JPH0674859A - Method for analyzing motion of fluid containing particle - Google Patents

Abstract

PURPOSE:To obtain a method for analyzing motion of a fluid containing a particle with a short analysis time. CONSTITUTION:In a method for shooting in time series with an image pick-up means by applying slit light to a plurality of particles floating in a fluid, using a particle image (starting image) at an arbitrary time, that (ending image) in a specific time, and an integral image (flow trace image) from the starting to ending points, and performing identification operation of the starting and ending images for each particle for analyzing the motion of the fluid according to the amount of traveling and a time difference, a window enclosing the flow trace image of each particle is set on identification operation of each particle so that the identification operation can be performed only within the window. Time required for identification operation of a particle inside the images which are shot in time series is reduced, thus speeding up the analysis processing by 10 times or faster.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of analyzing a motion of a fluid containing particles, and more particularly to a method of analyzing a motion of a fluid moving in a space containing particles with a reduced analysis time.

[0002]

2. Description of the Related Art Conventionally, as a device for measuring the velocity of a fluid, a contact type anemometer such as a Pitot tube, a hot wire anemometer and a propeller anemometer and a non-contact anemometer such as a laser Doppler anemometer and an ultrasonic anemometer are used. It has been known. However, these anemometers measure one point in principle, and a traverse is necessary to measure the velocity distribution, which requires a long measuring time. In particular, a great deal of labor is required to obtain the flow velocity distribution of a complicated three-dimensional flow used in an actual industrial product.

On the other hand, in a fluid in which particles are scattered, it is possible to clarify the state of the flow field by visualizing the movement of particles and measuring the moving speed.
A conventional example of this method is shown in FIGS. In the example of FIG. 6, the flow field of the fluid 3 flowing in the direction 2 in the flow path 1 is defined by the slit light source 9
Then, the tomographic illumination is performed in order to make the flow field two-dimensional, and the video camera 4 photographs it. Then, the video is guided to the image processing device 7,
The velocity vector of each particle is calculated. In this case, the position of the slit light source is changed and a plurality of cross sections are measured to obtain the entire volume of the three-dimensional flow. On the other hand, FIG.
In this example, two or more video cameras 4 and 5 are used for shooting, and the flow is stereoscopically shot for analysis.

[0004] These methods have recently attracted attention because they can grasp the entire cross-section at the same time without disturbing the flow. The most important part of this measurement is the image processing method, which is the association (identification operation) of a plurality of particles in the image at an arbitrary time and the subsequent image. Usually, the tracer needs to follow the flow, and therefore fine particles of 1 mm or less are used. Therefore, since the photographed images contain many particles of the same shape, the operation of identifying the same particle image in each photographed image is a complicated and time-consuming operation.

An example of the identification operation is shown in FIGS. (1) Image Capture As shown in the flowchart of FIG. 8, first, an image is captured, binarized, and then the logical sum is calculated. In the case of this example, k = 4 and four consecutive images are captured. This state is shown in FIG. Hatched portions represent particles. The figure in the left column shows images continuously captured from the top by the video camera at 33 msec intervals. The image sent from the video camera represents the brightness of each pixel as an analog signal of 0 to 1 volt, but each pixel of the image captured by the image processing device has a brightness of 0 to 256 by A / D conversion. It is a grayscale image represented by. The middle row is 2
It is a binarized image. The binarization is a process in which pixels having a brightness equal to or higher than the designated brightness are set to 1 and pixels below the specified brightness are set to 0, and particles are recognized as an aggregate of 1 in the image memory. t1 is a start point image and t4 is an end point image. The right column is the logical sum of the binarized images, that is, the integrated images. This is a trace image in fluid mechanics. (2) Identification Processing Next, identification processing is performed using the start point image at t1, the trace image at t4, and the end point image at t4. Figure 10
Shown in. The upper row shows the start point, the flow trace, and the end point image. First, the trace image is labeled. Thereby, each particle is numbered from i = 1 to nn, and the image processing apparatus can recognize each particle. The identification is performed by taking the logical product of the trace image and the start point image, and the trace image and the end point image. Now, if the fifth particle is selected, the trace image will be the image at the third central stage. The logical product 1 of this image and the starting point image is taken. That is, when the common portion is taken out, the image in the third row on the left column is obtained. On the other hand, if the logical product 2 is taken with the end point image, the image on the third row in the right column is obtained. By this, the start point and the end point corresponding to the fifth particle are obtained. Repeat this operation nn
When it is repeated, the identification operation of each particle is completed. However, the trace, start point, and end point do not always correspond well. For example, the third particle has a starting point but no ending point. On the contrary 7
The second particle has no starting point. Such a phenomenon does not occur in the case of a two-dimensional flow, but is a unique phenomenon that appears when a three-dimensional flow is irradiated with slit light. That is, the particles are particles that have flowed out of the slit illumination space during the shooting or have entered from the middle due to the flow velocity component in the direction perpendicular to the paper surface. In addition, the trace images may be connected like the ninth particle, and there may be a plurality of start points or end points. In such a case, since the starting point and the ending point are not one at a time, the vector calculation is not performed because it is abnormal. This method has a very strict logic and the results obtained are reliable. But,
On the other hand, there is a drawback that the processing time is long as described later.

