JP3728461B2 - Method for automatically processing position correction in image information - Google Patents

Description

  The present invention relates to a method for automatically processing position correction in image information, and in particular, when obtaining distribution information on a subject as image information, the displacement of the subject is detected to automatically correct the position of the image information. Related to technology.

  As a method of measuring the pressure applied to the rocket or aircraft body, place the aircraft model in the wind tunnel, ventilate in the supersonic region, image the light emission state of the pressure-sensitive paint applied to the aircraft surface, and from that image Obtaining pressure distribution information is performed. (See Patent Document 1 and Non-Patent Document 1.) The technique for obtaining pressure distribution data from an image of luminescence intensity performs preprocessing of dark subtraction and averaging on the image acquired as Step 1. In step 2, the ratio Iref / I for each pixel is calculated by dividing the reference image Iref when the ventilation is stopped and the measurement image I in the ventilation state. In step 3, the procedure is used to convert the Iref / I value to pressure using the calibration equation. However, this model has different positions and postures in the wind tunnel when ventilation is conducted in the wind tunnel and when ventilation is stopped. That is, when receiving a strong wind, the model cannot be held in a fixed position and is displaced. This displacement is not limited to a two-dimensional movement corresponding to a parallel movement, and generally includes a rotation component, so that an image of the captured model is accompanied by deformation. Therefore, in order to associate the pressure distribution information shown in the image with the initial image that is not ventilated, it is necessary to correct the displacement, but conventionally, the processing takes a lot of time.

  In the current method, the time required for image processing from the image obtained by imaging to the acquisition of aerodynamic data is as shown in FIG. 6. Filter processing, position correction, division image calculation, pressure / temperature correction, 3D mapping, aerodynamic data Procedures such as acquisition are taken. The time required for each process is about 2 seconds for the filter process, and the time from the calculation of the divided image to the acquisition of the aerodynamic data is 10 seconds to 1 minute. However, the position correction is performed manually using affine transformation. It is a situation that takes time. Incidentally, in this position correction, the number of markers to be attached to the surface of the body is set to 30, and two upper and lower surfaces are required. If the number of experimental cases is set to 20, the position information is 30 × 2 × 20 = 1200 pieces. If the processing time is 10 seconds per point, it takes 200 minutes as a whole.

To detect the movement of the airframe model, a plurality of marks are attached to the surface of the model, and the movement of the model is detected from the displacement to establish a correspondence relationship. When the image in the wind tunnel before ventilation is the reference image, the arbitrary point is (x _i , y _i ), and the position of the corresponding point in the measurement image after ventilation is (x _i ′, y _i ′). The affine transformation formula holds.
x _i = a ₁ + a ₂ x _i '+ a ₃ y _i ' + a ₄ x _i ' ² + a ₅ x _i ' y _i '+ a ₆ y _i ' ²
y _i = b ₁ + b ₂ x _i '+ b ₃ y _i ' + b ₄ x _i ' ² + b ₅ x _i ' y _i '+ b ₆ y _i ' ²
If the correspondence between the markers in the reference image and the measurement image is known, the coefficient of the affine transformation equation can be obtained from the position information in each image. As the number of markers increases, more accurate coefficients can be obtained. However, since manual correspondence between marks is too time-consuming, automation is strongly desired.
JP, 2001-249076, A September 14, 2001 publication "Highly functional pressure sensitive paint and element for oxygen concentration measurement" Yusuke Asai, “Measurement technology of pressure distribution with pressure-sensitive paint”, Visualization information, published by Japan Visualization Society Vol.18 No.69 April 1998

  The problem to be solved by the present invention is to provide a technique that can automatically and quickly take correspondence between the same positions on a subject between two images obtained by imaging the same subject. An object of the present invention is to provide a method capable of automatically and quickly correcting the position of a measurement image when measuring a pressure distribution acting on the surface of a coated model from a luminescence image.

The method of automatically performing position correction according to the present invention includes a step of recognizing a pattern in which the mark group is connected with a single stroke in a reference image of a subject having a number of mark points on a surface, and a mark in a measurement image. Select the pattern most similar to the previous recognition pattern from many single-stroke patterns that connect groups and identify the corresponding marks, and the affine transformation formula from the coordinate information in both images of each corresponding mark And a step of calculating a correction position by applying the obtained coefficient to points on the entire surface of the subject.
The method of automatically performing position correction according to the present invention is to specify one pattern in which marks are connected by a single stroke by an optimization method on the condition of the shortest distance.
Further, the optimization method of the present invention employs a genetic algorithm as a method for selecting a pattern most similar to the previous recognition pattern from among a large number of one-stroke writing patterns that connect mark groups in a measurement image.

