A floor engineering acceptance system based on image acquisition

Abstract

The application relates to the technical field of image processing, and particularly discloses a floor engineering acceptance system based on image acquisition, which is used for solving the problems that in floor acceptance images, measuring points and directions are random, near edges and holes are interfered, reconstruction noise and scales are not unified, virtual reading is easy to cross holes and borders, calibers are inconsistent, it is difficult to align with on-site reading data one by one, and traceable review is lacked, and the system comprises an imaging acquisition unit, a three-dimensional reconstruction calibration unit, a reference surface fitting and detrending unit, a height map gridding unit, a wavelet scale management unit, a dimension mapping unit, a boundary and hole processing unit, a candidate measuring area generation unit, a virtual reading determination unit and a data storage and output unit; through automatic reading consistent with on-site reading caliber, the floor acceptance image is prevented from crossing holes and borders, near edge interference is inhibited, traceable measurement alignment is established, error boundaries and reserved digits are determined, and unified statistical caliber and version records are established.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a floor slab engineering acceptance system based on image acquisition. Background Technology

[0002] In the conventional practice of floor slab acceptance, floor slabs are often inspected on-site using a two-meter straightedge and feeler gauge. However, because the specifications do not clearly specify the layout of measuring points and the direction of the inspection, inspectors place the straightedge in any direction and take the maximum height difference, resulting in random results that cannot fully reflect the flatness of the entire slab. At the same time, it is emphasized that the measuring points should not be close to the component cross-section. In engineering, an empirical threshold of no less than 20cm from the boundary is often used to reduce the edge effect. This situation has prompted current research to shift towards an automated acceptance path based on digital data. The literature published on CNKI (China National Knowledge Infrastructure) (the paper "Intelligent Segmentation and Flatness of Indoor Scenes Based on Point Cloud Data" by Cao Yuxing of Chongqing University) The "Integer Detection" method uses 148 sets of indoor scene point clouds from 15 residential buildings as a basis. First, the scenes are manually segmented and semantically labeled to form a dataset containing 8 types of components, such as walls and the upper and lower parts of the floor slabs. The component-level recognition ability of point cloud segmentation networks such as RandLA-Net is trained and verified as input for subsequent detection. Then, on the segmented target surface, two-dimensional wavelet transform is introduced to locate areas with significant undulations. Near-edge points are proposed by combining the principle of not being close to the edge and the density threshold. Finally, a simulated straightedge is used to output flatness values ​​of the same diameter as the manual flatness at the test points. Manual and simulated comparison experiments are carried out in multiple residential buildings to verify the effectiveness of the process. Given the continuous advancements in existing sensing hardware, reconstruction algorithms, and image processing technologies, a crucial task is to establish a one-to-one, reversible, and reproducible mapping between the target region of floor slab images, wavelet responses at different scales under constraints of unified sampling, detrending, and boundary / hole processing, and on-site ruler readings. Furthermore, it is essential to clarify the correspondence between pixel spacing and scale-spatial wavelength, response amplitude normalization, threshold values ​​for candidate regions, and the boundary values ​​for sub-pixel extreme value localization. This involves addressing the uncertainty propagation and consistent representation from reconstruction noise to the final reading, thereby forming a traceable metrological alignment baseline. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a floor slab engineering acceptance system based on image acquisition. By automatically reading the data with the same diameter as the on-site straightedge, the system avoids cross-hole boundary of the target area of ​​the floor slab acceptance image and suppresses near edge and noise interference, so that the flatness results are continuous and stable, the regions are comparable, and the versions are verifiable. The system establishes traceable measurement alignment, clarifies the error boundary and the number of digits to be retained, and unifies the statistical caliber and version records.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A floor slab engineering acceptance system based on image acquisition includes an imaging acquisition unit, a 3D reconstruction and calibration unit, a reference plane fitting and detrending unit, a height map rasterization unit, a wavelet scaling unit, a dimensional mapping unit, a boundary void processing unit, a candidate area generation unit, a virtual straightedge measurement unit, and a data storage and output unit. The imaging acquisition unit acquires multi-view floor slab images and simultaneously delivers camera position and time information to the 3D reconstruction and calibration unit. The 3D reconstruction and calibration unit performs intrinsic and extrinsic parameter calibration, distortion correction, attitude calculation, and dense reconstruction of the floor slab images, outputting a unified coordinate point cloud and calibration quality record. The reference plane fitting and detrending unit fits the image reference plane within the target area to generate a deviation field for the height map rasterization unit to process according to the image. The unit generates height maps and raster indexes based on the element spacing. The wavelet scale management unit calculates multi-scale responses within a preset scale set and establishes a scale-to-spatial wavelength correspondence table. The dimensional mapping unit constructs a piecewise monotone invertible mapping based on the scale codebook and standard sample blocks of the target measurement area in the floor image and retrieves entries using parameter fingerprints. The boundary hole processing unit generates a non-near-edge mask for the target area of ​​the floor image and repairs holes according to structural elements. The candidate area generation unit outputs a list of measurement area images and numbers according to the connected component threshold and minimum area. The virtual ruler measurement unit performs image sub-pixel extreme value interpolation within the selected scale and reading aperture and generates readings. The data storage and output unit archives the parameters of each unit, the acquisition batch identifier, the coordinates and readings of the measurement points, and the transformation version.

[0006] As a further aspect of the present invention, in the imaging acquisition unit, the target area is divided into a boundary zone, a transition zone, and a central zone based on the horizontal distance L to the nearest boundary, with the outer edge and internal void boundary of the horizontal projection of the floor image as references. The L values ​​of the boundary zone, transition zone, and central zone are less than or equal to 30cm, greater than 30cm but less than or equal to 80cm, and greater than 80cm, respectively. The number of sampling frames is calculated from the partition reference value and adjusted item by item according to the discrete conditions of angle coverage, baseline coverage, matching point density, reprojection residual, motion blur, brightness deviation, occlusion rate, and near-edge safety margin. The upper and lower limits of the boundary zone, transition zone, and central zone are set, and the frame number is rounded up step by step under the room-level frame number budget constraint. The camera position index is generated based on the results.

[0007] As a further aspect of the present invention, the three-dimensional reconstruction calibration unit performs sequence alignment of the multi-view images output by the imaging acquisition unit based on the frame time identifier and camera position index. It establishes an intrinsic parameter version record according to the shooting batch and lens status and writes in the focal length, principal point position, pixel size, radial distortion coefficient, and pixel non-orthogonal term. It solves the station attitude based on the circumferential and radial baseline positions and the line-of-sight angle sequence as geometric priors and generates a camera attitude record. It performs image distortion correction and dense reconstruction to obtain a unified coordinate system point cloud and generates a calibration quality record associated with the station number. The calibration quality record includes the reprojection residual quantile value, field-of-sight overlap ratio, baseline position coverage identifier, and angle coverage identifier. At the same time, it exports the parameter fingerprint field and writes it into the data storage output unit.

[0008] As a further aspect of this invention, the reference plane fitting detrending unit, under the constraints of unified coordinate point cloud, camera position attitude, baseline and angle coverage markers, boundary and hole masks, pixel spacing, partitioning rules, and height map grid index, generates fitting blocks and numbers them according to the boundary zone, transition zone, and center zone of the floor target area image with preset block size and step. Candidate blocks are screened out according to the hole ratio and near-edge buffer, and the baseline and angle coverage are checked. If they are not satisfied, they are adjusted or supplemented in the same partition according to the step. The in-point threshold is set according to the reprojection residual quantile value of the calibration quality record. The point cloud in the block is weighted by combining the view overlap and baseline coverage, and the weight attenuation is set according to the near-edge buffer distance. The fitting is completed, and the detrending bias field is output.

