Near-infrared displacement monitoring method and system based on multi-reference point correction

Abstract

The invention provides a near-infrared displacement monitoring method and system based on multi-reference point correction, and the method comprises the steps: obtaining a near-infrared target image sequence, and extracting a candidate reference point set and a measuring point set; performing multi-dimensional feature extraction on the two to obtain a feature vector set; inputting the feature vectors into a preset multi-dimensional similarity model for similarity evaluation, and generating a comprehensive similarity matrix; self-adaptive matching is carried out based on the matrix and a similarity threshold value, and a reference point-measuring point matching relation table is established; and finally, performing differential correction on the measuring point set according to the matching relation to obtain a measuring point displacement value set. According to the near-infrared displacement monitoring method and system based on multi-reference point correction provided by the invention, the reference stability is improved through multi-reference point screening and multi-dimensional feature extraction, and adaptive matching is performed through the preset multi-dimensional similarity model to ensure the matching accuracy, so that the displacement data accurately reflects the real deformation of the structure, and the accuracy of the real deformation of the structure is improved. And the accuracy and robustness of a monitoring result are improved.

Description

Technical Field

[0001] This invention belongs to the field of image acquisition and measurement technology, specifically relating to a near-infrared displacement monitoring method and system based on multi-reference point correction. Background Technology

[0002] Near-infrared displacement monitoring technology has been widely used in the health monitoring of large engineering structures such as bridges, dams, and tunnels due to its advantages such as strong anti-interference ability and suitability for long-distance non-contact measurement. However, existing near-infrared displacement monitoring methods still have several limitations in practical applications. Most methods rely on a single reference point to correct the displacement of the measuring point. When the reference point deviates due to changes in lighting, atmospheric disturbances, or local obstruction, its instability will be directly transmitted to all measuring points, leading to a systematic decrease in measurement accuracy. In addition, when there are multiple reference points and multiple measuring points within the monitoring field of view, existing technologies generally lack an effective automatic matching mechanism, often relying on manual presets or simple geometric rules. This makes it difficult to ensure the accuracy of the correspondence between reference points and measuring points in complex environments, easily leading to matching errors and affecting the reliability of subsequent displacement correction. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a near-infrared displacement monitoring method and system based on multi-reference point correction to solve the aforementioned problems. This method and system improve the stability of the reference reference by screening multiple reference points and extracting multi-dimensional features, and ensure matching accuracy by using a preset multi-dimensional similarity model for adaptive matching. This enables the displacement data to accurately reflect the true deformation of the structure, thereby improving the accuracy and robustness of the monitoring results.

[0004] To address the aforementioned technical problems, this invention provides a near-infrared displacement monitoring method based on multi-reference point correction, comprising the following steps: Obtain the original near-infrared target image sequence; Based on the original near-infrared target image sequence, a candidate reference point set and a measurement point set are obtained; Multidimensional feature extraction is performed on the candidate reference point set and the measurement point set to obtain the corresponding candidate reference point feature vector set and measurement point feature vector set; The candidate reference point feature vector set and the measurement point feature vector set are input into a preset multidimensional similarity model so that the multidimensional similarity model performs similarity evaluation based on the candidate reference point feature vector set and the measurement point feature vector set to obtain a comprehensive similarity matrix; Adaptive matching is performed based on the comprehensive similarity matrix and the preset similarity threshold to obtain a benchmark point-test point matching relationship table; Based on the reference point-measuring point matching relationship table, the measuring point set is differentially corrected to obtain the measuring point displacement value set.

[0005] In the above scheme, the original near-infrared target image sequence is first acquired. Based on the image sequence, a candidate reference point set and a measurement point set are obtained, and multi-dimensional feature extraction is performed to obtain a candidate reference point feature vector set and a measurement point feature vector set. Multi-reference point correction overcomes the susceptibility of traditional single-point sources to environmental interference, improving the reliability and stability of the reference reference. Then, the feature vector set is input into a preset multi-dimensional similarity model for similarity evaluation, generating a comprehensive similarity matrix. Based on the matrix and a preset similarity threshold, adaptive matching is performed to obtain a reference point-measurement point matching relationship table, avoiding the risk of mismatch caused by manual setting or simple rule matching, and ensuring matching accuracy in complex imaging environments. Finally, based on the matching relationship table, the measurement point set is differentially corrected to obtain a measurement point displacement value set, eliminating measurement errors introduced by interference such as jitter and illumination changes, ensuring that the displacement data accurately reflects the true deformation of the structure, and improving the accuracy and robustness of the monitoring results.

[0006] It should be noted that the process first involves acquiring the original near-infrared target image sequence, and then obtaining a candidate reference point set and a measurement point set based on the image sequence. Specifically, this can involve deploying multiple infrared reflective targets within the monitoring field of view. All targets adopt patterns with good geometric features and high reflectivity. By performing binarization, filtering, edge extraction, contour point extraction, and least-squares circle fitting on the image, the center coordinates of each target in the image coordinate system are extracted, thereby obtaining the candidate reference point set and the measurement point set.

[0007] Further, the step of performing multi-dimensional feature extraction based on the candidate reference point set and the measurement point set to obtain the corresponding candidate reference point feature vector set and measurement point feature vector set includes: Photometric features are extracted from the candidate reference point set and the measurement point set to obtain the corresponding candidate reference point photometric feature set and measurement point photometric feature set. Based on the candidate reference point set and the measurement point set, temporal features are extracted respectively to obtain the corresponding candidate reference point temporal feature set and measurement point temporal feature set; The candidate reference point feature vector set is constructed based on the candidate reference point photometric feature set and the candidate reference point temporal feature set; The measurement point feature vector set is constructed based on the measurement point photometric feature set and the measurement point temporal feature set.

[0008] In the above scheme, by extracting the photometric and temporal features of candidate reference points and measurement points respectively, and constructing corresponding feature vector sets, the limitations of insufficient description capability of single features are effectively overcome. This scheme comprehensively utilizes the photometric and temporal feature sets of the target to construct more discriminative candidate reference point feature vector sets and measurement point feature vector sets; it improves the comprehensiveness and accuracy of similarity assessment between reference points and measurement points, providing a reliable basis for subsequent adaptive matching. Through the multi-dimensional synergy of photometric and temporal features, the above scheme can more effectively distinguish different targets in complex imaging environments, enhancing the adaptability and robustness of reference point selection, thereby improving the overall anti-interference capability and measurement accuracy of the displacement monitoring method, especially suitable for severe engineering scenarios such as illumination changes and local occlusion.

[0009] It should be noted that the photometric features are used to characterize the grayscale characteristics of the target, and may include the average grayscale value, grayscale variance, and local grayscale histogram distribution. The temporal features are used to characterize the stability of the target over time, and may include brightness stability (i.e., the variance of the average grayscale over a period of time) and detection continuity (i.e., the proportion of targets successfully detected in consecutive frames). The combination of the two can effectively identify stable reference points with less interference. The candidate reference point feature vector set and the measurement point feature vector set can be standardized to unify different features to a scale of zero mean and unit variance, ensuring the objectivity of subsequent similarity. Further, the candidate reference point feature vector set and the measurement point feature vector set are input into a preset multidimensional similarity model, so that the multidimensional similarity model performs similarity evaluation based on the candidate reference point feature vector set and the measurement point feature vector set to obtain a comprehensive similarity matrix. The multidimensional similarity model includes a photometric feature similarity acquisition module, a temporal feature similarity acquisition module, and a comprehensive similarity construction module; wherein: The candidate reference point feature vector set and the measurement point feature vector set are input into a preset multidimensional similarity model so that the photometric feature similarity acquisition module can obtain the photometric feature similarity between the candidate reference point and the measurement point based on the candidate reference point feature vector set and the measurement point feature vector set, thereby obtaining a photometric feature similarity set. The candidate reference point feature vector set and the measurement point feature vector set are input to the temporal feature similarity acquisition module, so that the temporal feature similarity acquisition module obtains the temporal feature similarity between the candidate reference point and the measurement point based on the candidate reference point feature vector set and the measurement point feature vector set, and obtains the temporal feature similarity set; The photometric feature similarity set and the temporal feature similarity set are input into the comprehensive similarity construction module, so that the comprehensive similarity construction module performs a weighted sum based on the photometric feature similarity set and the temporal feature similarity set to obtain the comprehensive similarity matrix.

