Target positioning and identification method for high-precision photogrammetry of a pipe rack

Abstract

The application discloses a target positioning and identification method for high-precision photogrammetry of a jacket, and the method comprises the following steps: obtaining a plurality of images of the jacket through photogrammetry; drawing a histogram of the images; judging imaging conditions; coarsely detecting possible circular target areas by using a real-time target detection algorithm; detecting ellipses under different imaging conditions; positioning a circle center; and decoding a coding mark. The target positioning and identification method for high-precision photogrammetry of the jacket can effectively complete high-precision measurement work of a deepwater super-large jacket, and effectively improves the measurement precision of the deepwater super-large jacket.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of target detection of offshore oil engineering, and particularly relates to a target positioning and identification method for high-precision photogrammetry of a jacket. BACKGROUND

[0002] In the completion measurement of a deepwater super-large jacket, targets are uniformly pasted on the surface of the jacket at certain intervals before the photogrammetry is implemented. After the completion of the jacket, the jacket with the pasted targets is usually placed on a wharf, and an image of the jacket is acquired by using a camera to shoot the jacket with the pasted targets. The high-precision measurement of key points of the jacket is realized through target positioning and identification in the image.

[0003] However, in the process of implementing the photogrammetry, different air humidity and temperature will produce different atmospheric refraction. In addition, due to factors such as light and wind strength at the seaside, the image quality and measurement by the camera will be negatively affected.

[0004] Therefore, it is urgent to design a target positioning and identification method for high-precision photogrammetry of a jacket to solve the above-mentioned problems in the high-precision measurement of the jacket. SUMMARY

[0005] To solve the technical problem that different air humidity and temperature will produce different atmospheric refraction in the process of implementing the photogrammetry, and in addition, due to factors such as light and wind strength at the seaside, the image quality and measurement by the camera will be negatively affected, a target positioning and identification method for high-precision photogrammetry of a jacket is provided to solve the problem of high-precision measurement of the jacket.

[0006] To achieve the above-mentioned purpose, the specific technical scheme of the target positioning and identification method for high-precision photogrammetry of a jacket of the present application is as follows:

[0007] A target positioning and identification method for high-precision photogrammetry of a jacket, comprising the following steps: S1, obtaining a plurality of images of the jacket by photogrammetry;

[0008] S2, drawing a histogram of the image;

[0009] S3, judging the imaging condition;

[0010] S4, coarsely detecting possible circular target regions by using a real-time target detection algorithm;

[0011] S5, oval detection under different imaging conditions;

[0012] S6, positioning the center of the circle;

[0013] S7, decoding the coding mark.

[0014] Further, in S1, several images of the jacket are taken from different angles and positions using a UAV and are cropped into slices.

[0015] Further, in S2, a histogram of each image is calculated and plotted according to the images taken by the UAV, with the horizontal axis representing the brightness value of 0-255 and the vertical axis representing the number of pixels corresponding to the brightness in the photo.

[0016] Further, in S3, the imaging conditions of the image are determined to be normal imaging conditions or abnormal imaging conditions according to the histogram of the image; the normal imaging conditions refer to good lighting conditions and good overall contrast of the image, and the abnormal imaging conditions refer to the influence of the lighting and wind strength at the seaside on the imaging.

[0017] Further, in S4, the jacket target data set under the above two kinds of scenes is labeled and constructed respectively, and the two data sets are expanded using image enhancement technology; the data set is divided, and the corresponding training set is sent into the real-time target detection algorithm for training, and the weight of the last training of the model and the weight when the training effect is best are saved.

[0018] Further, the test image is sent into the trained model for target mark detection to obtain the confidence score and the position information of the corresponding target box; the detected target region is cropped from the image, and the position information is recorded.

[0019] Further, in S5, the photo taken under normal imaging conditions can be fitted to the boundary part of the ellipse according to the least square principle.

[0020] Further, in S5, the photo taken under abnormal imaging conditions depends on the flow conservation characteristics, and the gradient quantity entering through the arc line of the outermost ellipse must be equal to the gradient quantity flowing out through the arc line of the innermost ellipse, so as to realize the ellipse detection.

