Angular point detection method for phase consistency enhancement and adaptive geometric fitting

Abstract

The invention discloses a phase consistency enhancement and adaptive geometric fitting corner detection method. The method comprises the following steps: firstly, carrying out adaptive histogram equalization CLAHE contrast enhancement on a cross target area image detected by YOLO; calculating a phase consistency response of the enhanced image to obtain a corner candidate region insensitive to brightness and contrast change; and calculating Harris corner response on the enhanced image, combining the Harris response with the phase consistency response score to construct a comprehensive response diagram, and screening out a corner candidate set from the comprehensive response diagram by adopting adaptive multiple thresholds. According to the method, through combination of YOLO target detection and CLAHE region enhancement, illumination non-uniformity interference such as overexposure and shadow can be effectively suppressed in a target region, a comprehensive response diagram is constructed by adopting phase consistency and Harris response in a combined manner, corner candidates are screened by using adaptive multiple thresholds, and the luminosity invariant feature and gradient significance of corners are considered, so that the accuracy and the robustness of the target region are improved. And the angular point extraction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and image processing, and more specifically, to a corner point detection method combining phase consistency enhancement and adaptive geometric fitting. Background Art

[0002] With the development of health monitoring technology for civil engineering structures such as dams and bridges, displacement measurement methods based on visual sensing have been widely adopted. In practical applications, cross targets are often placed at key locations on the structure, and the targets are photographed by cameras. The coordinates of the target corners are then accurately obtained through target detection and corner extraction methods to calculate the tiny displacements of the structure. In recent years, deep learning target detection algorithms (such as YOLO) have performed well in real-time target recognition and can quickly locate target areas in images.

[0003] However, the current detection obtained by YOLO is a rough target bounding box, which cannot directly provide high-precision target corner coordinates. Therefore, it is usually necessary to further perform fine corner detection and sub-pixel positioning in the detected target area. Existing corner detection operators (such as Harris, Shi-Tomasi, etc.) are mainly based on grayscale gradient information and are sensitive to changes in illumination and contrast. Their performance degrades significantly in complex environments such as overexposure, shadows, raindrops, reflections, and motion blur. For example, when the image area is overexposed or has strong reflections, traditional Harris corner detection often cannot find the accurate corner position; under the interference of raindrops or blur, corner responses are prone to pseudo-points or loss. In addition, the complex background of the image area where the target is located will also introduce noise, affecting the accuracy of corner extraction.

[0004] Therefore, in real environments with complex interference, there is an urgent need for a new sub-pixel corner detection method that can combine depth detection and robust algorithms to improve the accuracy and reliability of target corner extraction. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a corner detection method of phase consistency enhancement and adaptive geometric fitting. By combining YOLO target detection and CLAHE area enhancement, it can effectively suppress the interference of illumination unevenness such as overexposure and shadow in the target area. By jointly constructing a comprehensive response map using phase consistency and Harris response, and using adaptive multi-threshold to screen corner point candidates, the photometric invariant characteristics and gradient significance of the corner points are taken into account, thereby improving the accuracy of corner point extraction to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a corner point detection method with phase consistency enhancement and adaptive geometric fitting, comprising the following steps: S1. First, adaptive histogram equalization (CLAHE) contrast enhancement is performed on the cross target area image detected by YOLO to improve the local contrast of the target area and suppress brightness unevenness. S2. Calculate the phase consistency response of the enhanced image and use a multi-scale Log-Gabor filter to extract brightness-insensitive structural features to obtain corner candidate regions that are insensitive to brightness and contrast changes; S3. Calculating Harris corner response on the enhanced image, jointly constructing a comprehensive response map by combining the Harris response with the phase consistency response score, and using adaptive multi-threshold to screen out a set of corner candidate points from the comprehensive response map; S4. Using the geometric constraints of the cross target structure, i.e., corner point pairs are orthogonal and the length of the connecting line is substantially equal, and combining the random sampling consistency RANSAC algorithm, outliers that do not meet the geometric constraints are removed from the corner point candidate set, and four corner points that meet the cross structure are extracted; S5. Perform weighted quadratic surface fitting on the grayscale image of the local neighborhood where the four corner points are extracted to achieve sub-pixel corner point positioning and improve the accuracy of the corner point coordinates; S6. Sort the extracted corner points according to the comprehensive response strength and fitting confidence of the corner points, and confirm the final four corner points.

