A dual-mode height measurement method and device based on partition weighted centroid multi-distance calibration

Abstract

The application discloses a dual-mode height measurement method and device based on partition weighted centroid multi-distance calibration. The phase distance measurement is susceptible to the influence of reflected signal attenuation and signal-to-noise ratio change in a specific distance interval, and the triangulation distance measurement is affected by imaging range, focusing and spot shape distortion under different distances and imaging conditions, resulting in unstable centroid. The application first uses phase distance measurement to obtain distance coarse measurement value and mode determination; in the triangulation distance branch, the spot image is grayed and normalized, multi-threshold segmented and the effective area is screened according to compactness, and after morphological optimization, the partition weighted centroid is used to obtain the centroid offset; based on the distance coarse measurement value, the height inversion model is selected or interpolated in the multi-distance calibration model library to obtain the triangulation height value; in the transition zone, the phase height value and the triangulation height value are weighted and fused to output. The application can maintain the continuity and stability of the height output under different distance measurement conditions, and improve the robustness and availability of long-distance height measurement.

Description

Technical Field

[0001] This invention relates to the field of engineering surveying and laser ranging technology, and in particular to a dual-mode fusion elevation measurement method for finished concrete surfaces, as well as a data processing device and storage medium for implementing the method. Background Technology

[0002] The quality of paving, leveling, and finishing during concrete construction directly affects the structural performance and subsequent processes. On-site, rapid elevation and surface shape checks are often required for large finished surfaces to assess flatness and consistency of construction quality. Existing engineering surveying equipment can achieve prism-free measurement using phase ranging, simultaneously acquiring geometric information such as angles and distances; laser triangulation offers high resolution in non-contact displacement measurement. However, in large spaces and under complex reflection conditions, single ranging methods are prone to accuracy degradation or failure within specific distance ranges. Furthermore, triangulation often relies on stable extraction of the spot centroid and calibration models for different distance conditions; otherwise, it is susceptible to optical distortion, noise, and changes in spot morphology, leading to unstable height inversion. Therefore, a dual-mode fusion elevation measurement scheme capable of stable output under different ranging conditions is needed.

[0003] In concrete pouring and leveling operations, elevation control is often characterized by "large working area, short time window, and high density of measuring points": on the one hand, it is necessary to quickly obtain the elevation distribution of multiple measuring points within a large working radius; on the other hand, it is desirable that the measurement results can be used for quality assessment or closed-loop control of construction equipment. Traditional point measurement methods, which mainly rely on manual aiming, are usually post-event inspections and are limited by personnel operation and sampling density, making it difficult to meet the requirements for real-time and high-density measurement.

[0004] In recent years, non-contact measurement solutions for construction sites have gradually adopted approaches such as laser scanning or multi-sensor ranging, achieving rapid coverage of target areas through mechanical scanning or array deployment. However, construction sites present factors such as varying lighting conditions, dust and water films, uneven surface reflectivity, and mechanical vibration. These factors can cause ranging signal attenuation, short-term loss of lock, or output drift, thereby affecting the stability and consistency of elevation measurements.

[0005] For imaging measurements based on triangulation, the gray-scale distribution of the light spot on the detector is affected by aberrations, uneven surface reflection, and local pits / defects, often resulting in blurred edges, asymmetry, or localized abnormal bright spots. If traditional algorithms such as single gray-scale weighted centroid or Gaussian fitting are directly used, the centroid is easily shifted due to the pull of local abnormal gray levels, thus introducing elevation inversion errors.

[0006] Furthermore, the mapping relationship between centroid offset and target height often varies with the measurement distance and exhibits nonlinear characteristics. If only a calibration model for a single distance is used, it is difficult to maintain inversion accuracy across the entire measurement range; therefore, it is necessary to establish a calibration model system that spans distances, and select or interpolate the corresponding model based on the real-time distance during the measurement process.

[0007] In summary, there is an urgent need for a dual-mode fusion elevation measurement method and device that uses phase ranging to provide coarse distance measurement and drive working mode selection, combines robust spot effective area extraction, partitioned weighted centroid and multi-distance calibration model inversion in the triangulation ranging branch, and achieves continuous output in the transition zone. Summary of the Invention

[0008] The present invention aims to solve at least the following problems: (1) the stability of elevation output decreases due to the influence of changes in the intensity of reflected signals and signal-to-noise ratio in phase ranging within a specific distance range; (2) the measurement failure may occur due to abnormal imaging range or spot shape under different distances or specific imaging conditions in triangulation ranging; (3) the sensitivity of spot centroid extraction to noise and the nonlinearity of height mapping relationship with distance in triangulation elevation inversion.