[0021] Further, in S6, according to the principle that the intersection point of two straight lines after the projective transformation is still the intersection point of the two straight lines after the transformation, the center of the mark point is located using the outermost circular arc segment.

[0022] Further, in S7, the region of interest where the coding mark point is located is selected, the annular band of the region of interest is extracted, the number represented by each annular band is calculated, the relative position relationship between the annular bands is calculated, the binary code value of the coding mark point is calculated, and the binary code value is converted into a decimal code number.

[0023] The target positioning and recognition method for high-precision photogrammetry of the jacket has the following advantages:

[0024] Through the target positioning and identification method, the high-precision measurement work of the deepwater super large jacket can be effectively completed, and the measurement precision of the deepwater super large jacket completion measurement is effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The whole flow chart of the target positioning and identification method for the high-precision photogrammetric measurement of the jacket is shown in the figure.

[0026] Figure 2 The target example diagram of the target positioning and identification method for the high-precision photogrammetric measurement of the jacket is shown in the figure.

[0027] Figure 3 The circular reference coding mark with a size of 50mm*50mm in the target positioning and identification method for the high-precision photogrammetric measurement of the jacket is shown in the figure.

[0028] Figure 4 The setting of the 50mm*50mm target at the center of the guide pipe in the target positioning and identification method for the high-precision photogrammetric measurement of the jacket is shown in the figure.

[0029] Figure 5 The circular reference coding mark with a size of 200mm*200mm in the target positioning and identification method for the high-precision photogrammetric measurement of the jacket is shown in the figure.

[0030] Figure 6 The elliptical transformation diagram of the target positioning and identification method for the high-precision photogrammetric measurement of the jacket is shown in the figure.

[0031] Figure 7 The overlapping area example diagram of the circular target in the target positioning and identification method for the high-precision photogrammetric measurement of the jacket is shown in the figure.

[0032] Mark description in the figure:

[0033] 1, 50mm*50mm coding mark; 2, 200mm*200mm coding mark; 3, steel bar; 4, guide pipe ring. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0035] Those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the claims, any of the claimed embodiments can be used in any combination.

[0036] The following is a reference to the appendix. Figure 1 To be continued Figure 7 This invention describes a target positioning and identification method for high-precision photogrammetry of catheter stents.

[0037] like Figure 1 As shown, the target positioning and identification method for high-precision photogrammetry of duct stents in this invention includes the following steps:

[0038] S1. Obtain several images of the guide frame through photogrammetry;

[0039] S2. Draw a histogram of the image;

[0040] S3. Determine the imaging conditions;

[0041] S4. Use a real-time target detection algorithm to coarsely detect possible circular target areas;

[0042] S5. Ellipse detection under different imaging conditions;

[0043] S6. Locate the center of the circle;

[0044] S7, Encoding flag decoding.

[0045] By using target positioning and identification methods, high-precision measurement of deep-water ultra-large jacket structures can be effectively completed, thus significantly improving the measurement accuracy of as-built measurements of deep-water ultra-large jacket structures.

[0046] Furthermore, such as Figures 1 to 5 As shown, in S1, several images of the pipe rack are taken using a drone from different angles and orientations and then cropped into slices; in S2, based on the images acquired by the drone, a histogram of each image is calculated and plotted, with the horizontal axis representing the brightness value from 0 to 255 and the vertical axis representing the number of pixels corresponding to the brightness in the photo; in S3, based on the image histogram, the imaging conditions of the image are determined to be divided into normal imaging conditions and abnormal imaging conditions; normal imaging conditions refer to good lighting conditions and good overall image contrast, while abnormal imaging conditions refer to situations where the intensity of light and wind at the seaside affects the imaging.

[0047] In this embodiment, S1 obtains several images of the guide pipe frame, including: designing two-diameter coded marks for the target, including a coded mark 1 example with a size of 50mm*50mm, which is mainly used for recognition in space-limited positions, and is attached at the center of the guide pipe circle.

[0048] Preferably, two steel bars 3 with a width greater than 5cm are provided above the guide pipe, the two steel bars are the diameter of a closed circle, and both pass through the center of the guide pipe circle 4; and a coded mark 2 example with a size of 200mm*200mm can be attached to any position of the object to be measured for photogrammetry.