[0010] The above scheme constructs a multi-dimensional similarity model comprising a photometric feature similarity acquisition module, a temporal feature similarity acquisition module, and a comprehensive similarity construction module, thereby achieving both component-based and comprehensive evaluation of the similarity between reference points and measurement points. In this multi-dimensional similarity model, the photometric feature similarity acquisition module first obtains the photometric feature similarity set between candidate reference points and measurement points, and the temporal feature similarity acquisition module obtains their temporal feature similarity set. Then, the comprehensive similarity construction module performs a weighted summation based on the photometric feature similarity set and the temporal feature similarity set to obtain the comprehensive similarity matrix. This scheme effectively overcomes the limitations of a single similarity measure by fusing similarity information from different dimensions through a weighted summation strategy, resulting in more comprehensive and reliable evaluation results. This improves the accuracy of determining the matching relationship between reference points and measurement points.

[0011] It should be noted that when performing weighted summation, a first weight parameter and a second weight parameter can be preset. By adjusting the first weight parameter and the second weight parameter, the contribution of photometric features and temporal features to the overall similarity can be balanced.

[0012] Further, the candidate reference point feature vector set and the measurement point feature vector set are input into a preset multidimensional similarity model, so that the photometric feature similarity acquisition module obtains the photometric feature similarity between the candidate reference point and the measurement point based on the candidate reference point feature vector set and the measurement point feature vector set, and obtains a photometric feature similarity set. The following steps are then performed within the photometric feature similarity acquisition module: For each candidate reference point feature vector in the candidate reference point feature vector set and each measurement point feature vector in the measurement point feature vector set, the average gray value, gray value variance, and local gray value histogram of the feature vector of that point are obtained respectively. For each candidate benchmark point and each measurement point, the Euclidean distance between the candidate benchmark point and the measurement point is obtained based on the average gray value and gray variance value corresponding to the candidate benchmark point and the measurement point. Based on the local gray-level histograms corresponding to the candidate reference point and the measuring point, the Batachalia distance between the candidate reference point and the measuring point is obtained. Based on the Euclidean distance and Battacharian distance between the candidate benchmark point and the measurement point, the photometric feature similarity is obtained through the Gaussian kernel function. The photometric feature similarity set is obtained based on several photometric feature similarities.

[0013] In the above scheme, the photometric feature similarity acquisition module obtains the average gray value, gray variance, and local gray histogram for each candidate reference point feature vector and measurement point feature vector, respectively. Based on the average gray value and gray variance of the candidate reference point and measurement point, the Euclidean distance is obtained, and based on its local gray histogram, the Patacharian distance is obtained. Then, the Euclidean distance and Patacharian distance are fused using a Gaussian kernel function to obtain the photometric feature similarity, ultimately resulting in the photometric feature similarity set. This scheme comprehensively considers multiple attributes such as average gray value, gray variance, and local gray histogram, and effectively fuses the Euclidean distance and Patacharian distance using a Gaussian kernel function, constructing a robust photometric feature similarity evaluation index. It effectively overcomes the limitations of a single distance metric, improves the accuracy and stability of photometric feature similarity calculation, and provides technical support for generating a reliable photometric feature similarity set.

[0014] Further, the step of inputting the candidate reference point feature vector set and the measurement point feature vector set into the temporal feature similarity acquisition module allows the temporal feature similarity acquisition module to obtain the temporal feature similarity between the candidate reference point and the measurement point based on the candidate reference point feature vector set and the measurement point feature vector set, thereby obtaining a temporal feature similarity set. Within the temporal feature similarity acquisition module, the following steps are performed: For each candidate reference point feature vector in the candidate reference point feature vector set and each measurement point feature vector in the measurement point feature vector set, the brightness stability and detection continuity of the feature vector are obtained respectively. For each candidate reference point and each measurement point, the brightness stability similarity between the candidate reference point and the measurement point is obtained based on the brightness stability corresponding to the candidate reference point and the measurement point. Based on the detection continuity between the candidate reference point and the test point, the detection continuity similarity between the candidate reference point and the test point is obtained. Based on the brightness stability similarity and detection continuity similarity corresponding to the candidate reference point and the measurement point, the temporal feature similarity is obtained by weighted summation.

[0015] In the above scheme, the temporal feature similarity acquisition module acquires the brightness stability and detection continuity of each candidate reference point feature vector and measurement point feature vector, respectively. Based on the brightness stability of the candidate reference point and the measurement point, a brightness stability similarity is acquired, and based on their detection continuity, a detection continuity similarity is acquired. Then, the brightness stability similarity and detection continuity similarity are fused by weighted summation to obtain the temporal feature similarity. This scheme comprehensively considers the brightness stability and detection continuity of the target in the time dimension and effectively fuses the two feature similarities using a weighted summation strategy, constructing a temporal feature evaluation index that reflects long-term stability. It effectively overcomes the limitations of relying solely on brightness or existence indicators, improving the comprehensiveness and reliability of temporal feature similarity assessment.

[0016] Furthermore, based on the comprehensive similarity matrix and the preset similarity threshold, adaptive matching is performed to obtain a benchmark point-test point matching relationship table; including: Each comprehensive similarity in the comprehensive similarity matrix is ​​compared with a preset similarity threshold using a mask. If the comprehensive similarity is less than the preset similarity threshold, the comprehensive similarity element is set to an invalid value, thus obtaining a masked comprehensive similarity matrix. A cost matrix is ​​obtained based on the mask comprehensive similarity matrix, wherein the cost value in the cost matrix is ​​negatively correlated with the comprehensive similarity in the comprehensive similarity matrix; Based on the baseline-test point pair corresponding to each cost value in the cost matrix, the candidate matching set is obtained by sorting the cost values ​​from smallest to largest. Based on the candidate matching set, each reference point-measure point pair in the candidate matching set is traversed sequentially and a matching relationship is established to obtain the reference point-measure point matching relationship table.

[0017] In the above scheme, a mask is used to compare the comprehensive similarity with a preset similarity threshold, filtering out invalid matching pairs below the threshold to obtain a masked comprehensive similarity matrix. A negatively correlated cost matrix is ​​constructed based on this masked comprehensive similarity matrix, and candidate matching sets are obtained by sorting the cost values ​​from smallest to largest. Finally, this set is traversed and matching relationships are established sequentially to obtain a reference point-measurement point matching relationship table. This scheme effectively eliminates low-reliability matching pairs through threshold mask preprocessing, reducing the risk of mismatches and overcoming matching uncertainties caused by environmental interference, thus ensuring the accuracy and stability of matching relationships under complex imaging conditions.

[0018] Further, the step of sequentially traversing each reference point-measure point pair in the candidate matching set and establishing a matching relationship to obtain a reference point-measure point matching relationship table includes: Each benchmark-test point in the candidate matching set is processed sequentially according to its value, from smallest to largest. For the current reference point-measurement point, determine whether the measurement point has been assigned a reference point. If the measurement point has not been assigned a reference point, then establish a matching relationship between the measurement point and the reference point. If the measuring point has already been assigned a reference point, skip the current reference point-measuring point pair and continue processing the next reference point-measuring point pair.