[0009] To address the aforementioned technical problems, this invention provides a dual-mode elevation measurement method using partitioned weighted centroid multi-distance calibration, comprising: acquiring phase distance measurement L as a coarse distance measurement value and determining the measurement mode; acquiring a spot image in triangulation mode, extracting the effective region, and obtaining the centroid offset using a partitioned weighted centroid algorithm; constructing a multi-distance calibration model library, and selecting or interpolating the height inversion model based on the coarse distance measurement value during measurement, converting the centroid offset into elevation; and performing weighted fusion of the phase elevation and triangulation elevation in transition fusion mode for output. This invention also provides a data processing device for implementing the above method and a computer-readable storage medium storing the program.

[0010] In one specific implementation, the mode determination uses a distance threshold L0 and a transition zone half-width ΔL to form a mode switching transition zone and suppress frequent jitter switching caused by distance estimation noise: when the coarse distance measurement value L is in the effective range of phase ranging, it enters the phase ranging mode; when L is in the effective range of triangular ranging, it enters the triangular ranging mode; when L is in the transition zone near the threshold, it enters the transition fusion mode.

[0011] In one specific implementation, the spot image undergoes grayscale and normalization processing before entering threshold segmentation: the color image is converted into a single-channel grayscale image, and the grayscale values ​​are linearly mapped to the [0,1] interval to reduce the grayscale range difference caused by different lighting conditions and exposure settings, thereby improving the robustness of subsequent segmentation.

[0012] In one specific implementation, the effective region extraction adopts a combination of multi-threshold traversal and compactness evaluation: multiple thresholds are used to perform binary segmentation on the same normalized grayscale image to obtain multiple candidate regions, and the area A and perimeter P of the candidate regions are calculated. The candidate regions are evaluated according to the compactness index C=4πA / P², and the candidate regions with a compactness closer to 1 and an area that meets the range constraint are selected as the effective regions.

[0013] In one specific implementation, the effective region is further subjected to morphological optimization processing, including operations such as hole filling, removal of small connected components, erosion or dilation, to form a more compact light spot region and reduce the impact of edge blurring and residual noise on centroid calculation; when the number of effective pixels after morphological processing is lower than a preset lower limit, the segmentation result before morphological processing can be retained as a fallback to ensure the feasibility of subsequent centroid calculation.

[0014] In one specific implementation, the partition-weighted centroid algorithm first determines the bounding box range of the effective area of ​​the light spot [x]. start ,x end ]×[y start ,y end The effective area is divided into multiple sub-regions along a preset direction (preferably horizontal); the local centroid of each sub-region is obtained by weighted summation of pixel coordinates based on grayscale values, and the total grayscale value within the sub-region is used as the weight w. i Then, the global centroid is obtained by weighting the local centroids of all valid sub-regions. If no valid pixels are found in any sub-region, the gray-scale weighted centroid of the entire region is used as a fallback solution.

[0015] In one specific implementation, the centroid offset Δv is the difference between the coordinates of the global centroid in the sensitive direction and the reference coordinates. The reference coordinates can be the centroid coordinates corresponding to zero height during calibration, the preset center coordinates of the detector coordinate system, or the reference coordinates determined by system initialization.

[0016] In one specific implementation, the multi-distance calibration model library uses multiple calibration distances L i The data (Δv, h) was collected and model-fitted; for each measured distance L... i Establish a highly inverted model h=f i (Δv) and store the corresponding model parameters; the model can be in the form of a high-order polynomial, a piecewise polynomial or a lookup table, where the order of the polynomial can be selected according to the fitting residuals and stability.

[0017] In one specific implementation, the online model selection or interpolation includes: when the coarse distance measurement L is close to a certain calibration distance L... i When the difference is less than the distance tolerance, directly select model f. iWhen L is located between adjacent calibration distances, the parameters of the adjacent models are interpolated to obtain the inversion model corresponding to the current distance, thereby improving the consistency of elevation inversion when the distance changes continuously.

[0018] In one specific implementation, the fusion weight w can vary with the position of the coarse distance measurement value L within the transition interval, preferably linearly. In addition, the fusion weight can also be corrected by combining the phase ranging output quality and the spot image quality. The image quality may include at least one of compactness, effective area, total gray level, and saturation pixel ratio.