[0049] Such a design not only takes into account the space requirements of different scenes, but also ensures the effectiveness of the object to be measured at various positions. From different angles and positions, the unmanned aerial vehicle takes several images of the guide pipe frame and crops them into slices with a size of 768*768 pixels.

[0050] S2 includes calculating and drawing a histogram of each image according to the images obtained by the unmanned aerial vehicle, the horizontal axis represents the brightness value of 0-255, and the vertical axis represents the number of pixels corresponding to the brightness in the photo.

[0051] S3 judges the imaging conditions of the image according to the histogram of the image, which is divided into two cases: the first case is that the shape of the image histogram is high in the middle and low at both ends, indicating that the lighting condition is good, and the overall contrast of the image is good. The histogram of the photo of the guide pipe frame taken by the camera in the normal scene should present this distribution; the second case is that the histogram distribution is concentrated at a certain gray value, and a very high peak appears at a certain place, while the peak value is very low or even 0, for example, when the data in the histogram is mostly concentrated near the right edge value, i.e. the gray level of the image is concentrated in the high brightness range, indicating that the image is overexposed.

[0052] When the data in the histogram is mostly concentrated near the left edge value, i.e. the gray level of the image is concentrated in the low brightness range, it indicates that the image is underexposed. In the process of implementing photogrammetry, different air humidity and temperature will produce different atmospheric refraction, causing image distortion; in addition, due to factors such as light and wind strength at the seaside, we classify these scenes as abnormal scenes, and the photos taken in these scenes may be blurred or have weak light, etc. Its histogram should present the shape described, and the image quality is poor.

[0053] Further, as Figure 1As shown, in S4, the catheter support target data sets in the above two scenarios are respectively labeled and constructed, and the two data sets are expanded using image enhancement techniques; the data sets are divided, and the corresponding training sets are respectively sent into the real-time target detection algorithm for training, and the weight of the last training of the model and the weight when the training effect is best are saved; the test image is sent into the trained model for target mark detection to obtain the confidence score and the position information of the corresponding target box; the detected target region is cropped from the image, and the position information is recorded.

[0054] In this embodiment, YOLOv5 network is used to coarsely detect possible circular target regions, including: labeling and constructing the catheter support target data sets in the above two scenarios respectively, and converting them into yolo format; using mosaic, random color change, random saturation change and other image enhancement techniques to expand the two data sets.

[0055] The two catheter support target data sets are respectively divided into training sets and test sets according to 8:2, the picture input size is defined as 768*768, the model training rounds and batchsize are defined, the corresponding training sets are respectively sent into the YOLOv5 network for training, and the weight is updated, and the weight of the last training of the model and the weight when the training effect is best are saved.

[0056] Finally, the test image is sent into the trained model for target mark detection to obtain the confidence score and the position information of the corresponding target box; according to the target box information predicted by the YOLOv5 network, the detected target region is cropped from the image, and the position information is recorded.

[0057] Further, as shown in Figure 1 , Figure 6 and Figure 7 , in S5, the photos taken under normal imaging conditions can fit the boundary part of the ellipse according to the least square principle; in S5, the photos taken under non-normal imaging conditions rely on the flow conservation characteristics, and the gradient quantity entering through the arc line of the outermost ellipse must be equal to the gradient quantity flowing out through the arc line of the innermost ellipse, so as to realize ellipse detection.

[0058] In S6, according to the principle that the intersection point of two straight lines after projection transformation is still the intersection point of the two straight lines after transformation, the center of the mark point is located using the outermost circular arc segment; in S7, the region of interest where the coding mark point is located is selected, the annular band of the region of interest is extracted, the number represented by each annular band is calculated, the relative position relationship between the annular bands is calculated, the binary code value of the coding mark point is calculated, and the binary code value is converted into a decimal code number.