[0009] As a further aspect of this invention, the heightmap rasterization unit establishes a multi-layer raster with differentiated partitions for the target region of the floor image under constraints of unified coordinate point cloud, detrended bias field, partition labels, boundary masks, hole masks, near-edge buffer distance, block size, and step size. The pixel spacing of the base layer is set according to partitions: 10 mm for the boundary zone, 20 mm for the transition zone, and 30 mm for the center zone. The correspondence between partitions and pixel spacing is recorded. The grid origin is taken as the coordinates of the lower left corner of the smallest bounding rectangle of the target region of the floor image, rounded down according to the pixel spacing. Row and column indices are generated for each layer in row priority order and a one-to-one correspondence is established with the fitted block number and partition category. Pixel aggregation adopts a medium... A bit-first strategy is adopted, and when the number of valid observations is less than five, the truncated mean is used and a no-value mark is output. Validity masks, boundary masks, hole masks, and non-near-edge masks are generated simultaneously. For each image cell, the number of observations, the number of outliers, and the aggregation window radius are recorded to form a raster index table for the floor image and write it into the parameter fingerprint field. The raster index table includes row and column ranges, layer numbers, cell spacing, origin coordinates, and index mode. The cell spacing setting, number of layers, origin coordinates, aggregation statistics method, valid observation threshold, mask type and version, and index mode are written into the parameter fingerprint field and stored in the data storage output unit.

[0010] As a further aspect of this invention, the wavelet scale management unit generates a scale set and a scale index under constraints such as the height map of the target region in the floor image, pixel spacing, partition labels, a list of fitted blocks, boundary masks, hole masks, near-side buffer distances, and reading apertures. The scale set consists of local scale sequences and global scale sequences, discretized geometrically with a scaling factor between 1:1.25 and 1:2. The spatial wavelength ranges of the local and global scales are 300 mm to 600 mm and 1000 mm to 3000 mm, respectively. A scale-spatial wavelength correspondence table is established using fixed coefficients. For each fitted block, the scale is processed based on the short side length and near-side buffer distance of that block. Effective scales are filtered. If the transform kernel radius corresponding to a scale exceeds 40% of the short side length of the block or crosses the near-side buffer, the scale is disabled in the block and its identifier is recorded. Normalization factors and cross-scale weights are configured for each scale. The normalization factors are set according to the scale power, and the power exponent is given by the parameter rule base. The cross-scale weights are set according to the logarithmic interval of the scales. When the reading aperture is between adjacent scales, the responses of adjacent scales are interpolated according to the weights and an aperture-bound scale index is generated. The wavelet scale management unit outputs the scale-to-spatial wavelength correspondence table, effective scale mask, scale normalization configuration, cross-scale weight table and aperture-bound index and writes them to the data storage output unit for subsequent unit calls.

[0011] As a further aspect of the present invention, the process by which the dimensional mapping unit constructs a piecewise monotonically invertible mapping based on the scale codebook of the target measurement area of ​​the floor slab image and standard sample blocks, and retrieves entries using parameter fingerprints, includes:

[0012] Step 1, Calibration Data Acquisition: Under the given scale set and reading aperture, collect wavelet response and millimeter reading paired samples from standard sample blocks or known true value sample areas, and record the pixel spacing, scale and spatial wavelength corresponding sequence number, reference surface model, detrending method identifier, boundary buffer distance, minimum repair aperture, reading aperture, sub-pixel interpolation method, normalized power exponent, cross-scale weight version, acquisition batch and equipment number.

[0013] Step 2, Parameter fingerprint generation: Encode the above fields in a fixed order to form a unique fingerprint, which is used as the primary index key of the mapping entry;

[0014] Step 3, Data Preprocessing: Standardize units and dimensions, configure response normalization according to scale, remove missing and incomplete records, and truncate outliers according to preset quantiles;

[0015] Step 4, Segmentation Strategy Definition: Based on the response distribution or reading caliber, divide the mapping domain into several adjacent intervals, clarify the upper and lower limits and monotonic direction of each interval, and set inter-segment continuity constraints.

[0016] Step 5, Function family fitting: Fit a function that is monotonic and invertible within each interval, fix the function type and parameters, and record the interval domain and extrapolation constraints;

[0017] Step 6, Entry Generation and Archiving: Generate mapping entries for each scale. The entry content includes fingerprint, function type, function parameters, interval domain, unit and precision caliber, extrapolation limit, residual statistics, acquisition batch and device identifier, and version number.

[0018] Step 7, Input and Indexing: Write the mapping entries into the scale codebook, establish a retrieval structure with parameter fingerprints as the main index, and add secondary indexes of scale code, caliber code, timestamp and device number;

[0019] Step 8, runtime retrieval: The processing flow generates parameter fingerprints based on the current data, retrieves matching entries in the scale codebook, and returns the entry number and function parameters if a match is found. When there are no completely matching entries, the adjacent entries are interpolated according to the scale and caliber nearest neighbor relationship of adjacent fingerprints and cross-scale weights to synthesize caliber-bound entries, generate new entry numbers, and cache them in the database.

[0020] Step 9, Version and Range Verification: Perform version consistency verification, parameter range verification, and domain coverage verification on the retrieved entries. If the requirements are not met, trigger a rollback strategy.

[0021] Step 10, Binding and Output: Establish a one-to-one binding relationship between the entry number and the virtual ruler reading caliper, generate a caliper binding index, and write the mapped entry number, function parameter, fingerprint, and version record into the data storage output unit.

[0022] As a further aspect of the present invention, the process of the boundary hole processing unit generating a non-near-edge mask for the target area of ​​the floor slab image and repairing holes according to structural elements includes:

[0023] First, a binary boundary map is generated based on the outer edge of the floor slab projection and the boundary of the internal opening. The distance field to the nearest boundary is calculated. A buffer expansion is applied to the boundary map based on the non-near-edge distance threshold to obtain a forbidden zone. This forbidden zone is then merged with an existing invalid cell mask to form an initial non-near-edge mask. Connectivity analysis is performed on the height map or voxel grid to extract hole regions, recording the hole area, major and minor axes, and equivalent diameter of the circle. Morphological closing operations or hole filling are performed on the initial mask according to preset structuring element types and sizes. Only holes with an equivalent diameter smaller than the minimum repairable hole diameter are repaired; holes larger than this threshold remain forbidden. Repairing connected holes that cross the boundary is prohibited within the near-edge buffer range. The updated non-near-edge mask and hole mask are output, and the structuring element type, structuring element size, distance threshold, minimum repairable hole diameter, connected zone threshold, buffer width, and mask version identifier are recorded.

[0024] As a further aspect of the present invention, the process by which the candidate area generation unit outputs a list and number of candidate area images according to the connected component threshold and the minimum area includes: inputting a multi-scale response map, an effectiveness mask, a non-near-edge mask, a hole mask, a raster index, and a connected component threshold and a minimum area; performing an AND operation between the response map and various masks to obtain a candidate binary map, uniformly selecting a four-connected connectivity aperture; marking the connected components of the candidate binary map to generate a connected component list and a pixel index set; calculating the area, effective pixel count, minimum bounding rectangle, principal direction, and centroid coordinates for each connected component; removing fragments according to the connected component threshold and removing fragments according to the minimum area threshold. Except for connected components with insufficient area, connected components with a minimum width less than the preset value are removed; morphological or topological merging is performed on adjacent connected components with a gap less than the preset distance, and statistical information and cell index are updated; the retained connected components are sorted according to the area priority rule, and region numbers are assigned sequentially from the beginning, and a mapping table between the number and cell index and geometric attributes is established; a survey area list is generated, which includes region number, partition, row and column range, boundary polygon vertex sequence, centroid, area, main direction, and cell index, and records the connectivity caliber used, connected component threshold, minimum area and version identifier, and writes it to the data storage output unit.