[0019] In one specific implementation, the method further includes a vibration compensation process based on acceleration feedback: synchronously acquiring the acceleration signal of the measurement system, performing spectral analysis on the acceleration signal to identify the dominant vibration frequency or target frequency band, and constructing a digital notch filter based on the dominant vibration frequency or target frequency band; using the digital notch filter to filter and compensate for at least one of the centroid offset Δv, the second elevation value h2, or the fused elevation value h, so as to suppress the measurement error caused by vibration and improve the measurement stability under dynamic working conditions.

[0020] In one specific implementation, the dual-mode elevation measurement process is controlled by a state machine, and the phase ranging result and the spot image quality are evaluated in each measurement cycle. The spot image quality may include at least one of compactness, effective area, total grayscale, and saturated pixel ratio. When the evaluation result meets abnormal conditions, an anomaly handling strategy is triggered to maintain the continuity and stability of the measurement results.

[0021] In one specific implementation, the anomaly handling strategy includes at least one of the following: using the full image gray-scale weighted centroid as the fallback centroid, re-acquiring the spot image, correcting the fusion weights to reduce the contribution of the second elevation value h2 in the fusion result, or switching to phase ranging mode output.

[0022] like Figure 6 As shown, the present invention also provides a dual-mode elevation measurement data processing device, which includes a distance acquisition and mode determination module, a phase elevation acquisition module, an image acquisition module, an effective area extraction module, a partitioned weighted centroid calculation module, a calibration model selection and inversion module, and a fusion output module; the above modules can be implemented by a software program executed by a processor, or by a programmable logic device or a dedicated hardware accelerator.

[0023] Compared with existing single ranging methods, the present invention has at least the following beneficial effects:

[0024] (1) By using a mode determination mechanism driven by coarse distance measurements, a more suitable measurement branch is adopted within the effective distance range of phase ranging and triangulation ranging, and transition fusion output is adopted near the threshold, thereby reducing the risk of output discontinuity caused by mode switching.

[0025] (2) By using multi-threshold traversal, compactness evaluation and morphological optimization, an effective region closer to the real light spot is obtained, reducing the impact of over-segmentation, under-segmentation and background noise on centroid calculation.

[0026] (3) By dividing the effective area of ​​the light spot into multiple sub-regions and calculating the local centroids separately through the partition weighted centroid algorithm, the influence of local abnormal gray levels, edge bright spots or uneven reflection on the global centroid can be weakened, and the positioning robustness under complex light spot conditions can be improved.

[0027] (4) By constructing a multi-distance calibration model library and selecting or interpolating model parameters according to distance in online measurement, it is possible to adapt to the nonlinear characteristics of the centroid offset-height mapping as a function of distance, thereby improving the consistency of elevation inversion across the entire range.

[0028] (5) By performing quality assessment, anomaly handling, and optional filtering and vibration suppression within the state machine framework, the controllability and engineering applicability of the continuous measurement process can be improved. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall process of the dual-mode elevation measurement method of the present invention.

[0030] Figure 2 This is a schematic diagram of the preprocessing and effective region extraction process for light spot images.

[0031] Figure 3 A schematic diagram of the principle of calculating the weighted centroid of a partition and a schematic diagram of the partition boundary.

[0032] Figure 4 A schematic diagram illustrating the process of constructing a multi-distance calibration model library, model selection, and coefficient interpolation inversion.

[0033] Figure 5 This is a schematic diagram of the state machine for measurement mode determination, transition fusion, and anomaly handling.

[0034] Figure 6 The structural block diagram of the dual-mode elevation measurement data processing device for implementing the method of the present invention is shown.

[0035] Figure 7 The fitted curve is for L=5m.

[0036] Figure 8 The value is the fitting residual when L=5m.

[0037] Figure 9The results show the accuracy verification for L=5, 10, 15, and 20m.

[0038] Figure 10 The image is the result of grayscale conversion and normalization preprocessing.

[0039] Figure 11 The results are from image thresholding and morphological processing.

[0040] Figure 12 This is a comparison of the results of traditional two-dimensional Gaussian fitting.

[0041] Figure 13 The result of the partition-weighted centroid method is shown in the figure.

[0042] Figure 14 Before and after images are shown to compensate for the previous results.