[0019] In the above scheme, each benchmark-test point pair in the candidate matching set is processed sequentially according to its algebraic value from smallest to largest. During the traversal, it is determined whether the test point has already been assigned a benchmark point. If it has not been assigned, a matching relationship is established; if it has been assigned, the current benchmark-test point pair is skipped, and the next benchmark-test point pair is processed. By introducing a unique test point allocation mechanism, the above scheme ensures that each test point establishes a one-to-one mapping relationship with the current best candidate benchmark point, effectively avoiding matching conflicts caused by multiple benchmark points competing for the same test point. This guarantees the determinism of the matching process, improves matching efficiency and rationality, and provides a stable benchmark-test point matching relationship table for subsequent differential correction.

[0020] Further, based on the candidate matching set, each reference point-measure point pair in the candidate matching set is traversed sequentially and a matching relationship is established to obtain a reference point-measure point matching relationship table, which is implemented through a first matching strategy or a second matching strategy, wherein: Under the first matching strategy, if the measuring point in the current benchmark-measuring point pair is not assigned a benchmark point, and the benchmark points in the current benchmark-measuring point pair have already established a matching relationship, then the measuring point is set as an invalid point and no matching relationship is established. Under the second matching strategy, if the measuring point in the current benchmark-measuring point pair is not assigned a benchmark point, and the benchmark points in the current benchmark-measuring point pair have already established a matching relationship, then the measuring point will still establish a matching relationship with the benchmark point to form a matching relationship of one benchmark point matching multiple measuring points.

[0021] The above scheme achieves flexible adaptation of matching relationships by introducing two configurable matching strategies. Under the first matching strategy, if a measuring point is not assigned a reference point but a matching relationship has been established with a reference point, the measuring point is set as an invalid point, ensuring a strict one-to-one correspondence between the reference point and the measuring point. Under the second matching strategy, under the same conditions, one reference point is allowed to match multiple measuring points, forming a one-to-many matching relationship. The above scheme can be selected according to the needs of the actual monitoring scenario. When reference point resources are sufficient, priority is given to ensuring correction accuracy, while when the number of reference points is limited, the matching coverage is ensured. This effectively solves the matching problem under different target deployment conditions, ensuring both data independence in high-precision scenarios and monitoring continuity in complex environments.

[0022] It should be noted that the first matching strategy can be used in scenarios with a sufficient number of reference points, prioritizing one-to-one relationships to improve correction accuracy; the second matching strategy can be used in scenarios with a limited number of reference points, allowing one-to-many relationships to ensure monitoring coverage.

[0023] Further, the step of performing differential correction on the set of measuring points based on the reference point-measuring point matching relationship table to obtain a set of measuring point displacement values ​​includes: For each matching relationship in the reference point-measurement point matching relationship table, perform the following steps: Obtain the initial frame coordinates of the measurement point, the initial frame coordinates of the reference point, the current frame coordinates of the measurement point, and the current frame coordinates of the reference point from the matching relationship; The original displacement vector of the measuring point is obtained based on the initial frame coordinates and the current frame coordinates of the measuring point. The displacement vector of the reference point is obtained based on the initial frame coordinates and the current frame coordinates of the reference point. The original displacement vector of the measuring point and the displacement vector of the reference point are used to perform a difference operation to obtain the corrected displacement vector of the measuring point. The displacement value set of the measuring points is constructed based on the displacement vector correction of several measuring points.

[0024] In the above scheme, differential correction is performed on each matching relationship based on the reference point-measuring point matching table: the initial frame coordinates and current frame coordinates of the measuring point and the reference point are obtained, the original displacement vector of the measuring point and the displacement vector of the reference point are calculated respectively, and then differential operation is performed to obtain the corrected displacement vector of the measuring point, finally constructing the measuring point displacement value set. This scheme effectively eliminates systematic errors introduced by factors such as camera shake and environmental changes by extracting the reference point displacement vector from the original displacement vector of the measuring point. This ensures that the corrected displacement vector accurately reflects the true deformation of the structure, thereby improving the accuracy and reliability of the displacement monitoring data. It ensures that the measuring point displacement value set can truly represent the actual deformation state of the structure, enhances the anti-interference capability of monitoring in complex engineering environments, and provides a highly reliable data foundation for structural health monitoring.

[0025] The present invention also provides a near-infrared displacement monitoring system based on multi-reference point correction, comprising: Image sequence acquisition module, used to acquire raw near-infrared target image sequences; The target recognition module is used to obtain a candidate reference point set and a measurement point set based on the original near-infrared target image sequence; The feature extraction module is used to perform multi-dimensional feature extraction based on the candidate reference point set and the measurement point set, respectively, to obtain the candidate reference point feature vector set and the measurement point feature vector set; The similarity evaluation module is used to input the candidate reference point feature vector set and the measurement point feature vector set into a preset multidimensional similarity model, so that the multidimensional similarity model performs similarity evaluation based on the candidate reference point feature vector set and the measurement point feature vector set to obtain a comprehensive similarity matrix; An adaptive matching module is used to perform adaptive matching based on the comprehensive similarity matrix and a preset similarity threshold to obtain a benchmark point-test point matching relationship table; The differential correction module is used to perform differential correction on the set of measuring points based on the reference point-measuring point matching relationship table to obtain a set of measuring point displacement values.

[0026] The above scheme achieves fully automated near-infrared displacement monitoring by integrating an image sequence acquisition module, target recognition module, feature extraction module, similarity evaluation module, adaptive matching module, and differential correction module into a collaborative architecture. First, the feature extraction module constructs a set of candidate reference point feature vectors and a set of measurement point feature vectors. Then, a comprehensive similarity matrix is ​​obtained using a preset multi-dimensional similarity model in the similarity evaluation module. Next, the adaptive matching module establishes a reference point-measurement point matching table based on a preset similarity threshold. Finally, the differential correction module obtains the accurate set of measurement point displacement values. This scheme effectively overcomes the problems of excessive manual intervention, strong single-point dependence, and weak anti-interference ability in traditional methods through the collaborative work of multiple modules. It improves the automation, accuracy, and environmental adaptability of displacement monitoring, providing a reliable technical means for the health monitoring of large-scale engineering structures. Attached Figure Description

[0027] Figure 1 This is a schematic flowchart of a near-infrared displacement monitoring method based on multi-reference point correction, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of a near-infrared displacement monitoring system architecture based on multi-reference point correction, provided as an embodiment of the present invention. Detailed Implementation

[0028] 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.

[0029] Please see Figure 1 This embodiment provides a near-infrared displacement monitoring method based on multi-reference point correction, including the following steps: Obtain the original near-infrared target image sequence; Based on the original near-infrared target image sequence, a candidate reference point set and a measurement point set are obtained; Multidimensional feature extraction is performed on the candidate reference point set and the measurement point set to obtain the corresponding candidate reference point feature vector set and measurement point feature vector set; The candidate reference point feature vector set and the measurement point feature vector set are input into a preset multidimensional similarity model so that the multidimensional similarity model performs similarity evaluation based on the candidate reference point feature vector set and the measurement point feature vector set to obtain a comprehensive similarity matrix; Adaptive matching is performed based on the comprehensive similarity matrix and the preset similarity threshold to obtain a benchmark point-test point matching relationship table; Based on the reference point-measuring point matching relationship table, the measuring point set is differentially corrected to obtain the measuring point displacement value set.