[0025] As a further aspect of this invention, the virtual ruler measuring unit, based on the detrended height map within the waiting area of ​​the floor image, performs omnidirectional sliding according to the measurement window set in the configuration table, with directional steps not exceeding five degrees and displacement steps not exceeding twenty centimeters. The window profile signal is extracted sequentially by the grid index and sub-pixel interpolation is performed using cubic splines. Extreme value positioning adopts a dual judgment method of neighborhood three-point fitting and spline extreme value verification. When the endpoint or interior falls into invalid pixels, holes, or non-near-edge masks, judgment is made based on the continuous invalid segment length threshold and the total invalid percentage threshold. When the continuous invalid segment length is not greater than 20 cm and the total invalid percentage is not greater than 20%, interpolation is used to fill the gap. If either threshold is exceeded, the endpoint is retreated by a fixed step size, and the number of repositioning times is recorded. When a window crosses an opening, the largest segment profile is used as the reading basis, and splicing across openings is prohibited. When the window approaches the boundary and causes an overshoot, the window is clipped in that direction with priority given to the mask that is not close to the edge. The correspondence between the window and the scale set is determined by the aperture binding index given by the wavelet scale management unit. When the reading aperture is between adjacent scales, the responses of adjacent scales are interpolated according to the cross-scale weight and bound to a single aperture index. The generated readings, window center coordinates, orientation angle, endpoint coordinates, spline kernel radius, orientation step and displacement step, invalid segment statistics, interpolation participation identifier, relocation record, reading retention digits and rounding rules are written to the data storage output unit and a one-to-one correspondence is established with the mapping entry number returned by the dimensional mapping unit.

[0026] The technical advantages of the image acquisition-based floor slab engineering acceptance system of the present invention are as follows:

[0027] This invention introduces frame number adaptation based on the distance to the nearest boundary and baseline and angle coverage constraints. It solidifies posture and scale through time-camera alignment, versioning of internal and external parameters, and quality recording. It uses partitioned sliding blocks, near-edge buffering, and a two-stage weighted fitting to output a detrended bias field. It unifies pixel spacing and indexing through partitioned differentiated multi-layer raster. It binds reading caliber to scale through scale-spatial wavelength correspondence, effective scale mask, and cross-scale weighting. It constructs a segmented monotonically reversible response-millimeter mapping with version verification using scale codebook and parameter fingerprint. It unifies forbidden areas through non-near-edge masking and hole repair. It forms a closed loop for reading generation through omnidirectional sliding of a virtual ruler, sub-pixel extrema, and invalid segment determination. Ultimately, it achieves a unified caliber and traceable output across the entire chain from acquisition, reconstruction, detrended, raster, scale, mapping to reading. This enables consistent, cross-scale, and cross-batch reuse of floor flatness determination in millimeter dimensions under complex conditions of multiple devices, multiple scenes, and boundaries and openings. Attached Figure Description

[0028] Figure 1 This is a system block diagram of the present invention;

[0029] Figure 2 This is a physical image of a 7m x 5m floor slab according to the present invention;

[0030] Figure 3 This is a schematic diagram of the detrending height map, zoning, and equipment well of the present invention;

[0031] Figure 4 Before processing, this is the virtual ruler 2-meter window range diagram (along the X direction) of the present invention;

[0032] Figure 5 This is a processed diagram of the 2-meter window range of the ruler (along the X direction) for the present invention. Detailed Implementation

[0033] 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.

[0034] like Figure 1As shown, the present invention proposes an image acquisition-based floor slab engineering acceptance system, comprising an imaging acquisition unit, a 3D reconstruction and calibration unit, a reference surface fitting and detrending unit, a height map rasterization unit, a wavelet scaling unit, a dimensional mapping unit, a boundary void processing unit, a candidate area generation unit, a virtual straightedge measurement unit, and a data storage and output unit. The imaging acquisition unit acquires multi-view floor slab images and simultaneously delivers camera position and time information to the 3D reconstruction and calibration unit. The 3D reconstruction and calibration unit completes the calibration of intrinsic and extrinsic parameters of the floor slab images, distortion correction, attitude calculation, and dense reconstruction, and outputs a unified coordinate point cloud and calibration quality record. The reference surface fitting and detrending unit fits the image reference surface within the target range to generate a deviation field for height map rasterization. The unit generates height maps and raster indexes according to pixel spacing. The wavelet scale management unit calculates multi-scale responses within a preset scale set and establishes a scale-spatial wavelength correspondence table. The dimensional mapping unit constructs a piecewise monotone invertible mapping based on the scale codebook and standard sample blocks of the target measurement area of ​​the floor image and retrieves entries using parameter fingerprints. The boundary hole processing unit generates a non-near-edge mask for the target area of ​​the floor image and repairs holes according to structural elements. The candidate area generation unit outputs a list of measurement area images and numbers according to the connected component threshold and minimum area. The virtual ruler measurement unit performs image sub-pixel extreme value interpolation within the selected scale and reading aperture and generates readings. The data storage and output unit archives the parameters of each unit, the acquisition batch identifier, the coordinates and readings of the measurement points, and the transformation version.

[0035] The system proposed in this invention uses a closed loop of acquisition, reconstruction, detrending, rasterization, scale management, dimensional mapping, candidate area, virtual ruler, and archiving to unify multi-view images into a detrended height map. Under the constraints of not being near the edge and hole masking, it achieves monotonically reversible conversion down to millimeter readings through wavelet multi-scale response, scale codebook, and parameter fingerprint. Through time and camera position alignment, internal and external parameter version and quality recording, omnidirectional sliding window and sub-pixel extreme value positioning, aperture binding, and version verification, it forms an automated, consistent, and traceable flatness evaluation in complex boundary and multi-device, multi-batch scenarios.

[0036] It should be noted that, as Figure 2 As shown, in the imaging acquisition unit, the target area is divided into a boundary zone, a transition zone, and a central zone based on the horizontal distance L from the nearest boundary, with the outer edge of the horizontal projection of the floor image and the boundary zone of the internal void boundary as references. The L values ​​of the boundary zone, transition zone, and central zone are less than or equal to 30cm, greater than 30cm but less than or equal to 80cm, and greater than 80cm, respectively. The number of sampling frames is calculated from the partition reference value and adjusted item by item according to the discrete conditions of angle coverage, baseline coverage, matching point density, reprojection residual, motion blur, brightness deviation, occlusion rate, and near-edge safety margin. The upper and lower limits of the boundary zone, transition zone, and central zone are set. Under the room-level frame number budget constraint, the frame is rounded up step by step, and the camera position index is generated based on the results.

[0037] Taking the upper surface of a 7m x 5m cast-in-place floor slab as the object, a 0.8m x 0.5m equipment well is set at the northwest corner, 0.2m from the edge of the slab. Zones are defined based on the distance to the nearest boundary: less than or equal to 0.30m is the boundary zone, between 0.30m and 0.80m is the transition zone, and greater than 0.80m is the center zone. The reference frame numbers are set to 6, 4, and 3 respectively. The measured point densities at the east side of the well and the beam edge are 2800 and 3200 points per square meter respectively (threshold not less than 4000), the median reprojection residuals are 0.62 and 0.55 pixels respectively (threshold not greater than 0.50), and the occlusion rate is 34%. With a 20% (threshold not greater than 20) brightness deviation between adjacent frames and a 12% (threshold not greater than 10) brightness deviation, and a near-edge safety margin of 8 cm, 2 additional frames are taken in each boundary zone, 1 additional frame is taken in the transition zone, and the center zone remains unchanged. The camera position index also specifies a circumferential baseline cycle of 30 / 60 / 120 cm, a radial baseline cycle of 40 / 80 cm, alternating principal angles of 32 and 38 degrees, seven levels of pitch from -10 to 20 degrees, and a roll dispersion of -3 / 0 / +3 degrees. The number of frames covered by the sampling points and the azimuth interval are recorded to ensure that the boundary zone is not less than 6 and 3, the transition zone is not less than 4 and 2, and the center zone is not less than 3 and 1.