[0043] Figure 15 Frequency analysis graphs before and after DisA data compensation. Detailed Implementation

[0044] Example 1 (Basic Example):

[0045] like Figure 1 As shown in the figure, this embodiment presents a dual-mode elevation measurement process with measurement points as processing units. In each measurement cycle, the system first acquires phase ranging as a coarse distance measurement value, and selects the phase ranging mode, triangulation ranging mode, or transition fusion mode accordingly. When entering the triangulation ranging-related branch, the centroid offset is obtained through spot image processing, and then the elevation value is obtained by inversion using a multi-distance calibration model library. The fusion result is then output in the transition zone.

[0046] (1) Input / output and data structure: In this embodiment, the measurement point is used as the processing unit. Each measurement point corresponds to at least a set of coarse distance measurement values ​​L, a set of phase elevation values ​​h1, and one or more frames of measurement point spot images I. The spot image I can be a grayscale image or a color image, which can be converted into a double-precision grayscale matrix before processing; the output is the measurement point elevation result h, where h can be h1 in phase mode, h2 in triangular mode, or the weighted fusion result in transition fusion mode.

[0047] (2) Preparation of multi-distance calibration model library: Select several calibration distances L i The coverage system's working range, for example, can cover 5m to 30m and sample at 0.5m or 1m step intervals; at each calibration distance L i Set multiple known heights h below j (Can be formed by standard blocks or lifting platforms), acquire light spot images at each height and obtain the centroid offset Δv according to the centroid algorithm described later. ij With (Δv) ij ,h jThe height inversion model h=f is obtained by fitting the sample at this distance. i (Δv), and correlate the model parameters with the corresponding L. i Associative storage forms a multi-distance calibration model library.

[0048] (3) Online measurement process:

[0049] S101. Obtain the phase ranging L;

[0050] S102. Determine the mode based on L, the preset threshold L0, and the half-width ΔL of the transition zone;

[0051] S103. When the phase ranging mode is determined, read or calculate the first elevation value h1 and output it;

[0052] S104. When the triangulation mode or transition fusion mode is determined, acquire the spot image I and complete the effective region extraction;

[0053] S105. Calculate the partitioned weighted centroid coordinates (x) based on the effective region. c ,y c ), and based on this, the centroid offset Δv=x is obtained. c -x ref , where x ref Preset reference coordinates;

[0054] S106. Based on L, select or interpolate from the multi-distance calibration model library to obtain the height inversion model, and calculate the second elevation value h2;

[0055] S107. When in triangulation mode, output h2. When in transition fusion mode, calculate and output the fused elevation according to h=w·h1+(1−w)·h2.

[0056] (4) Anomaly and quality control: When the image quality does not meet the requirements (e.g., the effective area is too small, the compactness is too low, the number of effective pixels after morphological processing is too small), it can be determined that the triangulation result is unreliable. In this case, it can trigger the re-acquisition of the spot image, reduce the weight of the triangulation result in the fusion, or directly switch to the phase ranging mode output to ensure the continuous availability of the elevation result.

[0057] The process for the multi-distance calibration model library is as follows: Taking L=5m as an example, the 5th-order polynomial fitting results are shown in Table 1. The fitting curve is as follows. Figure 7 As shown, the R² of the model with direction h=f(x) is 0.99478 and RMSE is 0.1445mm, indicating that the model is in high agreement with the experimental data.

[0058] Table 1. Parameters for 5th-order polynomial fitting at distance L=5m

[0059]

[0060] After completing a fifth-order polynomial fit, a systematic analysis of the fitting residuals at a distance of L=5m was performed to further evaluate the model accuracy. The residual value for each fitting point was calculated in Matlab, and the results are as follows: Figure 8 As shown.

[0061] The statistical properties of the residuals are shown in the table below:

[0062] Table 2 Statistical characteristics of residuals when L=5m

[0063]

[0064] The mean residual of −0.0032 mm is close to zero, indicating that the model has no systematic bias. The standard deviation of 0.1516 mm reflects the measurement accuracy, and the 3σ interval is ±0.455 mm, meaning that 99.73% of the measurements fall within this range.