[0030] In this embodiment, firstly, the original near-infrared target image sequence is acquired. Based on the image sequence, a candidate reference point set and a measurement point set are obtained, and multi-dimensional feature extraction is performed to obtain a candidate reference point feature vector set and a measurement point feature vector set. Multi-reference point correction overcomes the susceptibility of traditional single-point sources to environmental interference, improving the reliability and stability of the reference reference. Then, the feature vector set is input into a preset multi-dimensional similarity model for similarity evaluation, generating a comprehensive similarity matrix. Based on the matrix and a preset similarity threshold, adaptive matching is performed to obtain a reference point-measurement point matching relationship table, avoiding the risk of mismatch caused by manual setting or simple rule matching, and ensuring matching accuracy in complex imaging environments. Finally, based on the matching relationship table, differential correction is performed on the measurement point set to obtain a measurement point displacement value set, eliminating measurement errors introduced by interference such as jitter and illumination changes, ensuring that the displacement data accurately reflects the true deformation of the structure, and improving the accuracy and robustness of the monitoring results.

[0031] In one embodiment, a near-infrared displacement monitoring method based on multi-reference point correction is provided, which involves acquiring an original near-infrared target image sequence and acquiring a candidate reference point set and a measurement point set based on the original near-infrared target image sequence. This can be achieved in the following specific ways: Multiple specialized targets with high near-infrared reflectivity are deployed at stable locations (such as stationary components surrounding the structure under test) and at key locations of the displacement to be measured within the structural monitoring area. All targets employ patterns with good, stable geometric features (such as circles) to ensure high-precision identification and positioning in the images.

[0032] After acquiring the raw near-infrared image, the image is first preprocessed and target identified. Specifically, the input image undergoes binarization, filtering and noise reduction, edge extraction, and contour point set extraction to accurately segment the pixel region of each target. Subsequently, least-squares circle fitting is performed on the extracted target contour point set to calculate and obtain the center coordinates (x, y) of each target in the image coordinate system. The center points of all identified targets together constitute the candidate target set.

[0033] It should be noted that the candidate target set here includes both the reference points used for subsequent differential correction and the measurement points to be monitored. By deploying multiple targets and performing high-precision positioning in a unified manner, a data foundation is laid for subsequent reference point selection and matching based on multi-dimensional features, overcoming the dependence of traditional methods on a single, preset reference point.

[0034] It should be noted that in the multi-dimensional feature extraction based on the candidate reference point set and the measurement point set, the multi-dimensional features can be constructed based on the identified targets. Specifically, for each target with extracted center coordinates, a series of image features can be further calculated to form a feature vector. The features may include, but are not limited to: the radius of the target fitted circle, the gray centroid of the target image, the gray distribution moments of the target image in the main directions, and the invariant moments of the target contour.

[0035] Further, the step of performing multi-dimensional feature extraction based on the candidate reference point set and the measurement point set to obtain the corresponding candidate reference point feature vector set and measurement point feature vector set includes: Photometric features are extracted from the candidate reference point set and the measurement point set to obtain the corresponding candidate reference point photometric feature set and measurement point photometric feature set. Based on the candidate reference point set and the measurement point set, temporal features are extracted respectively to obtain the corresponding candidate reference point temporal feature set and measurement point temporal feature set; The candidate reference point feature vector set is constructed based on the candidate reference point photometric feature set and the candidate reference point temporal feature set; The measurement point feature vector set is constructed based on the measurement point photometric feature set and the measurement point temporal feature set.

[0036] In this embodiment, by extracting the photometric and temporal features of candidate reference points and measurement points respectively, and constructing corresponding feature vector sets, the limitations of insufficient single feature description capability are effectively overcome. This embodiment comprehensively utilizes the photometric and temporal feature sets of the target to construct a more discriminative set of candidate reference point feature vectors and measurement point feature vectors; it improves the comprehensiveness and accuracy of similarity assessment between reference points and measurement points, providing a reliable basis for subsequent adaptive matching. Through the multi-dimensional synergy of photometric and temporal features, this embodiment can more effectively distinguish different targets in complex imaging environments, enhancing the adaptability and robustness of reference point selection, thereby improving the overall anti-interference capability and measurement accuracy of the displacement monitoring method, especially suitable for severe engineering scenarios such as illumination changes and partial occlusion.

[0037] In one embodiment, the step of performing multi-dimensional feature extraction based on the candidate reference point set and the measurement point set to obtain the corresponding candidate reference point feature vector set and measurement point feature vector set includes the following specific steps: First, photometric features are extracted from the candidate reference point set and the measurement point set, respectively. These photometric features include: average gray value (…). ), grayscale variance ( ), and the local grayscale histogram distribution ( This yields the corresponding candidate reference point photometric feature set and the measurement point photometric feature set.

[0038] Secondly, based on the temporal information of the image sequence, temporal features of the candidate reference point set and the measurement point set are extracted respectively. Specifically, for each point, its brightness stability in consecutive frames of images is calculated. , defined as the variance of the average gray value of the point over a period of time) and detection continuity ( (i.e., the ratio of successful detection and localization of the point in consecutive frames), thereby obtaining the corresponding candidate reference point temporal feature set and measurement point temporal feature set.

[0039] Finally, based on the photometric and temporal feature sets extracted for each point, a multidimensional feature vector is constructed. Specifically, the above features are combined to form a feature vector that comprehensively describes the point. : Perform this operation on all candidate reference points and measurement points to obtain the feature vector sets of candidate reference points and measurement points.

[0040] In this embodiment, the photometric features of each target are extracted comprehensively ( , , ) and time series characteristics ( , This constructs a more discriminative feature vector. Photometric features describe the apparent properties of the target at the single-frame image level, while temporal features characterize the stability of its performance at the image sequence level. This effectively overcomes the limitations of insufficient descriptive power of single features and improves the comprehensiveness and accuracy of subsequent similarity assessments. Especially in complex engineering scenarios such as lighting fluctuations and temporary local occlusions, temporal features such as brightness stability and detection continuity can provide crucial evidence for distinguishing stable benchmark points from disturbed measurement points.

[0041] Further, the candidate reference point feature vector set and the measurement point feature vector set are input into a preset multidimensional similarity model, so that the multidimensional similarity model performs similarity evaluation based on the candidate reference point feature vector set and the measurement point feature vector set to obtain a comprehensive similarity matrix. The multidimensional similarity model includes a photometric feature similarity acquisition module, a temporal feature similarity acquisition module, and a comprehensive similarity construction module; wherein: The candidate reference point feature vector set and the measurement point feature vector set are input into a preset multidimensional similarity model so that the photometric feature similarity acquisition module can obtain the photometric feature similarity between the candidate reference point and the measurement point based on the candidate reference point feature vector set and the measurement point feature vector set, thereby obtaining a photometric feature similarity set. The candidate reference point feature vector set and the measurement point feature vector set are input to the temporal feature similarity acquisition module, so that the temporal feature similarity acquisition module obtains the temporal feature similarity between the candidate reference point and the measurement point based on the candidate reference point feature vector set and the measurement point feature vector set, and obtains the temporal feature similarity set; The photometric feature similarity set and the temporal feature similarity set are input into the comprehensive similarity construction module, so that the comprehensive similarity construction module performs a weighted sum based on the photometric feature similarity set and the temporal feature similarity set to obtain the comprehensive similarity matrix.

[0042] In this embodiment, a multi-dimensional similarity model is constructed, comprising a photometric feature similarity acquisition module, a temporal feature similarity acquisition module, and a comprehensive similarity construction module, to achieve both component-based and comprehensive evaluation of the similarity between reference points and measurement points. In this multi-dimensional similarity model, the photometric feature similarity acquisition module first obtains the photometric feature similarity set between candidate reference points and measurement points, and the temporal feature similarity acquisition module obtains their temporal feature similarity set. Then, the comprehensive similarity construction module performs a weighted summation based on the photometric feature similarity set and the temporal feature similarity set to obtain a comprehensive similarity matrix. This embodiment effectively overcomes the limitations of a single similarity measure by fusing similarity information from different dimensions through a weighted summation strategy, resulting in a more comprehensive and reliable evaluation result. This improves the accuracy of determining the matching relationship between reference points and measurement points.