[0007] In a preferred embodiment, when CLAHE contrast enhancement is performed on the detection frame area in step S1, the detection frame area is divided into several 8×8 grid sub-areas of equal size, and local histogram equalization is performed on each sub-area to suppress illumination non-uniformity, and a contrast-limited distribution is used to avoid noise amplification.

[0008] In a preferred embodiment, the phase consistency response in step S2 is calculated using a multi-scale Log-Gabor filter group, and the corner significance is determined by analyzing the consistency of the phase information of the image at different scales and directions, thereby improving the robustness of feature detection under complex lighting conditions.

[0009] In a preferred embodiment, the adaptive multi-threshold in step S3 is determined by Otsu's method or a dynamic threshold method based on response statistics to adapt to changes in corner point response distribution in different scenarios.

[0010] In a preferred embodiment, the RANSAC algorithm in step S4 is iterated at least 1000 times, and four corner points are randomly selected in each iteration to fit the cross structure model to maximize the number of points that meet the geometric constraints, ensuring that the corner points can still be correctly extracted when there is noise and false detection.

[0011] In a preferred embodiment, when weighted quadratic surface fitting is performed on the corner points in step S5, the weight is related to the local gradient size or the comprehensive response intensity, so that the fitting process is more sensitive to the grayscale changes at the corner points, thereby improving the sub-pixel positioning accuracy.

[0012] In a preferred embodiment, the comprehensive evaluation index used for sorting corner points in step S6 is composed of the weighted sum of the comprehensive response value of the corner point and the fitting residual. Corner points with higher comprehensive response and smaller fitting error have higher priority to improve the reliability and accuracy of corner point selection.

[0013] Technical effects and advantages of the present invention: By combining YOLO target detection with CLAHE area enhancement, the algorithm can effectively suppress interference from illumination unevenness such as overexposure and shadows in the target area, thereby improving the robustness of corner detection. The present invention adopts phase consistency and Harris response to jointly construct a comprehensive response map, and uses adaptive multi-threshold to screen corner point candidates, taking into account the photometric invariance characteristics and gradient significance of corner points, thereby improving the accuracy of corner point extraction. The present invention eliminates false positives by introducing the geometric constraints of a cross target and the RANSAC algorithm, ensuring that the four corner points conform to the preset geometric relationship; combined with weighted quadratic surface fitting, sub-pixel positioning is achieved, greatly improving the corner point positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a flowchart of the overall corner point detection method of the present invention; Figure 2 is a structural block diagram of the displacement measurement system of the present invention; Figure 3 This is a schematic diagram of extracting four corner points through geometric constraints and RANSAC screening in the present invention. DETAILED DESCRIPTION

[0015] 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.

[0016] As attached Figures 1 to 3 A corner detection method using phase consistency enhancement and adaptive geometric fitting is shown, which is applicable to a cross target image in a YOLO-based displacement measurement system and includes the following steps: S1. First, perform adaptive histogram equalization and CLAHE contrast enhancement on the cross target area image detected by YOLO; S2. Calculate the phase consistency response of the enhanced image to obtain corner candidate regions that are insensitive to brightness and contrast changes; S3. Calculating Harris corner response on the enhanced image, jointly constructing a comprehensive response map by combining the Harris response with the phase consistency response score, and using adaptive multi-threshold to screen out a set of corner candidate points from the comprehensive response map; S4. Using the geometric constraints of the cross target structure, i.e., corner point pairs are orthogonal and the length of the connecting line is substantially equal, and combining the random sampling consistency RANSAC algorithm, outliers that do not meet the geometric constraints are removed from the corner point candidate set, and four corner points that meet the cross structure are extracted; S5. Perform weighted quadratic surface fitting on the grayscale image of the local neighborhood where the four extracted corner points are located to achieve sub-pixel corner point positioning; S6. Sort the extracted corner points according to the comprehensive response strength and fitting confidence of the corner points, and confirm the final four corner points.

[0017] The corner detection method of phase consistency enhancement and adaptive geometric fitting includes a robust sub-pixel corner detection device, which includes: YOLO detection unit, used to detect the cross target in the image and output the target detection frame area; A CLAHE enhancement unit, which performs adaptive contrast enhancement on the image in the detection frame area; a phase consistency calculation unit, which calculates a phase consistency response in the enhanced image; Harris response calculation unit, which calculates Harris corner response in the enhanced image; The comprehensive response generation and screening unit combines the phase consistency response and Harris response to generate a comprehensive response map, and uses adaptive multi-threshold to screen out the corner point candidate set; The RANSAC geometric screening unit removes outliers from the corner candidate set and extracts corners that conform to the cross structure based on the geometric constraints of the cross target (mutually orthogonal and equal-length edges); The sub-pixel fitting unit performs weighted quadratic surface fitting on the grayscale image of the local neighborhood where the extracted corner points are located to achieve sub-pixel positioning; The sorting and confirmation unit sorts the extracted corner points according to the comprehensive response strength and fitting confidence of the corner points and confirms the final four corner points.