[0065] Table 3 below shows the calibration results for the entire distance;

[0066] Table 3. Statistical analysis of calibration results for the entire distance (5-30m)

[0067]

[0068] Across all 28 distance points, the system exhibited excellent goodness of fit and sub-millimeter accuracy. The overall mean R² value was 0.9876, with a standard deviation of 0.0328. For 23 distance points (82.1%), R² > 0.99, indicating that the 5th-order polynomial model has good adaptability across the entire measurement range. The mean RMSE was 0.685 mm, with a standard deviation of 0.520 mm, demonstrating that the system maintains sub-millimeter measurement accuracy at most distances.

[0069] To comprehensively evaluate the generalization ability and actual measurement accuracy of the established fifth-order polynomial calibration model h=f(x), a systematic accuracy verification experiment was conducted at four key working distances (L=5, 10, 15, 20m). Five independent test points were collected at each distance to form a verification set; these points were not used in the calibration. The verification results are summarized in […]. Figure 9 .

[0070] from Figure 9As can be seen, under various distance measurement conditions, the predicted values ​​and the actual values ​​exhibit a good linear relationship, and most data points are closely distributed near the ideal straight line (y=x), indicating that the system has high measurement consistency. Further analysis of the RMSE index shows that: at L=5m, the RMSE is 0.2928mm; at L=10m, it is 0.1285mm; at L=15m, it is 0.5866mm; and at L=20m, it is 0.4976mm. Overall, the system achieves sub-millimeter accuracy (<0.3mm) in the 5–10m range and maintains a measurement accuracy better than 0.6mm in the 15–20m range, meeting the requirements for high-precision elevation measurement.

[0071] In summary, the system's calibration model passed rigorous experimental verification within the range of 5–20m. In a series of multi-distance and multi-height calibration experiments, the goodness of fit of the nonlinear mapping model between the spot centroid offset and the target height reached R²>0.99, and the root mean square error RMSE<0.5mm. Furthermore, through digital notch filtering compensation based on acceleration feedback, the measurement error caused by mid-frequency vibration was reduced by more than 67%, thereby improving the reliability under dynamic operating conditions.

[0072] Example 2 (Preferred Example):

[0073] Based on Example 1, this example provides more detailed methods for extracting the effective area of ​​the light spot, calculating the centroid of the partition, interpolating the multi-distance model, and fusing the transition zone, so as to improve the robustness of complex light spot morphology, cross-distance working conditions and on-site disturbance conditions.

[0074] (1) Extraction of effective regions (see) Figure 2 S201 Image grayscale conversion: Converts the colored spot image into a single-channel grayscale image; S202 Grayscale normalization: Linearly maps pixel grayscale values ​​to the [0,1] interval, for example, I norm (x,y)=(I(x,y)−I min ) / (I max -I min (where I) min =min(I),I max =max(I)); S203 Multi-threshold traversal segmentation: Perform binary segmentation on the threshold set T to obtain candidate spot regions. The threshold set can be selected as {0.3,0.4,0.5,0.6}; S204 Compactness evaluation: Calculate the area A and perimeter P for each candidate region, and evaluate the segmentation quality according to compactness C=4πA / P². Select the candidate region with a compactness closer to 1 and an area that meets the range constraint as the effective region; S205 Morphological optimization: Perform hole filling, small connected component removal, and erosion on the effective region to obtain a more compact final effective region; When the number of effective pixels after erosion is too small, the result before erosion can be retained as a fallback.

[0075] (2) Partition-weighted centroid (see Figure 3 S211 determines the bounding box of the final effective region, obtaining the effective range [x] start ,x end ]×[y start ,y end S212 partition boundary generation, along the horizontal direction [x start ,x end Divide the partition into n equal segments to obtain n+1 partition boundary coordinates x. bound The preferred method is to generate pixel coordinates by linear division followed by rounding, and duplicate boundaries are removed and ranges are filtered. In S213, local centroid calculation is performed: a mask is constructed for each partition i, retaining only the valid pixels within that partition, and the local centroid (x{c,i}, y{c,i}) is obtained by weighting by grayscale, using the sum of the grayscale values ​​of the partitions as the weight w. i S214 Invalid partition removal, when w i When the value is 0, it is considered an invalid partition and does not participate in subsequent fusion; S215 global centroid fusion, the global centroid (x) is obtained by weighted averaging of all valid partitions. c ,y c ):x c =Σ(w i ·x{c,i}) / Σw i y c =Σ(w i ·y{c,i}) / Σw i If all partitions are invalid, then the gray-weighted centroid of the entire region is used as a fallback. In a preferred configuration, the number of partitions n can be 8 to balance anomaly resistance and computational efficiency.