[0043] Further, the candidate reference point feature vector set and the measurement point feature vector set are input into a preset multidimensional similarity model, so that the photometric feature similarity acquisition module obtains the photometric feature similarity between the candidate reference point and the measurement point based on the candidate reference point feature vector set and the measurement point feature vector set, and obtains a photometric feature similarity set. The following steps are then performed within the photometric feature similarity acquisition module: For each candidate reference point feature vector in the candidate reference point feature vector set and each measurement point feature vector in the measurement point feature vector set, the average gray value, gray value variance, and local gray value histogram of the feature vector of that point are obtained respectively. For each candidate benchmark point and each measurement point, the Euclidean distance between the candidate benchmark point and the measurement point is obtained based on the average gray value and gray variance value corresponding to the candidate benchmark point and the measurement point. Based on the local gray-level histograms corresponding to the candidate reference point and the measuring point, the Batachalia distance between the candidate reference point and the measuring point is obtained. Based on the Euclidean distance and Battacharian distance between the candidate benchmark point and the measurement point, the photometric feature similarity is obtained through the Gaussian kernel function. The photometric feature similarity set is obtained based on several photometric feature similarities.

[0044] In this embodiment, the photometric feature similarity acquisition module obtains the average gray value, gray variance, and local gray histogram for each candidate reference point feature vector and measurement point feature vector, respectively. Based on the average gray value and gray variance of the candidate reference point and measurement point, the Euclidean distance is obtained, and based on its local gray histogram, the Patacharian distance is obtained. Then, the Euclidean distance and Patacharian distance are fused using a Gaussian kernel function to obtain the photometric feature similarity, ultimately resulting in the photometric feature similarity set. This embodiment comprehensively considers multiple attributes such as average gray value, gray variance, and local gray histogram, and effectively fuses the Euclidean distance and Patacharian distance using a Gaussian kernel function, constructing a robust photometric feature similarity evaluation index. This effectively overcomes the limitations of a single distance metric, improves the accuracy and stability of photometric feature similarity calculation, and provides technical support for generating a reliable photometric feature similarity set.

[0045] Further, the step of inputting the candidate reference point feature vector set and the measurement point feature vector set into the temporal feature similarity acquisition module allows the temporal feature similarity acquisition module to obtain the temporal feature similarity between the candidate reference point and the measurement point based on the candidate reference point feature vector set and the measurement point feature vector set, thereby obtaining a temporal feature similarity set. Within the temporal feature similarity acquisition module, the following steps are performed: For each candidate reference point feature vector in the candidate reference point feature vector set and each measurement point feature vector in the measurement point feature vector set, the brightness stability and detection continuity of the feature vector are obtained respectively. For each candidate reference point and each measurement point, the brightness stability similarity between the candidate reference point and the measurement point is obtained based on the brightness stability corresponding to the candidate reference point and the measurement point. Based on the detection continuity between the candidate reference point and the test point, the detection continuity similarity between the candidate reference point and the test point is obtained. Based on the brightness stability similarity and detection continuity similarity corresponding to the candidate reference point and the measurement point, the temporal feature similarity is obtained by weighted summation.

[0046] In this embodiment, the temporal feature similarity acquisition module acquires the brightness stability and detection continuity of each candidate reference point feature vector and measurement point feature vector, respectively. Based on the brightness stability of the candidate reference point and the measurement point, a brightness stability similarity is acquired, and based on their detection continuity, a detection continuity similarity is acquired. Then, the brightness stability similarity and detection continuity similarity are fused by weighted summation to obtain the temporal feature similarity. This embodiment comprehensively considers the brightness stability and detection continuity of the target in the time dimension and effectively fuses the two feature similarities using a weighted summation strategy, constructing a temporal feature evaluation index that reflects long-term stability. This effectively overcomes the limitations of relying solely on brightness or existence indicators, improving the comprehensiveness and reliability of temporal feature similarity evaluation.

[0047] In one embodiment, the step of "inputting the candidate reference point feature vector set and the measurement point feature vector set into a preset multidimensional similarity model, so that the multidimensional similarity model performs similarity evaluation based on the candidate reference point feature vector set and the measurement point feature vector set to obtain a comprehensive similarity matrix" specifically includes the following steps: First, the components of the candidate reference point feature vector and the measurement point feature vector are standardized and preprocessed. Specifically, the feature vectors are... Each feature component in the model is standardized using the Z-score method (i.e., standard score) to transform it into a distribution with zero mean and unit variance.

[0048] It should be noted that this standardization process can eliminate the influence of different physical dimensions and numerical ranges on similarity calculations, making the subsequent similarity assessment results based on distance metrics more objective and fair.

[0049] Next, using the photometric feature similarity acquisition module in the multidimensional similarity model, the photometric feature similarity between the candidate reference point and the measurement point is calculated. The specific process is as follows: For each candidate reference point r and measurement point m, calculations are performed based on their standardized photometric characteristics: Calculate the Euclidean distance based on the average gray value and the gray value variance. : in, and These are the average gray values ​​of the standardized reference point and the measuring point, respectively. and Standardize its gray variance.

[0050] Calculate the Batachalia distance based on the local gray-level histogram distribution. : in, and These are the values ​​of the normalized grayscale histograms of the reference point and the measuring point in the kth bin, respectively.

[0051] The above distances are fused using a Gaussian kernel function and converted into photometric feature similarity. : in, is the smoothing parameter of the Gaussian kernel function. The closer the value is to 1, the more similar the photometric characteristics of the reference point r and the measuring point m are.

[0052] It should be noted that this embodiment comprehensively considers the target's average gray level, gray level variance, and gray level distribution histogram, and uses a Gaussian kernel function to effectively fuse Euclidean distance and Batachaljalian distance, thus constructing a robust photometric feature similarity index and overcoming the limitations of a single distance metric.

[0053] Then, using the temporal feature similarity acquisition module in the multidimensional similarity model, the temporal feature similarity between the candidate reference point and the measurement point is calculated. The specific process is as follows: Calculate brightness stability similarity : in, and These are the brightness stability indices (temporal variance of average grayscale value) for the reference point and the measurement point, respectively. This is a control factor.

[0054] Calculate the continuity similarity of the detection : in, , These are the continuity indicators for the benchmark point and the measuring point (with values ​​between 0 and 1).

[0055] By weighted summation and fusion of the above similarities, the temporal feature similarity is obtained. : in, Weighting coefficients (0≤ ≤1).

[0056] It should be noted that this embodiment, by comprehensively evaluating the consistency of brightness fluctuations and the continuous stability of detection over time, can effectively identify targets that exhibit consistent performance in long-term monitoring, thereby improving the robustness of the matching relationship to temporal disturbances.

[0057] Subsequently, using the comprehensive similarity construction module in the multidimensional similarity model, the photometric feature similarity and temporal feature similarity are weighted and fused to obtain the comprehensive similarity between the candidate reference point r and the measurement point m: in, , For the weight parameters, satisfying ≥0, ≥0, and + = 1.

[0058] By traversing all candidate reference points and measurement point pairs, the comprehensive similarity matrix can be obtained.

[0059] Finally, based on the comprehensive similarity matrix and the preset similarity threshold... Make a judgment: when the overall similarity between a pair of reference points and measuring points is... If the reference point and the measurement point are found to be highly similar in imaging characteristics and stability, they can be used as a valid candidate matching pair and proceed to the subsequent matching relationship establishment step.

[0060] This embodiment uses a preset threshold. Screening was conducted to ensure that only benchmark-measuring point pairs with high similarity were considered for subsequent differential correction, effectively avoiding the risk of errors introduced by mismatched or weakly correlated point pairs and ensuring the accuracy of the final displacement monitoring results.

