A center displacement measurement method based on the fusion of target detection and corner point area screening

Abstract

This invention discloses a center displacement measurement method that integrates target detection and corner region screening, comprising the following steps: using a YOLOv8 network to perform target detection on a structure image containing multiple targets, outputting the detection box coordinates for each target, and calculating the center coordinates of each detection box; for each obtained detection box, performing sub-pixel level corner detection within the image region of interest defined by the box, and extracting the coordinates of the four corners of the cross-shaped target; and calculating the corner coordinates by geometric fitting or averaging. This invention combines real-time target detection with sub-pixel fine measurement, improving detection efficiency and flexibility in multi-target environments (YOLOv8 achieves 28.6% mAP on the COCO dataset, enabling real-time image processing), while ensuring high measurement accuracy. The weighted fusion algorithm and anomaly removal mechanism further enhance the stability and accuracy of the algorithm in complex scenes.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and structural health monitoring technology, and more specifically, to a method for measuring center displacement that integrates target detection and corner region screening. Background Technology

[0002] Currently, optical methods are commonly used in structural health monitoring to measure minute displacements in structures such as dams and bridges. For example, displacement values ​​are obtained by setting up targets and using digital image processing. Traditional methods usually rely on camera calibration and image registration, converting changes in pixel coordinates in the image into actual displacements to complete the measurement.

[0003] However, current measurement processes involve complex camera calibration, and camera lens distortion can amplify measurement errors, making rapid deployment and maintenance on-site difficult. Furthermore, while existing technologies using deep learning object detection (such as the YOLO series) can achieve rapid localization of multiple targets, the detection box center is only a rough approximation with limited pixel-level accuracy, failing to meet the requirements for high-precision measurement. At the same time, although subpixel corner detection methods can locate target corners in images with extremely high precision, they are prone to failure in blurry, poorly lit, or occluded conditions due to their high requirements for image clarity. Current technologies lack measurement methods that combine deep learning object detection with high-precision corner extraction to balance detection robustness and localization accuracy.

[0004] Therefore, there is an urgent need for a new method that can achieve high-precision displacement measurement with little or no calibration requirements by combining the advantages of target detection and fine corner detection in multi-target situations. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a center displacement measurement method that integrates target detection and corner region screening. By combining real-time target detection with sub-pixel fine measurement, the method improves detection efficiency and flexibility in multi-target environments (YOLOv8 achieves an mAP of 28.6% on the COCO dataset, enabling real-time image processing) and ensures high measurement accuracy. The weighted fusion algorithm and anomaly removal mechanism further enhance the stability and accuracy of the algorithm in complex scenes, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a center displacement measurement method that integrates target detection and corner region screening, comprising the following steps:

[0007] Target detection and corner region filtering: First, the YOLOv8 network is used to detect structural images containing multiple targets. The approximate locations of the cross-shaped targets embedded within each ring are quickly identified from the original image, and the corresponding detection boxes are output. Regions of interest (ROIs) are cropped based on the spatial limitations of the detection boxes, and only the regions containing the targets are processed to remove interfering information, improving the efficiency and accuracy of corner extraction.

[0008] Subpixel corner extraction: Within the ROI corresponding to each detection box, corner detection is performed after image preprocessing (such as grayscale enhancement) to extract the four corners of the crosshair target. Subpixel-level refinement algorithms (such as Harris corner detection combined with cornerSubPix iterative optimization) are used to optimize the initial corner coordinates, so that the corner positions achieve subpixel accuracy.

[0009] Center position optimization fusion algorithm: Calculate the center of the YOLO detection box separately Fitting the center to the four corner points Then, a weight fusion strategy is constructed to obtain the optimization center. Specifically, weights are adaptively allocated based on metrics such as YOLO bounding box confidence and corner fitting error. The fusion formula is as follows: The outlier removal logic is designed to remove corner points or detection box data that are significantly deviated from the geometric structure before fusion, in order to improve robustness to blurred and interfering images;

[0010] Calibration-independent high-precision measurement: The measurement results of this invention are directly calculated based on the change in the center position within the image plane, eliminating the need for complex camera intrinsic parameter calibration and distortion correction. By calculating the pixel coordinate difference of the optimized center of the same target in adjacent frame images, high-precision two-dimensional displacement output is achieved, avoiding additional errors introduced by distortion and facilitating rapid on-site application.

[0011] In a preferred embodiment, in step S1, the YOLOv8 detection module trains a detection model for each target and filters the detection results based on the confidence of the output detection box, setting a confidence threshold to eliminate false detections.