[0076] (3) Multi-distance model selection and interpolation (see Figure 4 When the difference between the coarse distance measurement L and a certain calibrated distance Li is less than the distance tolerance, model f is directly selected. i When L is located at an adjacent calibration distance L i With L i+1 When the distance is between these parameters, interpolation can be performed on the model parameters to obtain the inversion model for the current distance. Taking a fifth-order polynomial as an example, the coefficients of each order can be interpolated separately:

[0077] a k (L)=a {k,i} ·(L i+1 −L) / (L i+1 -L i )+a {k,i+1} ·(L−L i ) / (L i +1−L i), where k=0…5. a is obtained by interpolating the coefficients. k After (L), calculate h2=a0(L)+a1(L)·Δv+…+a5(L)·(Δv)^5.

[0078] (4) Transition fusion and weight determination (see Figure 5 When L is in the transition interval (L0−ΔL,L0+ΔL), the phase elevation value h1 and the triangular elevation value h2 are obtained simultaneously, and the fused elevation h = w·h1 + (1−w) is calculated. · h2. Here, w can vary linearly with L, for example, w = (L0 + ΔL − L) / (2ΔL), and w is restricted to [0,1]. Furthermore, w can also be corrected using image quality metrics, which may include at least one of compactness C, effective area A, or total grayscale ΣI, used to reduce the weight of h2 when the spot quality is poor, thereby improving the stability of the fused output. The following is a demonstration of the effect of the spot image preprocessing workflow, such as... Figure 10 As shown, the original image is transformed into an image that can be processed under the same conditions after being converted to grayscale and normalized.

[0079] like Figure 11 As shown, after thresholding and morphological processing, a compact area of ​​light spot image is obtained, which completes the preprocessing stage for accurate calculation of the light spot centroid.

[0080] As a comparison method, the centroid coordinates were solved by applying a two-dimensional Gaussian fitting of the light spot, and the Matlab processing results are as follows: Figure 12 As shown;

[0081] It can be seen that two-dimensional Gaussian fitting needs to conform to the ideal state of the light spot, but the actual measured image light spot has asymmetry in the Y direction and abnormal gray level. The fitting parameters are shown in Table 4 below:

[0082] Table 4 Comparison of Gaussian Fitting Parameters

[0083]

[0084] According to the fitting parameters, only 66% of the grayscale changes in the light spot can be explained by the Gaussian model, while the remaining 34% of grayscale changes (such as local grayscale abrupt changes, edge blurring, and non-Gaussian shapes) cannot be described by the Gaussian function. The actual shape of the light spot differs significantly from the ideal Gaussian distribution (possibly due to aberrations, uneven target reflection, background noise interference, etc.). From the numerical perspective, the average deviation between the fitted value and the actual grayscale value is approximately 9.65 gray levels. This means that the Gaussian model cannot accurately capture the true grayscale distribution of the light spot, and the deviation will directly lead to the calculation offset of the center (centroid) of the Gaussian function, because the center parameter of the Gaussian model is solved based on the fitted grayscale distribution, rather than the grayscale distribution of the actual light spot, ultimately causing centroid extraction error.

[0085] The spot centroid optimization extraction method based on partition weighting was used, and the preprocessed image was calculated in MATLAB according to 8 partitions. The results are as follows. Figure 13 As shown, partitioning eliminates gray-level anomalies in the image, calculates the centroid for each partition separately, and then calculates the total centroid coordinates using gray-level weights, achieving sub-pixel resolution overall.

[0086] Example 3 (Improved Example - State Machine Control and Vibration Disturbance Suppression):

[0087] Based on Embodiment 1 or Embodiment 2, this embodiment presents a program organization method for continuous field measurement. The measurement process is implemented using a state machine, which ensures that the initialization, self-test, measurement, fusion, and anomaly handling stages have clear states and transition conditions, thereby guaranteeing the controllability of the system during long-term operation.

[0088] (1) State machine design: After the system is powered on, it enters the initialization state, completes parameter loading, communication link check and reference coordinate setting; then enters the phase coarse measurement state to obtain the distance coarse measurement value L, and triggers the transition to the phase ranging state, triangulation ranging state or transition fusion state according to L; after completing the elevation result release and data storage in the output state, it returns to the phase coarse measurement state to enter the next measurement cycle; when the image acquisition failure, effective area extraction failure or ranging output abnormality is detected, it enters the abnormal state to execute strategies such as re-acquisition, parameter reset or mode switching.