[0018] Wherein, step S1 includes a YOLO-based cross target displacement measurement system for structural health monitoring, which includes: An image acquisition unit, used to obtain real-time images of the monitoring scene; The YOLO detection unit performs target detection on the collected image and locates the cross target, outputting the target detection frame area; A robust sub-pixel corner detection device is used to perform CLAHE enhancement, phase consistency calculation, comprehensive response screening, RANSAC geometric screening and sub-pixel fitting on the image of the detection frame area to extract the four corner points of the cross target; The displacement calculation unit calculates the displacement change of the dam or bridge structure according to the positions of the four extracted corner points.

[0019] Among them, the YOLO-based target detection unit processes the captured image, quickly identifies and locates the cross target, outputs the detection frame coordinates of the target (i.e., ROI area), and uses the detection frame area as the input for subsequent fine corner point detection.

[0020] For the image area within the YOLO detection frame, the robust sub-pixel corner detection device of the present invention performs the following steps: (1) CLAHE adaptive enhancement: The contrast-limited adaptive histogram equalization (CLAHE) algorithm is applied to the detection frame area image for enhancement. Specifically, the detection area is divided into multiple sub-blocks of equal size (such as an 8×8 grid), and local histogram equalization is performed on each sub-block, and its contrast gain is limited to avoid noise amplification and over-enhancement. This processing can improve the local contrast of the target area, balance the brightness distribution, and make the edge and corner features more prominent; (2) Phase consistency response calculation: The phase consistency response map is calculated on the enhanced image. Specifically, the enhanced image is filtered using a Log-Gabor filter bank at multiple scales and directions to extract phase information. The phase consistency at each scale and in each direction is then normalized and accumulated to obtain the final phase consistency response map. The phase consistency response emphasizes the structural features in the image and is insensitive to grayscale and contrast changes. Therefore, it can reliably mark potential corner areas under complex lighting conditions. (3) Harris response and comprehensive response map construction: At the same time, the Harris operator is used to calculate the traditional corner response on the enhanced image. The Harris operator generates a corner response value for each pixel by analyzing the change of the image grayscale gradient in the local window, reflecting the significance of its surrounding gradient. The Harris response is then combined with the phase consistency response obtained in step 2 according to a certain weight (for example, weighted addition) to generate a comprehensive response map. This comprehensive response map combines the brightness invariance feature and gradient significance, taking into account the detection ability of different corner characteristics; (4) Adaptive threshold screening of candidate corner points: Multi-threshold segmentation is performed on the integrated response map to screen out the candidate set of corner points. The Otsu algorithm or the dynamic threshold method based on response distribution statistics can be used to determine the appropriate threshold. The peak area in the integrated response map that is greater than the threshold is used as the candidate corner point. Multi-threshold processing can take into account the characteristics of corner points at different intensity levels and ensure the integrity of the candidate set. (5) Geometric constraints and RANSAC to eliminate abnormal corner points: The inherent geometric structure characteristics of the cross target, that is, the lines connecting the four corner points are mutually orthogonal and have basically equal lengths, are used to screen the candidate corner points. Specifically, the random sampling consensus (RANSAC) algorithm is used: four points are randomly selected from the candidate corner point set, two mutually perpendicular straight lines are constructed and their intersection and length are calculated, and the two line segments are checked to see whether they meet the conditions of relatively close vertical angles and equal lengths. For the corner point group that meets the conditions, the number of inliers supported by it in the candidate set is calculated, and the above process is repeated several times. The group of corner points with the largest inlier support is used as the four corner points that meet the cross structure. This step can effectively eliminate pseudo corner points caused by noise, occlusion or false detection. (6) Sub-pixel positioning: For the four corner points extracted in step 5, grayscale weighted quadratic surface fitting is performed in their local neighborhood (for example, a 5×5 or 7×7 pixel window is taken at the center of each corner point). Specifically, the grayscale value and coordinates in the neighborhood are used to establish a quadratic surface model. , and introduce weights (such as Gaussian weights based on the gradient size or with the center of the corner as the peak) to perform weighted least squares fitting, solve the coefficients of the model, and calculate the first-order partial derivative of the fitted quadratic surface and set it to zero to obtain the extreme points of the quadratic surface, that is, the sub-pixel corner position offset, thereby achieving sub-pixel positioning. This fitting method can suppress the influence of noise on positioning and improve positioning accuracy; (7) Corner point sorting and confirmation: The four extracted corner points are sorted according to their comprehensive response strength and fitting confidence. A comprehensive evaluation index can be defined, such as using the weighted sum of the response value of the corner point in the comprehensive response graph and its fitting residual. Corner points with high response and small residual have higher scores. The final corner point set is determined by sorting, and the sub-pixel coordinates of the four corner points are output.