The method of automatically processing the position correction of the measurement image of the present invention comprises the step of recognizing one pattern in which the mark group is connected with a single stroke in a reference image of a subject having a number of mark points on the surface; The step of selecting the pattern most similar to the previous recognition pattern from a number of one-stroke patterns connecting the mark groups in the measurement image and identifying the corresponding marks, and the coordinate information of both images of the corresponding marks The step of calculating the coefficient of the affine transformation equation from the above and the step of calculating the correction position by applying it to the points on the entire surface of the subject using the calculated coefficient is not involved in pattern recognition. It is possible to identify the corresponding points of both images, find the affine transformation coefficient from the coordinate information, and correct the position of the measured image It can be automatically calculated.
The method of automatically processing the position correction of the measurement image of the present invention is a simple and easy-to-determine pattern by specifying one pattern in which a mark group is connected with a single stroke by an optimization method on the condition of the shortest distance. Since the corresponding pattern can be obtained directly in many cases under the same conditions for selecting the pattern of the measurement image, the correspondence of the same mark point in both images can be taken quickly.
Furthermore, the method of automatically processing the position correction of the measurement image of the present invention employs a genetic algorithm as an optimization method, so that it is possible to specify an appropriate pattern and a corresponding mark point in a short time. It can be carried out. Moreover, even if the number of marks increases, the calculation is effective within a practical time.

  The method of automatically processing the position correction of the measurement image according to the present invention is based on the luminescence image in which the pressure-sensitive paint applied to the model surface of the rocket or aircraft body installed in the wind tunnel emits light under ventilation. When applied to a wind tunnel test system that measures the distribution, the position correction processing, which has been a bottleneck for the time and work load, can now be performed automatically and quickly.

  In a typical embodiment of the present invention, a rocket or aircraft model (a triangular wing shown in the figure) formed by applying pressure-sensitive paint in a wind tunnel of an optical pressure distribution observation apparatus as shown in FIG. Then, an excitation light source and an observation CCD camera are arranged through the observation window, the ventilation state is set, a luminescence image of the model surface at that time is taken, and the position correction of the acquired measurement image is automatically performed by a computer Are to be processed. The excitation light source is a high-stability xenon lamp (300 W) driven by a constant current. A quartz optical fiber is connected to this, and an optical system such as a condensing lens and a band-pass filter of the wavelength of the excitation light are attached to the tip. In order to prevent the paint from deteriorating, a remote shutter is provided at the exit of the light source. For example, a cooled CCD camera having a resolution of 14 bits and a total number of pixels of 1000 × 1018 is adopted as a CCD camera used for capturing a luminescence image. A band-pass filter corresponding to the emission wavelength of the pressure-sensitive paint and a filter for preventing heat rays are attached to the camera lens.

  A plurality of mark points are provided on the surface of the model and installed in the wind tunnel to capture a reference image Iref (a in FIG. 2) when the ventilation is stopped. Next, the air volume set in the wind tunnel is blown, and a measurement image I (b in FIG. 2) in that state is captured. In order to normalize the detected pressure in the measurement image, a ratio Iref / I with the light emission state in the reference image is calculated. At this time, since the position of the model in the reference image and the position of the model in the measurement image are displaced, it is necessary to correct the position by calculating the amount of movement from the positional deviation between the images of the marks. A division image obtained by performing Iref / I calculation without adding this position correction is shown in FIG. Although it is not a color diagram, it is difficult to understand, but there are black discontinuities in the top and bottom contours of the model. On the other hand, an Iref / I division image with position correction added is shown in FIG. In this image, there is no unnatural discontinuous area on the model surface, and the overall natural pressure distribution is shown. As described above, it is understood that position correction is essential in the wind tunnel experiment because the model is moved.

  From the captured reference image Iref and the measured image I, by performing threshold processing, etc., a binarized image as shown in FIG. 3a is obtained to extract a mark point, and a marker extracted image as shown in FIG. Store on computer. The problem of searching for a one-stroke pattern that connects the mark groups of the reference image Iref with the shortest distance is given, and a calculation is performed on the computer to generate a solution as shown in FIG. 3c, and the pattern information is stored on the computer. Recognize. Subsequently, a pattern closest to the previously-written one-stroke pattern stored in the one-stroke pattern connecting the mark groups of the measurement image I is selected on the computer. In most cases, a similar pattern can be selected directly by solving the "single-stroke pattern". Since the point corresponding to the bending point of the pattern is the mark point, the correspondence between the same mark points can be taken. However, because the model is not a planar structure but a three-dimensional structure, there are rare cases in which adjacent mark points move back and forth on the image as the model rotates. In such a case, it becomes a geometric contradiction in the process of taking the coordinate correspondence of each mark point. At that time, by exchanging adjacent mark points, the contradiction can be resolved and the correspondence of the same mark can be taken.

The number of one-stroke patterns theoretically exists as many as the factorial of the number of marks. The larger the number of marks, the larger the number, and this work places a burden on the machine. Incidentally, the number of marks is 1.3 × 10 ⁵ , the number of marks 15 is 6.5 × 10 ¹¹ , the number of marks 20 is 1.2 × 10 ¹⁸ , and the number of marks 30 is 2.6 × 10 ^32. . Therefore, in the present invention, a genetic algorithm (abbreviated as GA) is used in the optimization processing technique to find the correct answer, thereby reducing the calculation time. By comparing the two patterns in both images, the correspondence between the mark points that are the inflection points of the pattern can be grasped. Therefore, the position (coordinate) information of each mark point in each image is determined, and the above-described affine transformation is performed. Substituting into the equation, the values of coefficients a ₁ to a ₆ and b ₁ to b ₆ are obtained. Using the coefficients thus obtained, the position of the measurement image I on the model is corrected to the model position on the reference image, and an Iref / I divided image is obtained.