[0038] It should be noted that the 3D reconstruction calibration unit aligns the multi-view images output by the imaging acquisition unit according to the frame time identifier and camera position index. It establishes an intrinsic parameter version record according to the shooting batch and lens status and writes the focal length, principal point position, pixel size, radial distortion coefficient, and pixel non-orthogonality term. It solves the station attitude based on the circumferential and radial baseline positions and the line-of-sight angle sequence as geometric priors and generates a camera attitude record. It performs image distortion correction and dense reconstruction to obtain a unified coordinate system point cloud and generates a calibration quality record, which is associated with the station number. The calibration quality record includes the reprojection residual quantile value, field of view overlap ratio, baseline position coverage identifier, and angle coverage identifier. At the same time, it exports the parameter fingerprint field and writes it to the data storage output unit.

[0039] Using a 7m x 5m floor slab as the target, the imaging acquisition unit acquired 96 frames of images, writing timestamps and camera position numbers to each frame (36 frames for the boundary zone, 36 frames for the transition zone, and 24 frames for the center zone). The 3D reconstruction calibration unit first generated an alignment record table based on the timestamps and camera position indexes, fixing the position of each frame in the sequence and its corresponding partition / sector; then, it established internal parameter versions according to the shooting batch (the first 48 frames were version one, and the last 48 frames were version two), recording the focal length, principal point position, pixel size, radial distortion, and pixel non-orthogonality for each frame; then, it used preset circumferential / radial baseline settings (0.30 / 0.60 / 1.20m, 0.40 / 0.80m) and line-of-sight angle sequences (32 / 38 degrees and 14 / 18 degrees) at the acquisition end. Using a geometric prior (degree), the attitude of each station is calculated in conjunction with the time-camera index to complete distortion correction and dense reconstruction, outputting a point cloud in a unified coordinate system. Simultaneously, for each station, calibration quality records (reprojection residual quantiles, view overlap ratio, baseline coverage, and angle coverage) are generated and associated, and parameter fingerprint fields (intrinsic and extrinsic parameter versions, pixel spacing / ground resolution estimation, detrending method identifier, boundary buffer and minimum repairable aperture, reading aperture and subpixel interpolation method, scale index, and spatial wavelength coefficient) are exported and written to data storage. By precisely aligning multi-view imagery with the geometric prior in both temporal and station dimensions, subsequent reference plane fitting, layered rasterization, wavelet scale filtering, and virtual ruler readings are all performed under a unified coordinate system, a unified intrinsic parameter version, and a traceable quality standard. This avoids inconsistencies in reading aperture and dimensional mapping caused by frame order drift, intrinsic and extrinsic parameter switching, or incomplete baseline / angle coverage.

[0040] It should be noted that, under the constraints of unified coordinate point cloud, camera position, baseline and angle coverage markers, boundary and hole masks, pixel spacing, partitioning rules, and height map raster index, the reference plane fitting detrending unit generates fitting blocks and numbers them according to the boundary zone, transition zone, and center zone of the floor target area image with preset block size and step. Candidate blocks are screened out according to the hole ratio and near-edge buffer, and the baseline and angle coverage are checked. If they are not satisfied, they are adjusted or supplemented in the same partition according to the step. The in-point threshold is set according to the reprojection residual quantile value of the calibration quality record. The point cloud in the block is weighted by combining the view overlap and baseline coverage, and the weight is attenuated according to the near-edge buffer distance. The fitting is completed, and the detrending bias field is output.

[0041] A 7m x 5m floor slab with a 0.8m x 0.5m equipment well and two beam strips on the upper surface; under unified coordinate point cloud, camera position attitude, baseline / angle coverage markings and boundary, and hole mask constraints, the reference plane fitting detrended unit generates fitting blocks by partition (block size 0.6m, step 0.3m). Candidate blocks with a hole pixel ratio greater than 1 / 4 or falling into the 0.3m near-edge buffer are directly eliminated. Blocks with insufficient baseline or angle coverage are shifted by one step or a supplementary block is inserted in the same partition; the in-point threshold is set with the reprojection residual quantile value given by the internal and external parameter calibration quality record (e.g., median 0.48 pixels, 90% 0.85 pixels). The point cloud within the block is weighted according to the view overlap and baseline coverage, and a distance attenuation weight is applied to points less than 0.5m from the boundary. After initial plane fitting, the residual is eliminated and then constrained plane / quadratic surface fitting is performed, outputting the detrended bias field and block-level parameter records. Compared with the methods of undivided blocks, unmasked blocks, and unweighted blocks, the bias field obtained on the same data shows that the fitting base plane bias of the 50 cm edge band is significantly converged, the residual stripes around the holes are suppressed, the detrending continuity at the block seams decreases, and the median absolute deviation of the overall bias field is reduced, ensuring that subsequent scale screening and virtual ruler readings are performed on a base plane with a uniform diameter.

[0042] It should be noted that the heightmap rasterization unit establishes a multi-layer raster with differentiated partitioning of the target area of ​​the floor image under the constraints of unified coordinate point cloud, detrended bias field, partition labels, boundary mask, hole mask, near-edge buffer distance, block size and step size. The pixel spacing of the base layer is set according to partitions: 10 mm for the boundary zone, 20 mm for the transition zone, and 30 mm for the center zone. The correspondence between partitions and pixel spacing is recorded. The grid origin is taken as the coordinates of the lower left corner of the smallest bounding rectangle of the target area of ​​the floor image, rounded down according to the pixel spacing. Row and column indices are generated for each layer in row priority order and a one-to-one correspondence is established with the fitted block number and partition category. The median is used for pixel aggregation. The priority strategy adopts a truncated mean and outputs a no-value mark when the number of valid observations is less than five. Simultaneously, a validity mask, boundary mask, hole mask, and non-near-edge mask are generated. For each image cell, the number of observations, the number of outliers, and the aggregation window radius are recorded to form a raster index table of the floor image and write it into the parameter fingerprint field. The raster index table includes row and column range, layer number, cell spacing, origin coordinates, and index mode. The cell spacing setting, number of layers, origin coordinates, aggregation statistics method, valid observation threshold, mask type and version, and index mode are written into the parameter fingerprint field and stored in the data storage output unit.

[0043] Taking a 7m x 5m floor slab with a 0.8m x 0.5m equipment well as an example, after the baseline is detrended, based on the partition label and near-edge buffer, the height map raster unit establishes a partitioned differential multi-layer raster: 1 cm for the boundary zone, 2 cm for the transition zone, and 3 cm for the center zone. Row and column indices are generated by aligning the lower left corner of the smallest bounding rectangle with the origin. Thus, the pixel size of each zone is approximately: 6.84 square meters ≈ 68,400 grids for the boundary zone, 9.8 square meters ≈ 24,500 grids for the transition zone, and 18.36 square meters ≈ 20,400 grids for the center zone. The raster index corresponds one-to-one with the fitted block number. During aggregation, the boundary zone adopts median priority, and the truncated mean is enabled when there are fewer than 5 valid observations. The mask layer is output synchronously by combining validity, boundary, hole and non-near-edge mask, and the number of observations, outliers and aggregation window radius are recorded in the pixel attributes. The sparseness and aliases caused by holes and occlusions in the boundary zone are captured and isolated by fine pixels. The proportion of non-value pixels around the holes is reduced from about 27% to about 9%, and the height jump at the joint of the fitted block is reduced from about 1 mm to less than 0.5 mm. For wavelet kernels at a local scale of 300 mm, the sampling density of the boundary zone reaches 30 pixels per kernel diameter (300 mm ÷ 1 cm), the transition zone is 15 pixels, and the central zone is 10 pixels, which meets the sampling requirements of subsequent scale-aperture binding and virtual ruler sliding window. At the same time, multi-layer indexing ensures that window sampling and dimensional mapping are called according to the same row and column coordinate system, avoiding reading aperture drift and scale mismatch caused by mixing different resolutions.