[0021] Through the above steps, the present invention achieves stable extraction and high-precision positioning of cross target corners in complex environments. The target area detected by YOLO is used as the algorithm processing range, which effectively reduces background interference and improves processing efficiency. The combination of CLAHE and phase consistency enhances adaptability to extreme lighting conditions. The fusion of Harris and phase consistency takes into account the characteristics of different types of corner points. Geometric screening and RANSAC eliminate false detection interference. Weighted fitting achieves sub-pixel accuracy. The results show that the algorithm can accurately extract corner points in complex interference scenarios such as overexposure, raindrops, blur, and reflections, significantly improving the displacement measurement accuracy of structural monitoring.

[0022] 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. Secondly: The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to conventional designs. The same embodiment and different embodiments of the present invention may be combined with each other without conflict. Finally: 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 in the scope of protection of the present invention.

Claims

1. A corner point detection method using phase consistency enhancement and adaptive geometric fitting, characterized by: This detection method is applicable to cross-target images in a YOLO-based displacement measurement system and includes the following steps: S1, perform adaptive histogram equalization CLAHE contrast enhancement on the cross target area image detected by YOLO; S2. Calculate the phase consistency response of the enhanced image to obtain corner candidate regions that are insensitive to brightness and contrast changes; S3, calculating the Harris corner response on the enhanced image, jointly constructing a comprehensive response map by combining the Harris response with the phase consistency response score, and screening out a set of corner candidate points from the comprehensive response map using an adaptive multi-threshold; S4. Using the geometric constraints of the cross target structure, i.e., corner point pairs are orthogonal and the length of the connecting line is equal, combined with the random sampling consistency RANSAC algorithm, outliers that do not meet the geometric constraints are removed from the corner point candidate set, and four corner points that meet the cross structure are extracted; S5. Perform weighted quadratic surface fitting on the grayscale image of the local neighborhood where the four extracted corner points are located to achieve sub-pixel corner point positioning; S6. Sort the extracted corner points according to the comprehensive response strength and fitting confidence of the corner points, and confirm the final four corner points.

2. The corner point detection method using phase consistency enhancement and adaptive geometric fitting according to claim 1, characterized in that: When performing CLAHE contrast enhancement on the detection frame area in step S1, the detection frame area is divided into several 8×8 grid sub-regions of equal size. Local histogram equalization is performed on each sub-region to suppress illumination non-uniformity, and contrast-limited distribution is used to avoid noise amplification.

3. The corner point detection method using phase consistency enhancement and adaptive geometric fitting according to claim 1, characterized in that: In step S2, the phase consistency response is calculated using a multi-scale Log-Gabor filter bank, and the corner saliency is determined by analyzing the consistency of the phase information of the image at different scales and directions.

4. The corner point detection method using phase consistency enhancement and adaptive geometric fitting according to claim 1, characterized in that: The adaptive multi-threshold in step S3 is determined by the Otsu method, i.e., the Otsu algorithm or a dynamic threshold method based on response statistics, so as to adapt to the changes in the corner point response distribution in different scenarios.

5. The corner point detection method of phase consistency enhancement and adaptive geometric fitting according to claim 1, characterized in that: The RANSAC algorithm in step S4 is iterated for at least 1000 times, and in each iteration four corner points are randomly selected to fit the cross structure model, so as to meet the number of points of the geometric constraint.

6. The corner point detection method of phase consistency enhancement and adaptive geometric fitting according to claim 1, characterized in that: When weighted quadratic surface fitting is performed on the corner points in step S5, the weight is related to the local gradient size or the comprehensive response strength.

7. The corner point detection method using phase consistency enhancement and adaptive geometric fitting according to claim 1, characterized in that: The comprehensive evaluation index used for sorting the corner points in step S6 is composed of the weighted sum of the comprehensive response value of the corner points and the fitting residual.

Images