[0061] Furthermore, based on the comprehensive similarity matrix and the preset similarity threshold, adaptive matching is performed to obtain a benchmark point-test point matching relationship table; including: Each comprehensive similarity in the comprehensive similarity matrix is ​​compared with a preset similarity threshold using a mask. If the comprehensive similarity is less than the preset similarity threshold, the comprehensive similarity element is set to an invalid value, thus obtaining a masked comprehensive similarity matrix. A cost matrix is ​​obtained based on the mask comprehensive similarity matrix, wherein the cost value in the cost matrix is ​​negatively correlated with the comprehensive similarity in the comprehensive similarity matrix; Based on the baseline-test point pair corresponding to each cost value in the cost matrix, the candidate matching set is obtained by sorting the cost values ​​from smallest to largest. Based on the candidate matching set, each reference point-measure point pair in the candidate matching set is traversed sequentially and a matching relationship is established to obtain the reference point-measure point matching relationship table.

[0062] In this embodiment, a mask is used to filter out invalid matching pairs below a preset similarity threshold by comparing the overall similarity with a comparison mask, resulting in a masked overall similarity matrix. A negatively correlated cost matrix is ​​then constructed based on this masked overall similarity matrix, and candidate matching sets are obtained by sorting the cost values ​​from smallest to largest. Finally, this set is traversed, and matching relationships are established sequentially to obtain a reference point-measurement point matching relationship table. This embodiment effectively eliminates low-reliability matching pairs through threshold mask preprocessing, reducing the risk of mismatches and overcoming the matching uncertainty caused by environmental interference, thus ensuring the accuracy and stability of matching relationships under complex imaging conditions.

[0063] Further, the step of sequentially traversing each reference point-measure point pair in the candidate matching set and establishing a matching relationship to obtain a reference point-measure point matching relationship table includes: Each benchmark-test point in the candidate matching set is processed sequentially according to its value, from smallest to largest. For the current reference point-measurement point, determine whether the measurement point has been assigned a reference point. If the measurement point has not been assigned a reference point, then establish a matching relationship between the measurement point and the reference point. If the measuring point has already been assigned a reference point, skip the current reference point-measuring point pair and continue processing the next reference point-measuring point pair.

[0064] In this embodiment, each benchmark-measure point pair in the candidate matching set is processed sequentially according to its algebraic value from smallest to largest. During the traversal, it is determined whether the measure point has already been assigned a benchmark point. If it has not been assigned, a matching relationship is established; if it has been assigned, the current benchmark-measure point pair is skipped, and the next benchmark-measure point pair is processed. This embodiment introduces a unique measure point allocation mechanism to ensure that each measure point establishes a one-to-one mapping relationship with the current best candidate benchmark point. This effectively avoids matching conflicts caused by multiple benchmark points competing for the same measure point, thereby ensuring the determinism of the matching process, improving matching efficiency and rationality, and providing a stable benchmark-measure point matching relationship table for subsequent differential correction.

[0065] Further, based on the candidate matching set, each reference point-measure point pair in the candidate matching set is traversed sequentially and a matching relationship is established to obtain a reference point-measure point matching relationship table, which is implemented through a first matching strategy or a second matching strategy, wherein: Under the first matching strategy, if the measuring point in the current benchmark-measuring point pair is not assigned a benchmark point, and the benchmark points in the current benchmark-measuring point pair have already established a matching relationship, then the measuring point is set as an invalid point and no matching relationship is established. Under the second matching strategy, if the measuring point in the current benchmark-measuring point pair is not assigned a benchmark point, and the benchmark points in the current benchmark-measuring point pair have already established a matching relationship, then the measuring point will still establish a matching relationship with the benchmark point to form a matching relationship of one benchmark point matching multiple measuring points.

[0066] In this embodiment, two configurable matching strategies are introduced to achieve flexible adaptation of matching relationships. Under the first matching strategy, if a measuring point is not assigned a reference point but a matching relationship has been established with a reference point, the measuring point is set as an invalid point, ensuring a strict one-to-one correspondence between the reference point and the measuring point. Under the second matching strategy, under the same conditions, one reference point is allowed to match multiple measuring points, forming a one-to-many matching relationship. This embodiment can select according to the needs of the actual monitoring scenario. When reference point resources are sufficient, priority is given to ensuring correction accuracy, while when the number of reference points is limited, the matching coverage is ensured. This effectively solves the matching problem under different target deployment conditions, ensuring both data independence in high-precision scenarios and monitoring continuity in complex environments.

[0067] In one embodiment, the adaptive matching based on the comprehensive similarity matrix and a preset similarity threshold to obtain the benchmark point-test point matching relationship table specifically includes the following steps: First, construct the cost matrix. Let the set of candidate benchmarks be... The set of measurement points is The comprehensive similarity matrix is The similarity is converted into a cost value, and a cost matrix C is constructed, where each element... The calculation formula is: In this embodiment, the above conversion makes the cost... Similarity negative correlation The smaller the value, the stronger the candidate benchmark. With measuring points The better and more reasonable the matching relationship, the better. The cost matrix provides a quantitative basis for subsequent optimal matching selection.

[0068] Next, the cost matrix is ​​masked based on a preset similarity threshold τ. This involves iterating through each element of the cost matrix C. If its corresponding comprehensive similarity Less than the preset threshold If the candidate pair is deemed unreliable, its corresponding cost will be deducted. Mark as invalid value (e.g., set to infinity); if Then keep The original value is obtained. This yields the masked cost matrix.

[0069] Then, a candidate matching set is generated and sorted by cost value. All baseline-test point pairs with non-invalid cost values ​​in the masked cost matrix are collected. This forms an initial candidate matching set. Each matching pair in this candidate matching set is then assigned a value... Sort them in ascending order.

[0070] Next, the sorted candidate matching set is traversed, and matching relationships are adaptively established. Each baseline-test point pair in the candidate matching set is processed sequentially in ascending order of cost value (i.e., in descending order of matching priority). The specific matching logic is determined according to a preset strategy, and one of the following two strategies can be adopted: The first matching strategy (i.e., one-to-one matching): During the traversal, for the currently processed benchmark-test point pair... : If the measuring point No reference points have been assigned yet, and reference points If no measuring points are assigned, then the measuring points will be... With reference point Establish a matching relationship.

[0071] If the measuring point A reference point has been assigned, or a reference point has been assigned. If a measurement point has already been assigned, skip the current reference point-measure point pair. Continue processing the next benchmark-measuring point pair.

[0072] In this implementation, the first matching strategy ensures that each measuring point is mapped to only one optimal reference point, and each reference point serves only one measuring point, forming a strict one-to-one matching relationship. When there are a sufficient number of stable reference points in the system, and each measuring point can find a unique and highly similar reference point, prioritizing this strategy helps avoid the loss of correction accuracy that may be introduced by multiple measuring points sharing the same reference point, ensuring the independence and high accuracy of data processing.

[0073] The second matching strategy (which allows one-to-many matching): During the traversal, for the currently processed base point-test point pair... : If the measuring point If no reference point has been assigned, then the measuring point will be... With reference point Establish a matching relationship.

[0074] If the measuring point If a reference point has already been assigned, skip the current pair. .

[0075] In this embodiment, the second matching strategy allows a reference point to establish a matching relationship with multiple measurement points simultaneously (i.e., one-to-many). When the number of measurement points in the monitoring field of view is greater than the number of stable reference points, and multiple measurement points show high similarity to the same reference point in both photometric and temporal characteristics (for example, multiple adjacent measurement points are distributed on the structural surface, while the number of stable reference points in the field of view is limited), it can ensure that all measurement points can obtain reference points for differential correction, thus guaranteeing the coverage and integrity of displacement monitoring results.