On the other hand, there is a method shown in FIG. The principle will be described with reference to FIG. The image in the right column of FIG. 12 is a binarized grayscale image that has been captured, and is the same as the image in the center column of FIG. Now, focusing on the fifth particle in the starting point image at t1, and assuming the maximum movement amount that is assumed in advance, the particle is within the virtual circle 10 centered on the starting point at the next time t2. In this example, particles of a and b are candidates. Now, assuming that the particle of a corresponds to the fifth particle, a virtual circle a ′ centering on the place of existence at time t3 can be written from the movement amount from t1 to t2. In this case, since there is one particle in the virtual circle, the time t is calculated from the movement amount from t2 to t3 for this particle.
A virtual circle a ″ centered on the place of existence in 4 can be written, and one particle still exists, and this assumption is quite valid, and it can be imagined that the traced particle is the particle No. 5. Virtual circle 1
When the particle of b is traced at 0, it can be seen that at time t3, there is no particle in the virtual circle of b ′ and it is not the particle of 5. This method is shown in the flow chart of FIG.
Although this method has a slightly lower reliability than the example of FIG. 8, it has an advantage that the processing time is short.

[0007]

[Problems to be Solved by the Invention] In the case of the example of FIG.
The time required from T to END is as follows. Required time = mm × (τ1 + nn × τ2) (1) where τ1 is the time required for the image capturing loop, and τ
2 is the time required for the loop of the identification processing. A general-purpose image processing device on the market, τ1 = 400 msec, τ2 =
It is about 460 msec. Further, nn is the number of particles in the trace image of about 500, and mm is the number of times required for averaging in the case of velocity fluctuation such as turbulence, which is about 100 times. Substituting this value into the equation (1), the required time is 6.4 hours.

On the other hand, in the example of FIG. 11, the following equation is obtained. Required time = mm × (τ1 + nn × τ3) (2) Here, mm and τ1 are the same as those in FIG. 8, but τ3 is about 200 msec and shorter than τ2. This is because although the number of internal repetitions is as large as n1 + n2 + n3 + n4, the area for performing logical product and labeling is limited to the virtual circle and the processing area is reduced. In the case of (2), the processing time is 2.8.
This is time-consuming and is shorter than the method using the trace image described above.

In any case, however, considering that the movement phenomenon of particles is on the order of several seconds at most, the time required for analysis is very long. As a method for compensating for these drawbacks, a method has been reported in which a part of the starting point image is cut out and the correlation with the ending point image, that is, pattern matching is used and the most coincident point is set as the moving point. This method shortens the time required for the analysis, but has the drawback that the reliability of the analysis result is lower than that of other methods.

As described above, the method of measuring the flow velocity distribution using the image processing takes less time than the conventional flow velocity measuring method, but it still has a drawback of requiring several hours.
An object of the present invention is to eliminate the above-mentioned problems of the prior art and to provide a method of analyzing the motion of a fluid containing particles, which has a short analysis time.

[0011]

To achieve the above object, the first invention of the present application is to time-sequentially photograph a plurality of particles suspended in a fluid by an image photographing means, and to obtain a particle image (start point image) at an arbitrary time. ), A particle image after a predetermined time (end point image), and an integrated image (trace image) from the start point to the end point, a start point image at an arbitrary time for each particle and a predetermined time after that. In the method of performing the identification operation with the end point image in Fig. 1, calculating the velocity vector from the movement amount and the time difference, and analyzing the motion of the fluid, in the identification operation of each particle, the window including the trace image of each particle is included. And a method of analyzing the motion of a fluid containing particles, characterized in that the identification operation is performed only within the window.