[0044] It should be noted that the wavelet scaling unit generates a scale set and scale index under the constraints of the target region height map, pixel spacing, partition labels, fitted block list, boundary mask, hole mask, near-side buffer distance, and reading aperture in the floor image. The scale set consists of a local scale sequence and a global scale sequence, discretized geometrically with a scaling factor between 1:1.25 and 1:2. The spatial wavelength ranges of the local and global scales are 300 mm to 600 mm and 1000 mm to 3000 mm, respectively. A scale-spatial wavelength correspondence table is established using fixed coefficients. For each fitted block, effective scaling is performed based on the short side length and near-side buffer distance of that block. Scale filtering: If the transform kernel radius corresponding to a scale exceeds 40% of the short side length of the block or crosses the near-side buffer, the scale is disabled in the block and its identifier is recorded. Normalization factors and cross-scale weights are configured for each scale. The normalization factors are set according to the scale power, and the power exponent is given by the parameter rule base. The cross-scale weights are set according to the logarithmic interval of the scales. When the reading aperture is between adjacent scales, the responses of adjacent scales are interpolated according to the weights and an aperture-bound scale index is generated. The wavelet scale management unit outputs the scale-to-spatial wavelength correspondence table, effective scale mask, scale normalization configuration, cross-scale weight table and aperture-bound index and writes them to the data storage output unit for subsequent unit calls.

[0045] Taking a 7m x 5m floor slab with a 0.8m x 0.5m equipment shaft as an example, the wavelet scale management unit generates local scale sets of 300, 450, and 600 mm, and global scale sets of 1500 and 2400 mm on the partitioned differential grid, and provides a scale-spatial wavelength table according to fixed coefficients; for the fitted block with a short side of 0.6m, the available scales are selected according to the rule that the kernel radius does not exceed 40% of the short side and does not cross the 0.3m near-side buffer: 300 mm (radius 150 mm) and 450 mm (radius 225 mm) are retained, and 600 mm (radius 300 mm) is disabled near the buffer; when the reading aperture At a distance of 2 meters, located between the 1500 and 2400 mm scales, aperture binding interpolation is performed on the responses of the two scales according to logarithmic interval weights, and power-law normalization is provided for each scale (given by the parameter table). As a result, the edges and hole neighborhoods will not produce spurious responses due to kernel cross-boundaries. The same reading aperture falls within the unified scale index and normalized aperture on different blocks, different sampling intervals, and different devices. The responses between local and global apertures are directly converted and aligned through the weight table. The output includes a complete set of records containing scale-wavelength tables, effective scale masks, normalization and weight configurations, and aperture binding indexes, which can be directly called and verified later.

[0046] It should be noted that the process by which the dimensional mapping unit constructs a piecewise monotonically invertible mapping based on the scale codebook of the target measurement area in the floor image and standard sample blocks, and retrieves entries using parameter fingerprints, includes:

[0047] Step 1, Calibration Data Acquisition: Under the given scale set and reading aperture, collect wavelet response and millimeter reading paired samples from standard sample blocks or known true value sample areas, and record the pixel spacing, scale and spatial wavelength corresponding sequence number, reference surface model, detrending method identifier, boundary buffer distance, minimum repair aperture, reading aperture, sub-pixel interpolation method, normalized power exponent, cross-scale weight version, acquisition batch and equipment number.

[0048] Step 2, Parameter fingerprint generation: Encode the above fields in a fixed order to form a unique fingerprint, which is used as the primary index key of the mapping entry;

[0049] Step 3, Data Preprocessing: Standardize units and dimensions, configure response normalization according to scale, remove missing and incomplete records, and truncate outliers according to preset quantiles;

[0050] Step 4, Segmentation Strategy Definition: Based on the response distribution or reading caliber, divide the mapping domain into several adjacent intervals, clarify the upper and lower limits and monotonic direction of each interval, and set inter-segment continuity constraints.

[0051] Step 5, Function family fitting: Fit a function that is monotonic and invertible within each interval, fix the function type and parameters, and record the interval domain and extrapolation constraints;

[0052] Step 6, Entry Generation and Archiving: Generate mapping entries for each scale. The entry content includes fingerprint, function type, function parameters, interval domain, unit and precision caliber, extrapolation limit, residual statistics, acquisition batch and device identifier, and version number.

[0053] Step 7, Input and Indexing: Write the mapping entries into the scale codebook, establish a retrieval structure with parameter fingerprints as the main index, and add secondary indexes of scale code, caliber code, timestamp and device number;

[0054] Step 8, runtime retrieval: The processing flow generates parameter fingerprints based on the current data, retrieves matching entries in the scale codebook, and returns the entry number and function parameters if a match is found. When there are no completely matching entries, the adjacent entries are interpolated according to the scale and caliber nearest neighbor relationship of adjacent fingerprints and cross-scale weights to synthesize caliber-bound entries, generate new entry numbers, and cache them in the database.

[0055] Step 9, Version and Range Verification: Perform version consistency verification, parameter range verification, and domain coverage verification on the retrieved entries. If the requirements are not met, trigger a rollback strategy.

[0056] Step 10, Binding and Output: Establish a one-to-one binding relationship between the entry number and the virtual ruler reading caliper, generate a caliper binding index, and write the mapped entry number, function parameter, fingerprint, and version record into the data storage output unit.

[0057] This dimensional mapping process establishes response-millimeters pairings using standard samples, maps entries using parameter fingerprints as unique indexes, preprocesses data using unified dimensions and normalized calibers, solidifies the domain and limits extrapolation using piecewise monotonically invertible functions, uses scale codebooks for database retrieval and nearest neighbor interpolation to fill in missing values, and verifies versions and parameter ranges to prevent mismatches. Furthermore, it binds each entry to a virtual ruler reading caliber and synchronously archives function coefficients and indexes, enabling wavelet responses from different devices, different acquisition batches, and different pixel spacings and preprocessing strategies to be stably converted into millimeter readings under unified scale, caliber, and version management, forming a traceable, verifiable, and reusable metrological alignment baseline.

[0058] It should be noted that the process of the boundary hole processing unit generating a non-near-edge mask for the target region of the floor slab image and repairing holes according to structuring elements includes:

[0059] First, a binary boundary map is generated based on the outer edge of the floor slab projection and the boundary of the internal opening. The distance field to the nearest boundary is calculated. A buffer expansion is applied to the boundary map based on the non-near-edge distance threshold to obtain a forbidden zone. This forbidden zone is then merged with an existing invalid cell mask to form an initial non-near-edge mask. Connectivity analysis is performed on the height map or voxel grid to extract hole regions, recording the hole area, major and minor axes, and equivalent diameter of the circle. Morphological closing operations or hole filling are performed on the initial mask according to preset structuring element types and sizes. Only holes with an equivalent diameter smaller than the minimum repairable hole diameter are repaired; holes larger than this threshold remain forbidden. Repairing connected holes that cross the boundary is prohibited within the near-edge buffer range. The updated non-near-edge mask and hole mask are output, and the structuring element type, structuring element size, distance threshold, minimum repairable hole diameter, connected zone threshold, buffer width, and mask version identifier are recorded.