[0076] Finally, all successfully established benchmark-measurement point matching relationships are recorded to generate a final benchmark-measurement point matching relationship table, which will be used for subsequent differential correction steps.

[0077] Further, the step of performing differential correction on the set of measuring points based on the reference point-measuring point matching relationship table to obtain a set of measuring point displacement values ​​includes: For each matching relationship in the reference point-measurement point matching relationship table, perform the following steps: Obtain the initial frame coordinates of the measurement point, the initial frame coordinates of the reference point, the current frame coordinates of the measurement point, and the current frame coordinates of the reference point from the matching relationship; The original displacement vector of the measuring point is obtained based on the initial frame coordinates and the current frame coordinates of the measuring point. The displacement vector of the reference point is obtained based on the initial frame coordinates and the current frame coordinates of the reference point. The original displacement vector of the measuring point and the displacement vector of the reference point are used to perform a difference operation to obtain the corrected displacement vector of the measuring point. The displacement value set of the measuring points is constructed based on the displacement vector correction of several measuring points.

[0078] In this embodiment, differential correction is performed on each matching relationship based on the reference point-measurement point matching table: the initial frame coordinates and current frame coordinates of the measurement point and the reference point are obtained, the original displacement vector of the measurement point and the displacement vector of the reference point are calculated respectively, and then differential operation is performed to obtain the corrected displacement vector of the measurement point, finally constructing the measurement point displacement value set. This embodiment effectively eliminates systematic errors introduced by factors such as camera shake and environmental changes by separating the reference point displacement vector from the original displacement vector of the measurement point, so that the corrected displacement vector accurately reflects the real deformation of the structure, thereby improving the accuracy and reliability of displacement monitoring data, ensuring that the measurement point displacement value set can truly represent the actual deformation state of the structure, enhancing the monitoring anti-interference capability in complex engineering environments, and providing a highly reliable data foundation for structural health monitoring.

[0079] In one embodiment, the step of "performing differential correction on the set of measuring points based on the benchmark-measuring point matching table to obtain a set of measuring point displacement values" specifically includes the following steps: For each established benchmark-measuring point pair in the matching table, perform the following differential correction calculation: First, obtain the center coordinates of the reference point and the measurement point in the image coordinate system of the matching pair in the initial reference frame (denoted as time t0), and the center coordinates of the image coordinate system in the current frame (denoted as time t), specifically: Reference point coordinates: Initial frame coordinates Current frame coordinates Measurement point coordinates: Initial frame coordinates Current frame coordinates Calculate the displacement vector of the reference point : Calculate the original displacement vector of the measuring point : In this embodiment, the reference points are located in a theoretically stable region that does not move with the structure. Therefore, the displacement calculated from these reference points mainly reflects common-mode system disturbances such as camera shake, slight offsets in the imaging system, and recognition deviations caused by changes in illumination, rather than the actual displacement of the structure itself. The original displacement vector of the measuring points simultaneously includes both the actual deformation of the structure and the errors introduced by the disturbances.

[0080] Secondly, a difference operation is performed between the original displacement vector of the measuring point and the displacement vector of the reference point to obtain the corrected displacement vector of the measuring point. : In this embodiment, differential correction involves subtracting the systematic common-mode error characterized by the displacement of the reference point from the original displacement of the measuring point, thereby effectively eliminating error components introduced by overall disturbances in the imaging system or environmental changes. The near-infrared displacement monitoring method based on multi-reference point correction described in this embodiment reduces the influence of non-structural deformation factors on the displacement monitoring results, enabling the final set of measuring point displacement values ​​to more accurately reflect the true deformation of the structure. This improves the stability, accuracy, and anti-interference capability of the displacement monitoring data in complex engineering environments, providing a reliable data foundation for the accurate assessment of structural health.

[0081] Please see Figure 2 This embodiment also provides a near-infrared displacement monitoring system based on multi-reference point correction, including: Image sequence acquisition module, used to acquire raw near-infrared target image sequences; The target recognition module is used to obtain a candidate reference point set and a measurement point set based on the original near-infrared target image sequence; The feature extraction module is used to perform multi-dimensional feature extraction based on the candidate reference point set and the measurement point set, respectively, to obtain the candidate reference point feature vector set and the measurement point feature vector set; The similarity evaluation module is used to input the candidate reference point feature vector set and the measurement point feature vector set into a preset multidimensional similarity model, so that the multidimensional similarity model performs similarity evaluation based on the candidate reference point feature vector set and the measurement point feature vector set to obtain a comprehensive similarity matrix; An adaptive matching module is used to perform adaptive matching based on the comprehensive similarity matrix and a preset similarity threshold to obtain a benchmark point-test point matching relationship table; The differential correction module is used to perform differential correction on the set of measuring points based on the reference point-measuring point matching relationship table to obtain a set of measuring point displacement values.

[0082] In this embodiment, a collaborative architecture integrating an image sequence acquisition module, a target recognition module, a feature extraction module, a similarity evaluation module, an adaptive matching module, and a differential correction module achieves fully automated processing of near-infrared displacement monitoring. First, the feature extraction module constructs a set of candidate reference point feature vectors and a set of measurement point feature vectors. Then, a comprehensive similarity matrix is ​​obtained using a preset multi-dimensional similarity model in the similarity evaluation module. Next, the adaptive matching module establishes a reference point-measurement point matching table based on a preset similarity threshold. Finally, the differential correction module obtains the accurate set of measurement point displacement values. This embodiment effectively overcomes the problems of excessive manual intervention, strong single-point dependence, and weak anti-interference ability in traditional methods through the collaborative work of multiple modules. It improves the automation, accuracy, and environmental adaptability of displacement monitoring, providing a reliable technical means for the health monitoring of large-scale engineering structures.

[0083] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A near-infrared displacement monitoring method based on multi-reference point correction, characterized in that, Includes the following steps: Obtain the original near-infrared target image sequence; Based on the original near-infrared target image sequence, a candidate reference point set and a measurement point set are obtained; Multidimensional feature extraction is performed on the candidate reference point set and the measurement point set to obtain the candidate reference point feature vector set and the measurement point feature vector set; The candidate reference point feature vector set and the measurement point feature vector set are input into a preset multidimensional similarity model so that the multidimensional similarity model performs similarity evaluation based on the candidate reference point feature vector set and the measurement point feature vector set to obtain a comprehensive similarity matrix; Adaptive matching is performed based on the comprehensive similarity matrix and the preset similarity threshold to obtain a benchmark point-test point matching relationship table; Based on the reference point-measuring point matching relationship table, the measuring point set is differentially corrected to obtain the measuring point displacement value set.

2. The near-infrared displacement monitoring method based on multi-reference point correction according to claim 1, characterized in that, The step involves performing multi-dimensional feature extraction based on the candidate reference point set and the measurement point set to obtain corresponding candidate reference point feature vector sets and measurement point feature vector sets; including: Photometric features are extracted from the candidate reference point set and the measurement point set to obtain the corresponding candidate reference point photometric feature set and measurement point photometric feature set. Based on the candidate reference point set and the measurement point set, temporal features are extracted respectively to obtain the corresponding candidate reference point temporal feature set and measurement point temporal feature set; The candidate reference point feature vector set is constructed based on the candidate reference point photometric feature set and the candidate reference point temporal feature set; The measurement point feature vector set is constructed based on the measurement point photometric feature set and the measurement point temporal feature set.