A second aspect of the invention is characterized in that, in the first aspect of the invention, the shape of the window in the window setting is the smallest rectangle including the trace image of each particle, and the movement of the fluid containing the particles is characterized. Regarding analysis method. A third invention is characterized in that, in the above-mentioned first invention or second invention, when the trajectory of particles is discontinuous, it is converted into a continuous trace image by thinning processing and expansion processing. The present invention relates to a motion analysis method for a fluid containing particles.

[0013]

In the method using the trace image, the analysis time is lengthened by the time τ2 required for the identification processing loop. The image processing commands used in the identification processing are logical product and labeling. In the general-purpose image processing device, these commands are processed by using a dedicated processor. When an image of 512 × 512 pixels which is normally used is executed by a general workstation without using a dedicated processor, both logical product and labeling are executed in about 1000 msec. On the other hand, if a dedicated processor is used, the logical product can be processed at a very high speed of 33 msec and the labeling is 200 msec.

However, a dedicated processor capable of processing at such a high speed also requires a long analysis time because it has a serious drawback that local processing cannot be performed in particle identification processing. In the case of identification processing, it suffices to focus only on the vicinity of the trace image of one particle, but since the dedicated processor is designed to process the whole at high speed, the processing range cannot be limited locally. , The whole 512 × 512 pixels are processed at once. As a result, the long-time processing as described above ends up occurring.

Therefore, if the identification processing is performed by software without using a dedicated processor, the local processing can be performed, so that the identification processing can be processed at high speed. When obtaining the flow velocity, the size of the trace image is about 100 pixels at most. Therefore, if only this area is set as the processing range, the processing time is proportional to the number of pixels, so that both logical product and labeling can be processed in 3 msec. Therefore, the processing time τ2 in the identification loop is about 20 msec. By substituting this value into the equation (1), if the general-purpose image processing device is not used at all, 0.
It will be 5 hours, and the processing time will be significantly shortened.

[0016]

FIG. 1 shows a flow chart of an embodiment. The outline is the same as in FIG. 8, but a window setting is added in the identification processing loop. In this example, image capture is 512 ×
A general-purpose image processor of 512 pixels was used, and the identification processing loop was performed in a 32-bit workstation. The visualized image was taken by mixing nylon particles with a particle diameter of 1 mm into a water flow flowing at 10 cm / sec through a 10 cm square rectangular duct and forming a slice with slit light. This image was photographed with a video camera of 30 frames per second, once recorded on a video recorder, and then framed and reproduced for processing.

The flowchart of FIG. 1 will be described below. (1) Capture 4 frames of the image, as shown in the top row of FIG.
The beginning is the start point image, the end is the end point image, and the logical sum of four frames is taken as the trace image. (2) Labeling processing is applied to the trace images, and each trace image is set to 0.
Number up to nn. 50 trace images in the screen
There are about 0, but since the image becomes complicated, it is omitted in FIG. 2 and only 9 are shown. When labeling is performed, the vertex coordinates of a rectangle surrounding each trace image are obtained at the same time. After that, the process of identifying each trace image and the start point / end point image is started. (3) Each trace image is sequentially cut out from the trace image. At this time,
If it is cut out normally, the whole screen is processed, so it takes about 30 mm / sec for one trace image. On the other hand, if you use the rectangle used for labeling to set the window and cut it out, it will be processed in a partial area, so it can be done at about 2 mm / sec. (4) The logical product of the cut out trace image and the starting point image is calculated,
Cut out the starting point image. Since the window is used also at this time, the place where it normally takes 30 mm / sec is 2 mm / sec. Then, the cut image is labeled and 1
It is checked whether or not only one piece can be cut out, and when only one piece is considered normal, the process proceeds to the next step. This determination is indispensable because in a complicated three-dimensional flow, trace images will inevitably overlap each other. At the same time as the labeling, the center of gravity of the particle image is obtained. (5) The logical product of the cut out trace image and the end point image is calculated,
Cut out the final image. Since the window is used also at this time, the time is shortened. Then, in the same manner, it is checked by labeling whether or not there is only one image cut out. If it is one, it is considered normal and the next step is entered. At this time, the center of gravity of the particles is also obtained. (6) The moving distance of the particle is calculated from the barycentric coordinates of the start point and the end point, and the velocity vector is obtained from the exposure time of four frames. (7) The operations of (3) to (6) are repeated as many times as the number of particles, that is, nn times to obtain velocity vectors of all particles.
FIG. 2 shows the state of the processing in which the window setting is performed. The top row is an image of the starting point, the trail, and the ending point. Although there are about 500 particles on the screen, they are shown in a schematic diagram because the image becomes complicated. The trace image is made by adding four frames.