  In the present invention, GA is used in the optimization processing method to search for the corresponding pattern. As the optimization processing method, annealing method (abbreviated as SA) and gradient method (abbreviated as HC). .) Was used for comparison. FIG. 4 shows the comparison result. In the graph shown in FIG. 4A, the horizontal axis represents the number of calculation steps, and the vertical axis represents the absolute error rate. In the GA method, a correct answer is obtained about 25 times, and in the SA method and the HC method, the error rate shows several percent even if the calculation is repeated about 500 times. As is apparent from this graph, the GA method shows a much faster convergence than HC and SA. FIG. 4b shows the number of marks on the horizontal axis and the calculation time on the vertical axis. From this graph, it can be confirmed that when the GA method is used, a pattern can be specified within a practical time even if the number of marks is large. It was confirmed by a test that the number of marks was about 1 to 2 seconds, and even if the number of marks was 100, the marks could be associated with each other in about 10 seconds.

In order to organize understanding of the method of the present invention, a procedure for automatically processing position correction in image information will be described with reference to a flowchart shown in FIG. In step 1, a reference image is acquired by photographing a subject placed under a known condition such as in the atmosphere where ventilation is stopped. In step 2, a measurement image is acquired by photographing a subject placed under a set environmental condition such as supersonic ventilation. In step 3, data processing is performed from the photographed reference image and measurement image, and an extracted image (marker extracted image) of the mark point group attached to the subject surface is obtained. In step 4, the computer is made to solve the problem of obtaining a pattern having the shortest distance among the one-stroke writing patterns connecting the mark points to the marker extracted image of the reference image using an optimization method. In step 5, the pattern which is the obtained solution is stored. In step 6, the computer is made to solve the problem of obtaining a pattern with the shortest distance among the one-stroke writing patterns connecting the mark points to the marker extracted image of the measurement image using an optimization method. In step 7, a pattern which is the obtained solution is stored. In step 8, the obtained pattern of the reference image and the pattern of the measurement image are matched, and the mark points that are the bending points are matched. In step 9, the position information between the corresponding points is substituted into the affine transformation formula to check if there is a geometric contradiction. If there is no contradiction, it is assumed that each mark point has been dealt with, and the process proceeds to step 12 to specify the values of the coefficients a ₁ to a ₆ and b ₁ to b ₆ of the affine transformation equation. If there is a contradiction, the process proceeds to step 10 and the position information of adjacent mark points is exchanged. In step 11, it is confirmed whether the geometrical alignment is obtained by the exchange. When the alignment is not achieved, the exchange of the mark point position information is further promoted until the geometric alignment is achieved, and when the alignment is achieved, the process proceeds to step 12. In step 13, position correction information for each pixel of the subject in the measurement image is calculated using the obtained coefficient affine transformation formula.

  The method of automatically processing the position correction in the image information of the present invention is as described above when measuring the pressure distribution acting on the surface of the model coated with the pressure sensitive paint installed in the wind tunnel from the luminescence image. Not only can it be used as a method that can automatically perform position correction quickly, but it can be widely applied to image position correction using a marker. The present invention can also be applied to a technique for determining the position and orientation of an object by drawing a predetermined pattern on the surface of the object, capturing and acquiring an image of the pattern, and correcting the position with the original pattern.

It is a block diagram of the optical pressure distribution measuring apparatus to which this invention is applied. It is a figure explaining the division image which carried out the picked-up image and data processing in this invention. It is a figure explaining the procedure which obtains a 1 stroke drawing pattern in this invention. It is a figure explaining the optimization method employ | adopted by this invention. It is a flowchart explaining the procedure which implements the method of this invention. It is a figure explaining the processing time in the pressure distribution measurement using a pressure sensitive paint.

Claims (4)

  A step of recognizing one pattern in which the mark group is connected with a single stroke in a reference image of a subject having a large number of mark points on the surface; A step of selecting a pattern most similar to the recognition pattern and identifying corresponding marks, a step of obtaining a coefficient of an affine transformation equation from coordinate information in both images of each corresponding mark, and using the obtained coefficient A method of automatically processing position correction of a measurement image, which includes the step of calculating a correction position by applying it to points on the entire surface of the subject surface.   The method for automatically processing position correction of a measurement image according to claim 1, wherein one pattern in which mark groups are connected by a single stroke is specified by an optimization method on the condition of the shortest distance.   The method of automatically processing position correction of a measurement image according to claim 2, wherein a genetic algorithm is adopted as an optimization method.   The object is a model of a rocket or an aircraft body, and a pressure-sensitive paint is applied to the surface of the model, and the object is applied to a system for measuring pressure distribution from the luminescence image in the wind tunnel. A method for automatically processing position correction of a measurement image according to any one of the above.

Images