[0059] In this embodiment, the photo taken under normal imaging conditions, first to cut out the possible circular target area of the pixel by pixel 8-neighborhood analysis, and then according to the center point around the 8 adjacent pixels in the direction of the gray difference to calculate the precise edge point A' offset in that direction, and finally according to the least square principle can be fitted out the boundary part of the ellipse, assuming the ellipse center is (x0, y0), the long half axis is a, the short half axis is b, the angle between the long half axis and the x axis is θ, the coordinate system is translated and rotated to form a positive ellipse with (x0, y0) as the center and the angle between the long half axis and the x axis as 0, then the point (x, y) on the ellipse is transformed into (xr, yr).

[0060]

[0061]

[0062] The transformed point (xr, yr) satisfies the equation:

[0063] =1

[0064] The unknowns are xr, yr, a, and b. The ellipse equation is nonlinear for the unknowns, and linearization is performed to obtain the error equation. The initial values of the unknowns (x0, y0), a, and b are determined by the edge points obtained from each region. According to the least square principle, the unknowns are iterated until the correction number is less than the set threshold. Let the function be:

[0065] -1

[0066] Let v1 be 4 / a2; v2 be 16 / a2. Substitute all the edge points of the region of interest into the function f, if the number of points whose absolute value of f is less than the set threshold v1 exceeds 95% of the total number of points, and the maximum absolute value of f does not exceed the set threshold v2, that is, most of the edge points are within the threshold v1 of the fitted ellipse, and the farthest point does not exceed the threshold v2, if the above two conditions are met, it is considered that the edge points constitute an ellipse.

[0067] The photo taken under abnormal imaging conditions uses the 45-degree overlapping region of the three-layer circular arc in the target, relies on the flow conservation characteristics, and realizes ellipse detection by the gradient quantity entering the outermost elliptical arc being equal to the gradient quantity flowing out of the innermost elliptical arc.

[0068] ​​​​First, pixel selection and region pairing are performed to create paths in the image in the form of 6 connected edge pixel sequences (3 overlapping circular arc segments), such that the direction of the path segment starting from pixel p is given by the image gradient p. Then, the pixel with the most occurrences is selected as the path endpoint by a voting procedure.

[0069] The selected endpoint is grouped into the inner region, forming a convex arc approximation polygon, and the associated starting point is grouped into the outer region. A voting procedure is applied to the real image, and then region pair selection is performed, only selecting region pairs that satisfy the "conservative constraint", which basically guarantees that the gradient magnitude inflow through the outer region must equal the gradient magnitude outflow through the inner region.

[0070] Preferably, in S6, the principle that the intersection of two straight lines after a projective transformation is still the intersection of the two transformed straight lines is used to determine the center of the marker point, and the outermost circular arc segment is used for circle center positioning. First, according to the extracted possible circular target region, the centroid of the innermost elliptical ring is extracted, and a straight line is connected between the centroid of the innermost elliptical ring and the centroid of the outermost encoding ring belt connected domain.

[0071] The distance of all edge points on the connected domain to the straight line is calculated, a distance threshold is set, and the number of points less than the distance threshold is counted; the straight line is rotated 180 degrees in the direction of the straight line around the circle center with a fixed step size, and the number of points less than the distance threshold is counted at each rotation, and the number of points less than the distance threshold is counted at each rotation. The number of points less than the distance threshold is counted at each rotation, and the number of points less than the distance threshold is counted at each rotation. The number of points less than the distance threshold is counted at each rotation, and the number of points less than the distance threshold is counted at each rotation.

[0072] Finally, the same operation is performed on each connected domain on the encoding ring belt to obtain all straight line edges; then the Gaussian fitting method with good noise resistance and positioning accuracy is selected for sub-pixel edge positioning. In the gradient direction of the image edge, the first derivative of the gray scale distribution is approximately Gaussian, and the center of the Gaussian distribution is the position with the highest edge point accuracy.

[0073] exp[- ]

[0074] In the relationship, x is the coordinate point in the edge gradient direction, y is the first derivative of the gray scale value, μ is the mean of the Gaussian function, σ is the standard deviation of the Gaussian function, and k is the amplitude of the Gaussian function. The sub-pixel position of the edge can be obtained by obtaining the position parameter μ.