In the present processing, a window is set around the trace image like the particle No. 5 shown in the third row. In this case, a rectangular window inscribed in the trace image as shown in FIG. 3 is used. The average number of pixels in the window was 100 pixels. Therefore, the execution time of the logical sum of the trace image and the start point image and the end point image and the trace image was 2 msec. The window was set for each particle.
As with the conventional example of FIG. 9, vector calculation is not performed on the images from the third stage to the fifth stage.

As a result, the processing time for each instantaneous image was 10 seconds, and the processing time for 100 times was 1040 seconds. Although a general-purpose image processing apparatus is used in the present embodiment, if omitted, the processing time for 100 times becomes 1800 sec, which is somewhat long, but the apparatus can be made inexpensive. Although the window setting was performed for each particle, it takes time if a general-purpose image processing device is not used, so the pixel number of the maximum trace image is estimated and set uniformly as shown in FIG. You may. In the future, if a window with the same shape as the trace image can be set, it can be processed at higher speed.

As described above, according to the present invention, the shortest processing time can be achieved by the strictest method using the trace image. In the present embodiment, the exposure time of the video camera is 33 msec, but depending on the model, the exposure time may be short even if the number of frames is the same. In this case, the trace image becomes discontinuous as shown in the upper part of FIG. 5, but it can be corrected by using the image processing command. For example, each image may be made into a line, thinning processing may be performed, and then expansion processing for expanding each image may be performed. When the expansion process is started from the beginning, each figure expands uniformly,
Since there is a high probability that the particles will be connected to the adjacent particles, the thinning process extends in the longitudinal direction, and thus the connection can be prevented.

Further, the present invention can naturally be applied to the case where two or more video cameras are used.

[0022]

According to the present invention, in the flow velocity distribution analysis using the visualization method, it is possible to process the flow velocity distribution at a much higher speed than in the past with the strict logic using the particle trace.

[Brief description of drawings]

FIG. 1 is a flowchart showing an embodiment of the present invention.

FIG. 2 is a schematic diagram showing a state of image processing in the present invention.

[Fig. 3]

FIG. 4 is a diagram showing how to apply a window.

FIG. 5 is a diagram showing an example of creating a trace image.

FIG. 6 is a diagram showing a flow velocity distribution measuring method in which image processing is added to a conventional visualization method.

FIG. 7 is a diagram showing a conventional technique using two video cameras.

FIG. 8 is a diagram showing a flowchart of a conventional image processing method.

FIG. 9 is a schematic diagram showing a state of conventional image processing.

FIG. 10 is a schematic diagram showing a conventional image processing method.

FIG. 11 is a flowchart showing another conventional image processing method.

12 is a schematic diagram showing a state of image processing in the method of FIG.

Claims (3)

[Claims] 1. A plurality of particles suspended in a fluid are time-sequentially photographed by an image photographing means, and a particle image (starting point image) at an arbitrary time, a particle image after a predetermined time (end point image), and a starting point. Using the integrated image (trace image) up to the end point, the identification operation of the start point image at an arbitrary time and the end point image after a predetermined time is performed for each particle, and the velocity vector is calculated from the movement amount and the time difference. In the method of calculating and analyzing the motion of the fluid, when identifying each particle, a window including a trace image of each particle is set, and the identifying operation is performed only within the window. A method for analyzing the motion of a fluid containing particles. 2. The motion analysis method for a fluid containing particles according to claim 1, wherein the shape of the window in the window setting is a minimum rectangle including the trace image of each particle. 3. The movement of a fluid containing particles according to claim 1 or 2, wherein when the trajectory of particles is discontinuous, the particles are converted into a continuous trace image by thinning processing and expansion processing. analysis method.