sum of the residual value of the trace line that should be removed and the occurrence value of the trace line that should appear. The edge cost of the occlusion source end is the sum of the hit value of the trace line that should be retained and the occurrence value of the trace line that should appear. The edge cost of the temporary occupation source end is the sum of the occurrence value of the trace line that should appear plus the absolute number of the difference between the hit value of the trace line that should be retained and the residual value of the trace line that should be removed. The edge cost of the offset suspected source end is the sum of the hit value of the trace line that should be retained and the residual value of the trace line that should be removed. When the source end is not established, the edge cost is the sum of the number of real scene points, the hit value of the trace line that should be retained, the residual value of the trace line that should be removed, and the occurrence value of the trace line that should be removed. S4-4. Generate adjacent edges between real scene fragments whose projected pixel ranges are connected. Subtract the absolute number of the difference between the same type of trace values ​​from the sum of the trace values ​​of two real scene fragments, and divide by two to obtain the common trace value. Use the sum of the three types of common trace values ​​as the adjacent edge cutting cost. Perform three-trace segmentation clustering in the order of process source end, occlusion source end, temporary occupation source end, and offset suspected source end. Use the sum of the label end edge cost and the cut-off cost of the adjacent edge to be used as the total cost. If the total cost after expansion is less than the total cost before expansion, use the expanded label. Otherwise, retain the label before expansion. If the label end names are the same in both rounds, stop. Otherwise, enter the next round of expansion. Merge real scene fragments with the same label end name and whose projected pixel ranges are connected into process fragment clusters, occlusion fragment clusters, temporary occupation fragment clusters, and offset suspected fragment clusters. Output the source fragment cluster table.

9. The smart construction site BIM real-scene 3D fusion monitoring method according to claim 8, characterized in that: S5 includes: S5-1. Read the projected pixel range and BIM as-built model from the source patch table, generate the BIM pixel range based on the collected pose, determine the area in the BIM pixel range that is not occupied by the projected pixel range as the gap difference area, determine the area in the projected pixel range that does not fall into the BIM pixel range as the incremental difference area, and output the difference area table. S5-2. Read the difference area table and the source patch table, associate the notch difference area with the source patch that is connected to the projected pixel range, associate the incremental difference area with the source patch that overlaps with the projected pixel range, and extract the source patch category, the residual value of the trace line that should be removed and the appearance value of the trace line that should not appear, and output the difference judgment table. S5-3. Read the difference judgment table. If the process area cluster has a zero value for the forbidden trace line, it is judged as a reasonable process state difference. If it does not meet the requirements, it is judged as an occlusion difference, a temporary occupation difference, a collection invisible difference, and a construction deviation difference in the following order: occlusion area cluster, temporary occupation area cluster, unassociated source area cluster, and suspected offset area cluster. If it still does not meet the requirements, it is judged as a suspected change difference. Output the BIM real scene monitoring results.

10. A smart construction site BIM real-scene 3D fusion monitoring system, characterized in that, include: The index generation module is used to acquire the BIM as-built model, construction process records and on-site 3D data of the target construction area, determine the current process number according to the component number, process time period and acquisition pose and generate a component index table. The process conversion module converts the geometric points of components in the BIM as-built model into process surfaces bound to the current process number based on the component number, the current process number, and the acquisition pose in the component index table. It also generates lines to be left, lines to be removed, and lines to be prohibited from being left based on the process surfaces and outputs a process surface rule table. The trace enhancement module projects the process surface, traces to be retained, traces to be removed, and traces to be prohibited onto the field image based on the acquired pose, generates a line center pixel sequence and original edge position, performs image enhancement through line center accumulation, adjacent gray freezing, and edge position recalculation, and classifies the original edge position and enhanced edge position into the same pixel interval, and outputs a trace block table. The fragment segmentation module is used to generate a three-trace fragment table by combining the enhanced trace blocks in the trace block table with the actual scene 3D data according to the acquisition pose. It generates a source end judgment table based on the hit value of the trace line that should be retained, the residual value of the trace line that should be removed, and the occurrence value of the trace line that should not appear. It also constructs a three-trace conservation cut map and merges the actual scene fragments into process fragment clusters, occluded fragment clusters, temporary occupied fragment clusters, and suspected offset fragment clusters through three-trace segmentation clustering, and outputs a source fragment cluster table. The difference determination module projects the BIM as-built model based on the collected pose to generate the BIM pixel range, compares the gaps and increments with the projected pixel range in the source cluster table, and distinguishes the differences according to the source cluster category, the residual value of the trace line, and the appearance value of the prohibited trace line, and outputs the BIM real-scene monitoring results.

Images