[0075] First, straight line fitting is performed on each edge, the gradient direction of the straight line edge is expressed by the normal direction of the straight line edge, and the sub-pixel edge position is obtained by calculating the gray difference value of the point along the normal direction and fitting the Gaussian model. Finally, based on the Hough transform idea and the random sample consensus (RANSAC) algorithm for center coordinate, the obtained sub-pixel edge points are mapped to the parameter space; then all intersection points of the corresponding mapping curve of each straight line point are obtained, in order to reduce the influence of noise points and mis-extracted points on the result, the intersection points are fitted in the parameter space by using the RANSAC algorithm, and the curve mapping to the original space is the center coordinate.

[0076] Preferably, S7 selects the region of interest where the encoding mark point is located, the region contains the encoding ring band information; the ring band of the region of interest is extracted; the number represented by each ring band is calculated; according to the proportion of the center circle radius and the inner and outer diameters of the ring band when the encoding mark point is designed, the proportion of the area of the smallest unit ring band and the area of the center circle is calculated, and the number represented by each ring band is obtained; the relative position relationship between the ring bands is calculated through the included angle between the center of mass of the connected domain and the center line of the center circle; finally, the binary code value of the encoding mark point is calculated and converted into a decimal code number, that is, the decoding is realized.

[0077] The target positioning and identification method based on high-precision photogrammetry of the guide pipe rack obtains a plurality of images of the guide pipe rack through photogrammetry, draws a histogram of the images and judges the imaging condition, uses a real-time target detection algorithm to coarsely detect possible circular target regions, performs elliptical detection under different imaging conditions, positions the center, and decodes the encoding mark. Among them, the imaging condition is divided into normal imaging condition and abnormal imaging condition, the image obtained under the normal imaging condition is detected through edge detection to detect the ellipse, and the image obtained under the abnormal imaging condition is detected through the three-layer overlapping region according to the flow conservation principle to effectively realize the high-precision measurement of the key points of the guide pipe rack.

[0078] Obviously, the above embodiments of the present application are only examples for clearly illustrating the present application, and are not intended to limit the embodiments of the present application. Based on the above description, other different forms of changes or variations can be made by those skilled in the art. Here, all the embodiments are not required to be exhausted. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the claims of the present application.

Claims

1. A target positioning and identification method for high-precision photogrammetry of a pipe rack, characterized in that, The method comprises the following steps: S1, obtaining several images of the jacket through photogrammetry, taking several images of the jacket from different angles and positions using a UAV and cutting them into slices; S2, drawing a histogram of the image, calculating and drawing a histogram of each image according to the image obtained by the UAV, wherein the horizontal axis represents the brightness value of 0-255, and the vertical axis represents the number of pixels corresponding to the brightness in the photo; S3, judging the imaging condition, judging the imaging condition of the image according to the histogram of the image, wherein the normal imaging condition refers to good lighting condition and good overall contrast of the image, and the abnormal imaging condition refers to the situation that the light and wind strength at the seaside will affect the imaging; S4, using a real-time target detection algorithm to coarsely detect possible circular target regions, respectively labeling and constructing the jacket target data set under the above two imaging condition scenarios, and expanding the two data sets using image enhancement technology; Dividing the data set, sending the corresponding training set into the real-time target detection algorithm for training, saving the weight of the last training of the model and the weight when the training effect is best; Sending the test image into the trained model for target mark detection to obtain the confidence score and the position information of the corresponding target box; cutting the detected target region from the image and recording the position information; S5, ellipse detection under different imaging conditions; under the normal imaging condition, fitting the boundary part of the ellipse according to the least square principle; under the abnormal imaging condition, relying on the flow conservation characteristics, the gradient quantity entering through the arc line of the outermost ellipse must be equal to the gradient quantity flowing out through the arc line of the innermost ellipse, to realize ellipse detection; S6, positioning the center; according to the principle that the intersection of two straight lines is still the intersection of the two straight lines after the projective transformation, the center of the mark point is positioned using the outermost circular arc segment; S7, encoding mark decoding; selecting the region of interest where the encoding mark point is located, extracting the annular band of the region of interest, calculating the number represented by each annular band, calculating the relative position relationship between the annular bands, calculating the binary code value of the encoding mark point, and converting it into a decimal code number.

Images