[0060] Taking a 7m x 5m floor slab as an example (with a 0.8m x 0.5m equipment well in the northwest corner, and several 5 to 9cm pipe holes along the wall and the edge of the opening), the distance field is calculated and buffered with an expansion of 0.3m as the non-near-edge distance threshold to obtain the forbidden zone. Existing invalid pixels are superimposed to form an initial non-near-edge mask. Connectivity analysis is performed on the height map. After recording the hole attributes, morphological closing operation and hole filling are performed with a minimum repairable hole diameter of 150 mm and a structural element disk radius of 10 mm. Only the pipe holes are repaired, and the equipment well and door opening are kept as forbidden zones. At the same time, any cross-edge connected repairs are prohibited within the near-edge buffer range. Processing results: The boundary forbidden zone area is approximately 6.84 square meters (35−6.4×4.4), combined with the retained equipment well area of ​​0.4 square meters, the final total forbidden zone area is approximately 7.24 square meters; the number of fragmented connected domains caused by edge burrs and small holes in the original mask has been reduced from 36 to 12, the number of cross-hole bridging cases has been reduced from 11 to zero, and the shortest distance between the boundary of the candidate area and the hole has been unified to no less than 0.3 meters; the virtual ruler 2-meter window no longer generates cross-hole splicing in the candidate area, and the window endpoint crossover rate has been reduced from 8.5% to 0. The structural element size, distance threshold, minimum repairable hole diameter, and mask version are recorded together for subsequent fitting, scale screening, and reading steps to call according to the same forbidden zone diameter.

[0061] It should be noted that the process of generating a candidate area by outputting a list and number of candidate area images based on the connected component threshold and minimum area includes: inputting a multi-scale response map, validity mask, non-near-edge mask, hole mask, raster index, and connected component threshold and minimum area; performing a bitwise AND operation between the response map and various masks to obtain a candidate binary map, uniformly selecting a four-connectivity aperture; marking the connected components of the candidate binary map to generate a connected component list and a cell index set; calculating the area, number of valid pixels, minimum bounding rectangle, principal direction, and centroid coordinates for each connected component; removing fragments based on the connected component threshold and removing surfaces based on the minimum area threshold. For connected components with insufficient area, remove those with a minimum width less than a preset value; perform morphological or topological merging on adjacent connected components with a gap less than a preset distance, and update statistical information and cell indexes; sort the retained connected components according to the area priority rule, assign region numbers sequentially from the beginning, and establish a mapping table between the numbers and cell indexes and geometric attributes; generate a survey area list, including region number, partition, row and column range, boundary polygon vertex sequence, centroid, area, main direction, and cell index, while recording the connectivity caliber used, connected component threshold, minimum area, and version identifier, and write it to the data storage output unit.

[0062] Taking the upper surface of the same 7m×5m floor slab as an example, after screening out more than 5200 candidate pixels using the wavelet response threshold, 3940 pixels were obtained by ANDing with the validity, non-near edge, and hole mask. The candidate area generation unit used four-connected labeling to obtain 68 connected domains, and the area, centroid, principal direction, and minimum bounding rectangle of each connected domain were calculated. After removing fragments with fewer than 50 pixels according to the connected domain threshold, 33 remained. Then, the thin strips and noisy edges were screened out according to the minimum area threshold of 0.04 square meters and the minimum width threshold of 6 centimeters, leaving 24. Adjacent connected domains with a gap of less than 3 centimeters were merged, and finally 17 test areas were formed, which were numbered from 01 to 17 according to area priority, and their boundary polygons, row and column ranges, and principal directions were output. The fragmented responses and narrow pseudo-regions along the edge hole are uniformly filtered, and the responses that are adjacent and physically continuous are grouped into stable measurement areas. This ensures that the subsequent virtual ruler sliding window is only executed within the numbered list, and the window will not cross into the prohibited measurement area or the narrow pseudo-region. The coordinates and readings of the measurement points can correspond one-to-one with the measurement area number and be verified according to the version record.

[0063] It should be noted that the virtual ruler measurement unit, based on the detrended height map within the floor slab image waiting area, performs omnidirectional sliding according to the measurement window set in the configuration table, with directional steps not exceeding five degrees and displacement steps not exceeding twenty centimeters. The window profile signal is extracted sequentially from the grid index and interpolated into subpixels using cubic splines. Extreme value positioning adopts a dual judgment method of neighborhood three-point fitting and spline extreme value verification. When the endpoint or interior falls into invalid pixels, holes, or non-near-edge masks, judgment is made based on the continuous invalid segment length threshold and the total invalid percentage threshold. When the continuous invalid segment length is not greater than 20cm and the total invalid percentage is not greater than 20%, interpolation is used to fill the gap. If either threshold is exceeded, the endpoint is retreated by a fixed step size, and the number of repositioning times is recorded. When crossing holes, the largest segment profile is used as the reading basis and cross-hole splicing is prohibited. When the reading is close to the boundary and causes an overtravel, the window is clipped in that direction with priority given to the mask that is not close to the edge. The correspondence between the window and the scale set is determined by the aperture binding index given by the wavelet scale management unit. When the reading aperture is between adjacent scales, the responses of adjacent scales are interpolated according to the cross-scale weight and bound to a single aperture index. The generated readings, window center coordinates, orientation angle, endpoint coordinates, spline kernel radius, orientation step and displacement step, invalid segment statistics, interpolation participation identifier, relocation record, reading retention digits and rounding rules are written to the data storage output unit and a one-to-one correspondence is established with the mapping entry number returned by the dimensional mapping unit.

[0064] Taking the 17 measurement areas obtained after partitioning and rasterization on the upper surface of a 7m×5m floor slab as an example, the virtual straightedge measurement unit slides omnidirectionally over a 2m window in each measurement area with a directional step of no more than 5 degrees and a displacement step of no more than 20 cm, generating more than 3,200 candidate windows. For windows whose endpoints or interiors fall into invalid pixels, holes, or non-near-edge masks, the decision to use cubic spline interpolation is made based on the aperture where the continuous invalid segment does not exceed 20 cm and the total invalid percentage does not exceed 20%. If any threshold is exceeded, the endpoint is retreated by a fixed step size or the direction is abandoned, and the number of repositionings is recorded. Windows that cross equipment wells are prohibited from being spliced ​​across holes, and only the largest segment profile is used as the reading basis. When close to the edge of the slab, priority is given to cutting based on the non-near-edge mask. The readings in that direction are cut; the correspondence between the window and the scale set is determined by the caliber binding index. When the reading caliber is between adjacent scales, the response of the adjacent scales is interpolated according to the cross-scale weight and uniformly bound to a single caliber index; finally, about 3,000 valid readings are generated, and for each reading, the window center coordinates, orientation angle, endpoint coordinates, spline kernel radius, step parameters, invalid segment statistics, interpolation and relocation identifiers, and rounding rules are recorded synchronously; this example shows that invalid data in holes, near edges, and sparse areas will not be introduced into the readings by crossing holes or going out of bounds. The consistency between direction and caliber is fixed by binding index and recording fields, so that the dispersion of readings remeasured in different directions within the same survey area is controlled under a unified caliber and has traceability.