[0012] The size of the detection frame can be appropriately expanded to ensure that the entire area containing the crosshair target is included, thereby defining the ROI area for corner extraction and improving the accuracy of corner detection.

[0013] In a preferred embodiment, in step S2, preprocessing is used to enhance the contrast of the ROI image, initial corner points are detected using Harris or Shi-Tomasi corner detection algorithms, and corner coordinates are optimized using subpixel thinning algorithms (such as OpenCV's cornerSubPix) to ensure that the accuracy of the extracted corner coordinates reaches the subpixel level.

[0014] In a preferred embodiment, the weight w in step S4 is determined by considering the confidence p of the YOLO detection box and the corner fitting error e, for example, it can be defined as follows: The weighting function is given by f(e), which is inversely proportional to the corner fitting error, and g(p), which is proportional to the confidence of the detection box. This gives greater weight to the corner detection when it is reliable, and more reliance on the center of the YOLO box when the image is blurred or occluded.

[0015] Before merging, outlier removal is performed on the corner coordinate set, that is, outliers that are too far away from the intersection geometry or other corners are removed.

[0016] In a preferred embodiment, the displacement measurement is the relative displacement of the target on the image plane. There is no need to use camera intrinsic parameters for calibration and distortion correction. The two-dimensional displacement of the structure is calculated by directly comparing the pixel coordinate difference of the target optimization center at different times, which directly avoids the error amplification introduced by distortion.

[0017] The technical effects and advantages of this invention are as follows:

[0018] This invention proposes an innovative fusion strategy that breaks through the limitations of traditional single-image measurement processes. By combining real-time target detection with sub-pixel fine measurement, it improves detection efficiency and flexibility in multi-target environments (YOLOv8 achieves 28.6% mAP on the COCO dataset, enabling real-time image processing) while ensuring high measurement accuracy. The weighted fusion algorithm and anomaly removal mechanism further enhance the stability and accuracy of the algorithm in complex scenarios. This method can achieve high-precision displacement measurement in the health monitoring of structures such as dams and bridges without frequent calibration and can operate around the clock, demonstrating significant innovation and practical value. Attached Figure Description

[0019] Figure 1 It is a schematic diagram of the overall structure of the present invention;

[0020] Figure 2 This is a schematic diagram of the actual measured value of the target x according to the present invention;

[0021] Figure 3 This is a schematic diagram of the actual measured value of the target y in this invention. Detailed Implementation

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0023] As attached Figure 1 The method for center displacement measurement that integrates target detection and corner region screening is shown. First, a pre-made circular ring with an embedded cross-shaped target is attached to the surface of the monitored object. Each target is marked with a cross at its center, and the remaining areas are blank contrast colors. Then, a high-resolution industrial camera is used to acquire monitoring images of the structure at different time points. For any frame of the image, the following processing steps are performed:

[0024] Step 1: YOLOv8 Object Detection. The acquired color image is input into a pre-trained YOLOv8 network model. YOLOv8 will output several bounding boxes, each locating the position of a circular or crosshair target, and providing the coordinates of the four vertices (x, y, y). min ,y min ,x max ,y max ) and confidence level p, calculate the center coordinates of each detection box: To improve the inclusiveness of corner detection, the detection box can be appropriately expanded to define a larger ROI region for subsequent processing. Low-confidence boxes can be filtered out by setting a confidence threshold to reduce false detections.

[0025] Step 2: Subpixel corner detection. For the ROI region corresponding to each detection box in Step 1, image preprocessing (such as equalization, filtering, contrast enhancement, etc.) is first performed to improve corner features. Then, the Harris corner detection algorithm or Shi-Tomasi algorithm is used to extract initial corner points. Theoretically, the circular crosshair target should correspond to four corner point positions. The initial corner point coordinates are fed into a subpixel optimization algorithm (such as cornerSubPix) to iteratively refine the corner point positions, achieving subpixel-level coordinate accuracy. Let the obtained four corner point coordinates be (x... i ,y i (i = 1, 2, 3, 4) If more than four corner points are detected, redundant points can be removed based on the relative position of the corner points or a threshold; if fewer than four are detected, the target can be temporarily skipped or only the YOLO center can be used for processing.