[0089] (2) Online quality assessment: In the triangulation branch, indicators such as the compactness, area, total grayscale, and saturated pixel ratio of the effective area of ​​the light spot can be calculated and compared with preset thresholds to obtain the image quality level; in the phase ranging branch, the phase quality level can be formed based on the stability of the ranging signal or the validity of the data. The quality assessment results can be used to: ① determine whether to trigger re-acquisition; ② correct the transition fusion weights; ③ directly adopt a single mode output in extreme cases to ensure continuity.

[0090] (3) Vibration interference suppression: Considering that there may be mid-frequency vibrations at the construction site, which may cause periodic fluctuations in the elevation results, sensor signals reflecting vibration (such as acceleration signals) can be selectively collected synchronously to identify the dominant vibration frequency or target frequency band, and a digital notch filter can be constructed to filter the measurement sequence. The filtering process can be applied to the centroid offset Δv, the triangular elevation value h2, or the fused elevation value h to suppress the target frequency band disturbance; without configuring vibration sensors, the system can still complete the measurement according to Example 1 or Example 2.

[0091] To objectively evaluate the effectiveness of the vibration compensation method based on acceleration feedback, a verification strategy that does not rely on external reference signals was adopted. Figure 14 The effects before and after compensation are shown.

[0092] The quantification results are shown in Table 5:

[0093] Table 5 Quantitative Results of Oscillation Suppression Effect

[0094]

[0095] As can be seen, over 67% of the unwanted oscillations were effectively suppressed, and the stability of the DisA signal was significantly improved. Here, DisA represents the elevation error sequence used to evaluate the vibration suppression effect before and after compensation. It is worth noting that the compensation process did not introduce trend distortion; it only removed interference components at specific frequencies, preserving the effective low-frequency information of the displacement signal.

[0096] Spectral analysis further verified the notch filtering effect.

[0097] like Figure 15 As shown, DisA exhibits a sharp peak at 6.35Hz before compensation; after compensation, this frequency component is deeply attenuated (theoretical suppression depth > −60dB), while the amplitudes of other frequency bands remain essentially unchanged, indicating that the filter has good frequency selectivity. Furthermore, due to the use of bidirectional zero-phase filtering, the compensated signal has no phase delay or lead in the time domain, ensuring event time alignment and avoiding the introduction of spurious dynamic characteristics due to filtering.

Claims

1. A dual-mode elevation measurement method with partitioned weighted centroid multi-distance calibration, characterized in that, Comprising: obtaining a phase ranging distance L and determining a phase mode, a triangular mode or a transition fusion mode according to a preset threshold; obtaining a first elevation value h1 in the phase mode or the transition fusion mode; collecting a light spot image in the triangular mode or the transition fusion mode, extracting an effective region and obtaining a centroid offset Δv through a partition weighted centroid method; selecting or interpolating to obtain a height inversion model from a multi-distance calibration model library based on the phase ranging distance L, and converting Δv into a second elevation value h2; outputting h1 in the phase mode, outputting h2 in the triangular mode, and performing weighted fusion on h1 and h2 according to a fusion weight w to obtain an elevation value h in the transition fusion mode; The multi-distance calibration model library is obtained through the following calibration process: at multiple calibration distances L i Next, adjust the height of the corresponding measuring points to multiple known heights h. j Acquire corresponding spot images and calculate centroid offset Δv ij Based on each calibration distance L i The following sample data is used to establish the height inversion model h=f corresponding to the calibration distance. i (Δv), which represents the model parameters of the height inversion model and the corresponding calibration distance L. i Associative storage forms a multi-distance calibration model library, where L i h is the i-th calibration distance. j For the j-th known height, Δv ij To be at the calibration distance L i and known height h j The centroid offset sample obtained below, f i To the calibration distance L i The corresponding height inversion model, where h is the elevation value obtained from the centroid offset Δv based on the height inversion model, h=f i (Δv) represents the functional relationship between elevation h and centroid offset Δv.

2. The dual-mode elevation measurement method for partitioned weighted centroid multi-distance calibration according to claim 1, characterized in that, the preset thresholds at least comprise a preset distance threshold L0 and a transition region half-width parameter ΔL; when L≤(L0-ΔL), the phase ranging mode is determined; when L≥(L0+ΔL), the triangular ranging mode is determined; when (L0-ΔL)<L<(L0+ΔL), the transition fusion mode is determined; L0 and ΔL are preset and stored mode determination parameters.