3. The near-infrared displacement monitoring method based on multi-reference point correction according to claim 1, characterized in that, The process involves inputting the candidate reference point feature vector set and the measurement point feature vector set into a preset multidimensional similarity model. This allows the multidimensional similarity model to perform similarity evaluation based on the candidate reference point feature vector set and the measurement point feature vector set, resulting in a comprehensive similarity matrix. The multidimensional similarity model includes a photometric feature similarity acquisition module, a temporal feature similarity acquisition module, and a comprehensive similarity construction module. Wherein: The candidate reference point feature vector set and the measurement point feature vector set are input into a preset multidimensional similarity model so that the photometric feature similarity acquisition module can obtain the photometric feature similarity between the candidate reference point and the measurement point based on the candidate reference point feature vector set and the measurement point feature vector set, thereby obtaining a photometric feature similarity set. The candidate reference point feature vector set and the measurement point feature vector set are input to the temporal feature similarity acquisition module, so that the temporal feature similarity acquisition module obtains the temporal feature similarity between the candidate reference point and the measurement point based on the candidate reference point feature vector set and the measurement point feature vector set, and obtains the temporal feature similarity set; The photometric feature similarity set and the temporal feature similarity set are input into the comprehensive similarity construction module, so that the comprehensive similarity construction module performs a weighted sum based on the photometric feature similarity set and the temporal feature similarity set to obtain the comprehensive similarity matrix.

4. The near-infrared displacement monitoring method based on multi-reference point correction according to claim 3, characterized in that, The candidate reference point feature vector set and the measurement point feature vector set are input into a preset multidimensional similarity model, so that the photometric feature similarity acquisition module obtains the photometric feature similarity between the candidate reference point and the measurement point based on the candidate reference point feature vector set and the measurement point feature vector set, and obtains a photometric feature similarity set. The following steps are performed in the photometric feature similarity acquisition module: For each candidate reference point feature vector in the candidate reference point feature vector set and each measurement point feature vector in the measurement point feature vector set, the average gray value, gray value variance, and local gray value histogram of the feature vector of that point are obtained respectively. For each candidate benchmark point and each measurement point, the Euclidean distance between the candidate benchmark point and the measurement point is obtained based on the average gray value and gray variance value corresponding to the candidate benchmark point and the measurement point. Based on the local gray-level histograms corresponding to the candidate reference point and the measuring point, the Batachalia distance between the candidate reference point and the measuring point is obtained. Based on the Euclidean distance and Battacharian distance between the candidate benchmark point and the measurement point, the photometric feature similarity is obtained through the Gaussian kernel function. The photometric feature similarity set is obtained based on several photometric feature similarities.

5. The near-infrared displacement monitoring method based on multi-reference point correction according to claim 3, characterized in that, The step involves inputting the candidate reference point feature vector set and the measurement point feature vector set into the temporal feature similarity acquisition module. This allows the temporal feature similarity acquisition module to obtain the temporal feature similarity between the candidate reference point and the measurement point based on the candidate reference point feature vector set and the measurement point feature vector set, resulting in a temporal feature similarity set. Within the temporal feature similarity acquisition module, the following steps are performed: For each candidate reference point feature vector in the candidate reference point feature vector set and each measurement point feature vector in the measurement point feature vector set, the brightness stability and detection continuity of the feature vector are obtained respectively. For each candidate reference point and each measurement point, the brightness stability similarity between the candidate reference point and the measurement point is obtained based on the brightness stability corresponding to the candidate reference point and the measurement point. Based on the detection continuity between the candidate reference point and the test point, the detection continuity similarity between the candidate reference point and the test point is obtained. Based on the brightness stability similarity and detection continuity similarity corresponding to the candidate reference point and the measurement point, the temporal feature similarity is obtained by weighted summation.

6. The near-infrared displacement monitoring method based on multi-reference point correction according to claim 1, characterized in that, Based on the comprehensive similarity matrix and the preset similarity threshold, adaptive matching is performed to obtain a benchmark point-test point matching relationship table; including: Each comprehensive similarity in the comprehensive similarity matrix is ​​compared with a preset similarity threshold using a mask. If the comprehensive similarity is less than the preset similarity threshold, the comprehensive similarity element is set to an invalid value, thus obtaining a masked comprehensive similarity matrix. A cost matrix is ​​obtained based on the mask comprehensive similarity matrix, wherein the cost value in the cost matrix is ​​negatively correlated with the comprehensive similarity in the comprehensive similarity matrix; Based on the baseline-test point pair corresponding to each cost value in the cost matrix, the candidate matching set is obtained by sorting the cost values ​​from smallest to largest. Based on the candidate matching set, each reference point-measure point pair in the candidate matching set is traversed sequentially and a matching relationship is established to obtain the reference point-measure point matching relationship table.

7. The near-infrared displacement monitoring method based on multi-reference point correction according to claim 6, characterized in that, The step of sequentially traversing each reference point-measure point pair in the candidate matching set and establishing matching relationships to obtain a reference point-measure point matching relationship table includes: Each benchmark-test point in the candidate matching set is processed sequentially according to its value, from smallest to largest. For the current reference point-measurement point, determine whether the measurement point has been assigned a reference point. If the measurement point has not been assigned a reference point, then establish a matching relationship between the measurement point and the reference point. If the measuring point has already been assigned a reference point, skip the current reference point-measuring point pair and continue processing the next reference point-measuring point pair.

8. The near-infrared displacement monitoring method based on multi-reference point correction according to claim 7, characterized in that, The process involves sequentially traversing each reference point-measure point pair in the candidate matching set and establishing matching relationships to obtain a reference point-measure point matching relationship table. This is achieved through either a first matching strategy or a second matching strategy, wherein: Under the first matching strategy, if the measuring point in the current benchmark-measuring point pair is not assigned a benchmark point, and the benchmark points in the current benchmark-measuring point pair have already established a matching relationship, then the measuring point is set as an invalid point and no matching relationship is established. Under the second matching strategy, if the measuring point in the current benchmark-measuring point pair is not assigned a benchmark point, and the benchmark points in the current benchmark-measuring point pair have already established a matching relationship, then the measuring point will still establish a matching relationship with the benchmark point to form a matching relationship of one benchmark point matching multiple measuring points.

9. The near-infrared displacement monitoring method based on multi-reference point correction according to claim 1, characterized in that, The step of performing differential correction on the set of measuring points based on the reference point-measuring point matching table to obtain a set of measuring point displacement values ​​includes: For each matching relationship in the reference point-measurement point matching relationship table, perform the following steps: Obtain the initial frame coordinates of the measurement point, the initial frame coordinates of the reference point, the current frame coordinates of the measurement point, and the current frame coordinates of the reference point from the matching relationship; The original displacement vector of the measuring point is obtained based on the initial frame coordinates and the current frame coordinates of the measuring point. The displacement vector of the reference point is obtained based on the initial frame coordinates and the current frame coordinates of the reference point. The original displacement vector of the measuring point and the displacement vector of the reference point are used to perform a difference operation to obtain the corrected displacement vector of the measuring point. The displacement value set of the measuring points is constructed based on the displacement vector correction of several measuring points.

10. A near-infrared displacement monitoring system based on multi-reference point correction, characterized in that, include: Image sequence acquisition module, used to acquire raw near-infrared target image sequences; The target recognition module is used to obtain a candidate reference point set and a measurement point set based on the original near-infrared target image sequence; The feature extraction module is used to perform multi-dimensional feature extraction based on the candidate reference point set and the measurement point set, respectively, to obtain the candidate reference point feature vector set and the measurement point feature vector set; The similarity evaluation module is used to input the candidate reference point feature vector set and the measurement point feature vector set into a preset multidimensional similarity model, so that the multidimensional similarity model performs similarity evaluation based on the candidate reference point feature vector set and the measurement point feature vector set to obtain a comprehensive similarity matrix; An adaptive matching module is used to perform adaptive matching based on the comprehensive similarity matrix and a preset similarity threshold to obtain a benchmark point-test point matching relationship table; The differential correction module is used to perform differential correction on the set of measuring points based on the reference point-measuring point matching relationship table to obtain a set of measuring point displacement values.

Images