[0026] Step 3: Calculate the corner centers. Based on the obtained coordinates of the four corner points, calculate the fitted center of the corners. For example, a simple arithmetic mean can be used: Ideally, the above average value is the coordinate of the intersection of the cross lines. Alternatively, a more robust center can be obtained by fitting a straight line to the line connecting the two sets of opposite corner points and finding the intersection. During the calculation process, it can be checked whether each corner point is geometrically consistent with the cross structure. If a corner point deviates too much (such as the distance from the average center exceeds the threshold), it is regarded as an outlier and removed, and then the calculation is repeated.

[0027] Step 4: Center fusion and optimization to obtain two center positions for each target: YOLO detection box center. and corner fitting center Considering the differences in advantages and disadvantages between the two methods, a weight w is introduced to fuse them, and the center after fusion is defined as follows: The fusion weight w can be jointly determined by the YOLO confidence p and the corner fitting error e. One feasible setting is to set the corner reliability f. 角 =1 / (e+∈), frame reliability f 框 =p, then calculate the normalized weights w = f 角 / (f 角 +f 框 Thus, when the corner fitting error is small, w tends to 1, and the corner center is preferred; if the corner fitting quality is poor or the confidence level p is large, w is appropriately reduced, and more weight is given to the YOLO center. Furthermore, outlier removal is performed before fusion: for example, if... or If the threshold is exceeded, the fusion result is deemed unreliable. In this case, the corner results can be ignored and only the YOLO center can be used, or the corner extraction step can be re-evaluated.

[0028] Step 5: Displacement calculation. For the same target, repeat the above steps in frames at different times to obtain the corresponding optimization center. The center coordinates in the initial frame (reference frame) are marked as The center of a frame at any given time is denoted as The pixel displacements of the target in the horizontal and vertical directions are: Pixel displacement is converted into actual physical displacement through a pre-determined calibration or proportional relationship (the corresponding length of each pixel can be calibrated in one step), thus obtaining a high-precision two-dimensional displacement measurement value. Since this method directly uses the differences in image plane coordinates and the center accuracy after fusion is high, complex distortion correction is not required, and the error is negligible within a small field of view.

[0029] In step S1, the YOLOv8 detection module trains a detection model for each target and filters the detection results based on the confidence of the output detection box, setting a confidence threshold to eliminate false detections.

[0030] The size of the detection frame can be appropriately expanded to ensure that the entire area containing the crosshair target is included, thereby defining the ROI area for corner extraction and improving the accuracy of corner detection.

[0031] In step S2, preprocessing is used to enhance the contrast of the ROI image, the initial corner points are detected using Harris or Shi-Tomasi corner detection algorithms, and the corner coordinates are optimized using subpixel thinning algorithms (such as OpenCV's cornerSubPix) to ensure that the accuracy of the extracted corner coordinates reaches the subpixel level.

[0032] In step S4, the weight w is determined by considering the confidence level p of the YOLO detection box and the corner fitting error e. For example, it can be defined as follows: The weighting function is given by f(e), which is inversely proportional to the corner fitting error, and g(p), which is proportional to the confidence of the detection box. This gives greater weight to the corner detection when it is reliable, and more reliance on the center of the YOLO box when the image is blurred or occluded.

[0033] Before fusion, outlier removal is performed on the corner coordinate set, that is, outliers that are too far away from the intersection geometry or other corners are removed.

[0034] The displacement measurement is the relative displacement of the target on the image plane. It does not require camera intrinsic parameters for calibration and distortion correction. The two-dimensional displacement of the structure is calculated by directly comparing the pixel coordinate differences of the target optimization center at different times, thus avoiding the error amplification introduced by distortion.

[0035] Through the above steps, this invention can achieve stable and accurate displacement measurement of multiple monitoring targets without continuous calibration. In practical applications, the YOLOv8 model and parameters for each monitoring area can be specifically trained and adjusted to adapt to targets of different sizes and materials. Simultaneously, the corner detection algorithm can select appropriate thresholds and window sizes based on the target design. Experimental results show that this method can maintain sub-pixel-level displacement measurement accuracy even under conditions of slight image blur, illumination changes, and environmental clutter interference, demonstrating significant advantages over traditional single-technology solutions.

[0036] In summary, this invention provides an innovative fusion strategy that achieves high-precision, calibration-free structural deformation and displacement measurement in complex environments by filtering corner regions through target detection and combining it with a weighted, optimized center fusion algorithm. This technology is characterized by its simplicity, real-time performance, and robustness, making it highly suitable for long-term online displacement monitoring needs in structural health monitoring fields such as dams and bridges.