3. The dual-mode elevation measurement method for partitioned weighted centroid multi-distance calibration according to claim 1, characterized in that, Based on the phase ranging L, a height inversion model corresponding to the current measured distance is selected or generated, including: when the phase ranging L is close to a certain calibration distance L... i When the difference is not greater than the preset distance tolerance, select the distance L from the calibration distance. i The corresponding height inversion model f i When the phase ranging F satisfies L i <L<L i+1 At that time, according to the phase ranging, F is in the interval [L] i ,L i+1 The relative position in the figure, relative to the calibration distance L i and L i+1 Interpolate the parameters of the same-order models of the two corresponding height inversion models to obtain the height inversion model corresponding to the current measurement distance, and use the height inversion model to convert the centroid offset Δv into the second elevation value h2.

4. The dual-mode elevation measurement method for partitioned weighted centroid multi-distance calibration according to claim 2, characterized in that, in the transition fusion mode, the fused elevation value satisfies: h=w·h1+(1-w)·h2; wherein w is a fusion weight with a value between [0,1].

5. The dual-mode elevation measurement method for partitioned weighted centroid multi-distance calibration according to claim 4, characterized in that, the fusion weight w is determined according to the position of the current phase ranging distance L in the transition interval, continuously changes along with the current phase ranging distance L in the transition fusion interval, and satisfies: w=(L0+ΔL-L) / (2ΔL).

6. The dual-mode elevation measurement method for partitioned weighted centroid multi-distance calibration according to claim 1, characterized in that, obtaining light spot centroid coordinates by adopting the partition weighted centroid algorithm comprises: determining a light spot boundary range based on the effective region; dividing the effective region into a plurality of sub-regions along a centroid offset sensitive direction; respectively calculating local centroid coordinates of each sub-region; determining corresponding weights according to gray distribution of effective light spot pixels in each sub-region, and eliminating or reducing the weight of sub-regions without effective light spot pixels, with local gray abnormality or local distortion; fusing the local centroid coordinates of the remaining sub-regions according to the weights to obtain global centroid coordinates.

7. The dual-mode elevation measurement method based on partition weighted centroid and multi-distance calibration according to claim 1, characterized in that, synchronously collecting measured acceleration signals, and identifying a vibration main frequency or a target frequency band according to the acceleration signals; constructing a digital notch filter according to the vibration main frequency or the target frequency band; performing filtering compensation on at least one of the centroid offset Δv, the second elevation value h2 or the fused elevation value h by using the digital notch filter.

8. The dual-mode elevation measurement method for partitioned weighted centroid multi-distance calibration according to claim 1, characterized in that, the dual-mode elevation measurement process is controlled by a state machine, and phase ranging and light spot image quality are evaluated in each measurement cycle; when an evaluation result meets an abnormal condition, an abnormal processing strategy is triggered; the abnormal processing strategy at least comprises one of the following: adopting a full-image gray weighted centroid as a fallback centroid, re-collecting a light spot image, correcting the fusion weight w to reduce the coefficient of h2, or switching to a phase ranging mode for output.

9. A dual-mode elevation measurement data processing device, characterized in that... a device for implementing the method according to any one of claims 1 to 8, comprising a distance acquisition and mode determination module, a phase elevation acquisition module, an image acquisition module, an effective region extraction module, a partition weighted centroid calculation module, a calibration model selection and inversion module, and a fusion output module; The distance acquisition and mode determination module is used to acquire the phase ranging L and determine the phase, triangulation or transition fusion mode according to a preset threshold. The phase elevation acquisition module obtains the first elevation value h1 in phase or transition fusion mode; The image acquisition module acquires a light spot image; The effective region extraction module performs multi-threshold traversal segmentation on the normalized spot image and combines compactness evaluation to collaboratively select the region that is closer to the real spot outline as the effective region from multiple candidate spot regions. The partition-weighted centroid calculation module uses a partition-weighted centroid algorithm to obtain the centroid coordinates of the light spot. The calibration model selection and inversion module selects or interpolates the height inversion model from a multi-distance calibration model library based on phase ranging L; The aforementioned fusion output module outputs h1 in phase mode, h2 in triangular mode, and in transition fusion mode, it weights and fuses h1 and h2 according to fusion weight w to obtain the elevation value h.

Images