[0065] like Figure 3 As shown, after detrending and zonal raster constraints were performed on the height field under unified coordinates and aperture, the overall tilt and large-scale bulges were stripped away into fine-scale undulations with near-zero mean, and the color distribution converged and became continuous. The 30-centimeter forbidden zone at the boundary and the equipment well area did not participate in the aggregation, and the pseudo-height differences near the wall and openings were isolated. After using incremental pixel spacing for the transition zone and the central zone, the edge details were more densely expressed while the center remained smooth, and there were no visible breaks in the cross-block connections, indicating that the indexing and detrending were consistent. At the same time, the weights internalized by the quality record suppressed edge strip noise, the connectivity of the waiting area was good, and the color scales mainly fell within ± millimeters, indicating that noise transmission was controlled. Thus, only the geometric shape that can correspond one-to-one with the field straightedge aperture was retained, laying the foundation for the comparability, verifiability, and traceability of subsequent virtual straightedge readings.

[0066] like Figure 4 and Figure 5 As shown, the improved results stem from the technical means of this invention, which involves the coordinated application of a boundary hole processing unit, a candidate area generation unit, a wavelet scale management unit, a virtual ruler measurement unit, and a dimensional mapping unit under the same quantization caliber: First, small holes with equivalent diameters smaller than the threshold are buffered and repaired according to the structural element to form a unified forbidden mask; then, stable measurement areas are screened out using connected domains and minimum area calibers; subsequently, the effective scale set of the two-meter window is limited according to the scale-spatial wavelength correspondence table and the caliber binding index; finally, the range value is obtained in the omnidirectional sliding window using cubic spline sub-pixel interpolation, dual threshold judgment of continuous invalid segments and total proportion, endpoint backtracking, and cross-hole prohibition rules, and mapped to millimeter readings by parameter fingerprint retrieval, thereby blocking cross-hole splicing and near-edge overtravel, reducing strip artifacts and isolated abnormal patches, increasing the effective window ratio, and solidifying the reading caliber and version, so that the window range map along the horizontal direction of the room changes from discontinuous and noisy to a continuous, consistent, and traceable result distribution.

[0067] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0068] In conclusion, the above description is only 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 floor slab engineering acceptance system based on image acquisition, characterized in that, The system includes an imaging acquisition unit, a 3D reconstruction and calibration unit, a reference surface fitting and detrending unit, a height map rasterization unit, a wavelet scaling unit, a dimensional mapping unit, a boundary hole processing unit, a candidate area generation unit, a virtual straightedge measurement unit, and a data storage and output unit. The imaging acquisition unit acquires multi-view floor images and simultaneously delivers camera position and time information to the 3D reconstruction and calibration unit. The 3D reconstruction and calibration unit performs intrinsic and extrinsic parameter calibration, distortion correction, attitude calculation, and dense reconstruction of the floor images, outputting a unified coordinate point cloud and calibration quality record. The reference surface fitting and detrending unit fits the image reference surface within the target area to generate a deviation field, which is then used by the height map rasterization unit to generate a height map according to the pixel spacing. The raster indexing and wavelet scaling management unit calculates multi-scale response and establishes a scale-to-spatial wavelength correspondence table within a preset scale set. The dimensional mapping unit constructs a piecewise monotonically reversible mapping based on the scale codebook and standard sample blocks of the target area of ​​the floor image and retrieves entries using parameter fingerprints. The boundary hole processing unit generates a non-near-edge mask for the target area of ​​the floor image and repairs holes according to structural elements. The candidate area generation unit outputs a list of test area images and numbers according to the connected component threshold and minimum area. The virtual ruler measurement unit performs image sub-pixel extreme value interpolation within the selected scale and reading aperture and generates readings. The data storage and output unit archives the parameters of each unit, the acquisition batch identifier, the coordinates and readings of the measurement points, and the transformation version. In the imaging acquisition unit, the target area is divided into a boundary zone, a transition zone, and a central zone based on the horizontal distance L from the nearest boundary, with the outer edge of the horizontal projection of the floor image and the boundary zone of the internal void boundary as reference. The L values ​​of the boundary zone, transition zone, and central zone are less than or equal to 30cm, greater than 30cm but less than or equal to 80cm, and greater than 80cm, respectively. The number of sampling frames is calculated from the partition reference value and adjusted item by item according to the discrete conditions of angle coverage, baseline coverage, matching point density, reprojection residual, motion blur, brightness deviation, occlusion rate, and near-edge safety margin. The upper and lower limits of the boundary zone, transition zone, and central zone are set. Under the room-level frame number budget constraint, the frame is rounded up step by step, and the camera position index is generated based on the results. The 3D reconstruction calibration unit aligns the multi-view images output by the imaging acquisition unit according to the frame time identifier and camera position index. It establishes an intrinsic parameter version record according to the shooting batch and lens status and writes the focal length, principal point position, pixel size, radial distortion coefficient, and pixel non-orthogonality term. It solves the station attitude based on the circumferential and radial baseline positions and the line-of-sight angle sequence as geometric priors and generates a camera attitude record. It performs image distortion correction and dense reconstruction to obtain a unified coordinate system point cloud and generates a calibration quality record associated with the station number. The calibration quality record includes the reprojection residual quantile value, field of view overlap ratio, baseline position coverage identifier, and angle coverage identifier. At the same time, it exports the parameter fingerprint field and writes it to the data storage output unit. The virtual ruler measurement unit performs omnidirectional sliding within the floor image waiting area based on the detrended height map, according to the measurement window set in the configuration table, with a directional step of no more than five degrees and a displacement step of no more than twenty centimeters. The window profile signal is extracted sequentially by the grid index and sub-pixel interpolation is performed using cubic splines. The extreme value positioning adopts a dual judgment method of neighborhood three-point fitting and spline extreme value verification. When the endpoint or interior falls into invalid pixels, holes, or non-near-edge masks, the judgment is made according to the continuous invalid segment length threshold and the total invalid percentage threshold. When the length of consecutive invalid segments is no greater than 20cm and the total invalid percentage is no greater than 20%, interpolation is used to fill the gaps. When any threshold is exceeded, the endpoint is backed up by a fixed step size and the number of repositionings is recorded. When the window crosses a hole, the maximum segment profile is used as the reading basis and cross-hole splicing is prohibited. When the window is close to the boundary and causes an overtravel, the window is clipped in that direction with priority given to the mask that is not close to the edge. The correspondence between the window and the scale set is determined by the aperture binding index given by the wavelet scale management unit. When the reading aperture is between adjacent scales, the responses of adjacent scales are interpolated according to the cross-scale weight and bound to a single aperture index. The generated readings, window center coordinates, direction angle, endpoint coordinates, spline kernel radius, direction step and displacement step, invalid segment statistics, interpolation participation identifier, repositioning record, reading retention digits and rounding rules are written into the data storage output unit and a one-to-one correspondence is established with the mapping entry number returned by the dimensional mapping unit.

2. The floor slab engineering acceptance system based on image acquisition according to claim 1, characterized in that, Under the constraints of unified coordinate point cloud, camera position, baseline and angle coverage markers, boundary and hole masks, pixel spacing, partitioning rules, and height map raster index, the reference plane fitting detrending unit generates fitting blocks and numbers them according to the boundary zone, transition zone, and center zone of the floor target area image with preset block size and step. Candidate blocks are screened out according to the hole ratio and near-edge buffer, and the baseline and angle coverage are checked. If they are not satisfied, they are adjusted or supplemented in the same partition according to the step. The in-point threshold is set according to the reprojection residual quantile value of the calibration quality record. The point cloud in the block is weighted by combining the view overlap and baseline coverage, and the weight is attenuated according to the near-edge buffer distance. The fitting is completed, and the detrending bias field is output.