[0037] The high-precision center position optimization displacement measurement device included in the above-mentioned center displacement measurement method for target detection and corner region filtering fusion includes: an image acquisition module, a YOLOv8 target detection module, a corner detection module, a center optimization fusion module, and a displacement calculation module; the image acquisition module is used to acquire a structural deformation image containing multiple circular rings embedded with cross targets; the YOLOv8 target detection module is used to perform multi-target detection on the image and output the detection box and the center coordinates of the detection box for each target; the corner detection module is used to perform sub-pixel level corner extraction within the ROI region corresponding to each detection box, obtain the corner coordinates of the cross target, and calculate the corner center; the center... The optimization and fusion module compares the center of the detection box with the center of the corner point, assigns weights based on confidence and fitting error, and fuses the two center positions to obtain the optimized center coordinates. The displacement calculation module compares the position changes of the optimized center in different frame images and outputs the two-dimensional displacement measurement results of the target. The YOLOv8 target detection module realizes real-time detection of multiple scales and categories. The corner detection module may further include an image preprocessing submodule and a corner refinement submodule to improve the stability and accuracy of corner extraction. The center optimization and fusion module includes an outlier filtering unit to identify and remove outliers in the detection box or corner calculation, thereby improving the robustness of the overall algorithm.

[0038] One of the center displacement measurement methods that integrates target detection and corner region screening also includes a computer-readable storage medium that stores a program that causes the processor to execute the above five steps S1 to S5. When the program is executed, the processor completes each of the above steps in sequence, thereby realizing the high-precision displacement measurement function.

[0039] Finally, a few points should be explained: First, in the description of this application, it should be noted that, unless otherwise specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense, and may refer to mechanical or electrical connections, internal communication between two components, or direct connection. "Up," "down," "left," and "right" are only used to indicate relative positional relationships. When the absolute positions of the objects being described change, the relative positional relationships may also change.

[0040] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0041] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for measuring center displacement that integrates target detection and corner region screening, characterized in that, Includes the following steps: S1. Use the YOLOv8 network to perform target detection on a structured image containing multiple targets, output the bounding box coordinates of each target, and calculate the center coordinates of each bounding box. S2. For each detection box obtained in step S1, perform sub-pixel level corner detection within the image region of interest defined by the box, and extract the coordinates (x, y, y) of the four corner points of the crosshair target. i ,y i ), i = 1, 2, 3, 4; S3. Obtain the corner fitting center of the crosshair target by performing geometric fitting or averaging on the corner coordinates. S4. Based on the center coordinates of the detection box in step S1 Fitting the center of the crosshair target in step S3 To address the positional discrepancies between the two center positions, an error-weighted fusion algorithm is constructed. This algorithm adaptively assigns weights w to fuse the two center positions and calculates the optimized center position. S5. Using the displacement change of the optimized target center in consecutive image frames, the two-dimensional displacement measurement value of the structural target is obtained. The weight w in step S4 is determined by considering the confidence p of the YOLO detection box and the corner fitting error e, and can be defined as follows: The weighting function is given by f(e), which is inversely proportional to the corner fitting error, and g(p), which is proportional to the confidence of the detection box. It gives greater weight when the corner detection is reliable, and depends more on the center of the YOLO box when the image is blurred or occluded. Before fusion, outlier removal is performed on the corner coordinate set, that is, outliers that are too far away from the intersection geometry or other corners are removed. Displacement measurement is the relative displacement of the target on the image plane. It does not require camera intrinsic parameters for calibration and distortion correction. The two-dimensional displacement of the structure is calculated by directly comparing the pixel coordinate differences of the target optimization center at different times, thus avoiding the amplification of errors introduced by distortion.

2. The center displacement measurement method for target detection and corner region screening fusion according to claim 1, characterized in that: In step S1, the YOLOv8 detection module trains a detection model for each target, filters the detection results based on the confidence of the output detection box, and sets a confidence threshold to eliminate false detections. The size of the detection frame can be expanded to ensure that the entire area containing the crosshair target is included, defining the ROI area for corner point extraction.

3. The center displacement measurement method for the fusion of target detection and corner region screening according to claim 1, characterized in that: In step S2, preprocessing is used to enhance the contrast of the ROI image, the initial corner points are detected using Harris or Shi-Tomasi corner detection algorithms, and the corner coordinates are optimized using a sub-pixel thinning algorithm to ensure that the accuracy of the extracted corner coordinates reaches the sub-pixel level.

Images