3. The floor slab engineering acceptance system based on image acquisition according to claim 1, characterized in that, The heightmap rasterization unit establishes a multi-layer raster with differentiated partitioning of the target region of the floor slab image under the constraints of unified coordinate point cloud, de-trending bias field, partition labels, boundary mask, hole mask, near-edge buffer distance, block size, and step size. The pixel spacing of the base layer is set according to partitions: 10 mm for the boundary zone, 20 mm for the transition zone, and 30 mm for the center zone. The correspondence between partitions and pixel spacing is recorded. The grid origin is taken as the coordinates of the lower left corner of the smallest bounding rectangle of the target region of the floor slab image, rounded down according to the pixel spacing. Row and column indices are generated for each layer in row priority order and a one-to-one correspondence is established with the fitted block number and partition category. Pixel aggregation adopts a median-first strategy. Furthermore, when the number of valid observations is less than five, the truncated mean is used and a no-value mark is output. Simultaneously, a validity mask, boundary mask, hole mask, and non-near-edge mask are generated. For each image cell, the number of observations, the number of outliers, and the aggregation window radius are recorded to form a raster index table of the floor image and written into the parameter fingerprint field. The raster index table includes row and column range, layer number, cell spacing, origin coordinates, and index mode. The cell spacing setting, number of layers, origin coordinates, aggregation statistics method, valid observation threshold, mask type and version, and index mode are written into the parameter fingerprint field and stored in the data storage output unit.

4. A floor slab engineering acceptance system based on image acquisition according to any one of claims 2 or 3, characterized in that, The wavelet scaling unit generates a scale set and a scale index under constraints such as the height map of the target region in the floor image, pixel spacing, partition labels, fitted block list, boundary mask, hole mask, near-edge buffer distance, and reading aperture. The scale set consists of local and global scale sequences, discretized geometrically with a scaling factor between 1:1.25 and 1:

2. The spatial wavelength ranges of the local and global scales are 300 mm to 600 mm and 1000 mm to 3000 mm, respectively. Fixed coefficients are used to establish the scale and spatial... A wavelength correspondence table is used to filter effective scales for each fitted block based on the length of its short side and the distance to the near-side buffer. If the radius of the transform kernel corresponding to a scale exceeds 40% of the short side length of the block or crosses the near-side buffer, the scale is disabled in that block and its identifier is recorded. Normalization factors and cross-scale weights are configured for each scale. The normalization factor is set according to the scale power, and the power exponent is given by the parameter rule library. The cross-scale weights are set according to the logarithmic interval between scales. When the reading aperture is between adjacent scales, the responses of adjacent scales are interpolated according to the weights, and an aperture-bound scale index is generated. The wavelet scale management unit outputs a scale-to-spatial wavelength correspondence table, an effective scale mask, a scale normalization configuration, a cross-scale weight table, and an aperture binding index, and writes them to the data storage output unit for subsequent unit calls.

5. The floor slab engineering acceptance system based on image acquisition according to claim 1, characterized in that, The process of constructing a piecewise monotonically invertible mapping based on the scale codebook and standard sample blocks of the target measurement area in the floor image and retrieving entries using parametric fingerprints includes: Step 1, Calibration Data Acquisition: Under the given scale set and reading aperture, collect wavelet response and millimeter reading paired samples from standard sample blocks or known true value sample areas, and record the pixel spacing, scale and spatial wavelength corresponding sequence number, reference surface model, detrending method identifier, boundary buffer distance, minimum repair aperture, reading aperture, sub-pixel interpolation method, normalized power exponent, cross-scale weight version, acquisition batch and equipment number. Step 2, Parameter fingerprint generation: Encode the fields in a fixed order to form a unique fingerprint, which is used as the primary index key for the mapping entry; Step 3, Data Preprocessing: Standardize units and dimensions, configure response normalization according to scale, remove missing and incomplete records, and truncate outliers according to preset quantiles; Step 4, Segmentation Strategy Definition: Based on the response distribution or reading caliber, divide the mapping domain into several adjacent intervals, clarify the upper and lower limits and monotonic direction of each interval, and set inter-segment continuity constraints. Step 5, Function family fitting: Fit a function that is monotonic and invertible within each interval, fix the function type and parameters, and record the interval domain and extrapolation constraints; Step 6, Entry Generation and Archiving: Generate mapping entries for each scale. The entry content includes fingerprint, function type, function parameters, interval domain, unit and precision caliber, extrapolation limit, residual statistics, acquisition batch and device identifier, and version number. Step 7, Input and Indexing: Write the mapping entries into the scale codebook, establish a retrieval structure with parameter fingerprints as the main index, and add secondary indexes of scale code, caliber code, timestamp and device number; Step 8, runtime retrieval: The processing flow generates parameter fingerprints based on the current data, retrieves matching entries in the scale codebook, and returns the entry number and function parameters if a match is found. When there are no completely matching entries, the adjacent entries are interpolated according to the scale and caliber nearest neighbor relationship of adjacent fingerprints and cross-scale weights to synthesize caliber-bound entries, generate new entry numbers, and cache them in the database. Step 9, Version and Range Verification: Perform version consistency verification, parameter range verification, and domain coverage verification on the retrieved entries. If the requirements are not met, trigger a rollback strategy. Step 10, Binding and Output: Establish a one-to-one binding relationship between the entry number and the virtual ruler reading caliper, generate a caliper binding index, and write the mapped entry number, function parameter, fingerprint, and version record into the data storage output unit.

6. The floor slab engineering acceptance system based on image acquisition according to claim 1, characterized in that, The process of the boundary hole processing unit generating a non-near-edge mask for the target region of the floor slab image and repairing holes according to structuring elements includes: First, a binary boundary map is generated based on the outer edge of the floor slab projection and the boundary of the internal opening. The distance field to the nearest boundary is calculated. A buffer expansion is applied to the boundary map based on the non-near-edge distance threshold to obtain a forbidden zone. This forbidden zone is then merged with an existing invalid cell mask to form an initial non-near-edge mask. Connectivity analysis is performed on the height map or voxel grid to extract hole regions, recording the hole area, major and minor axes, and equivalent diameter of the circle. Morphological closing operations or hole filling are performed on the initial mask according to preset structuring element types and sizes. Only holes with an equivalent diameter smaller than the minimum repairable hole diameter are repaired; holes larger than this threshold remain forbidden. Repairing connected holes that cross the boundary is prohibited within the near-edge buffer range. The updated non-near-edge mask and hole mask are output, and the structuring element type, structuring element size, distance threshold, minimum repairable hole diameter, connected zone threshold, buffer width, and mask version identifier are recorded.

7. The floor slab engineering acceptance system based on image acquisition according to claim 1, characterized in that, The process of generating a candidate area image list and numbering unit based on connected component thresholds and minimum area includes: inputting a multi-scale response map, validity mask, non-near-edge mask, hole mask, raster index, connected component threshold, and minimum area; performing a bitwise AND operation between the response map and various masks to obtain a candidate binary map, uniformly selecting a four-connected connectivity aperture; marking connected components in the candidate binary map to generate a connected component list and pixel index set; calculating the area, number of valid pixels, minimum bounding rectangle, principal direction, and centroid coordinates for each connected component; removing fragments based on the connected component threshold and removing insufficient area based on the minimum area threshold. Connected components are processed by removing those with a minimum width less than a preset value. Morphological or topological merging is performed on adjacent connected components with a gap less than a preset distance, and statistical information and cell indexes are updated. The remaining connected components are sorted according to area priority rules, assigned region numbers sequentially from the beginning, and a mapping table is established between the numbers and cell indexes and geometric attributes. A survey area list is generated, containing the region number, its partition, row and column range, boundary polygon vertex sequence, centroid, area, principal direction, and cell index. Simultaneously, the connectivity standard used, connected component threshold, minimum area, and version identifier are recorded and written to the data storage output unit.

Images