JPH06137974A - Image analyzer for object - Google Patents

Abstract

PURPOSE:To stably and precisely catch movement of an object even at a position difficult to see it and reduce cost with regard to an image analyzer for the object. CONSTITUTION:A pressure sensor 22 for outputting a pressure distribution of an object as a load image, a position measuring means 23A for classifying the load image in response to a concentration value of each pixel, calculating a projection value in each classified group, weighting a calculated projection value and calculating a load center as the center of gravity and an operation means 24A for calculating a major axis of a moment centering the calculated center of the gravity and finding a direction of the object and a change of the direction are equipped.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an object image analysis apparatus. In recent years, TVs such as automatic inspection, unmanned monitoring, and motion analysis
Image processing devices that process images captured by cameras are becoming widespread. A desired object can be extracted from the image and information such as the position, size, and color of the object can be captured. As a result, the image processing apparatus can reduce the work manually performed, can realize the work that cannot be entered by a human, and can capture a phenomenon that the human cannot understand. It is attracting attention as a non-contact sensor that fulfills the above requirements.

In addition to these non-contact sensors, there is a demand for detailed image processing of motion states of invisible objects with contact sensors. In particular, attention is focused on a pressure sensor as a contact sensor that captures the pressure distribution of an object applied to a certain area.

[0003]

2. Description of the Related Art In a conventional object image analysis apparatus,
Attach a marker to the object to extract the movement of the object,
This is done by extracting the locus of the marker from the image captured by the TV camera. Specifically, a plurality of color markers whose movements can be easily discriminated are used as markers, and the position of a desired color marker can be calculated by image extraction, color extraction, projection, and calculation of the center of gravity. For example, when the change in the direction of the foot is calculated, the direction of the foot can be obtained by mounting the markers on the toes and the heel of the foot and calculating the positions of the two markers by the above image processing. Furthermore, the above algorithm is easy to process moving images, can be processed at high speed at the speed of TV camera (60 fields per second), and can calculate the change of the foot direction in detail at the sampling interval of 60 points / second. You can

FIG. 13 shows a position measuring side portion for analyzing the movement of an object. In FIG. 13, first, the color marker 1 is attached to a portion of the golfer 2 whose position is to be measured. For example, wear it on the club head, body head, both shoulders, both knees, both feet. A /
The D converter 3 converts the image signal input from the TV camera 4 into an RGB digital signal.

In the color extraction circuit 5, as shown in FIG. 14, the R, G, B image data output from the TV camera 4 is converted using the table 6 to obtain image data of a plurality of specific colors. Only, the pixel value is output with a specific value corresponding to that color, and for other colors, the pixel value is output with 0.
For example, the color extraction circuit 5 is realized by using the "video rate color extraction device" described in JP-A-63-314988.

The noise removing circuit 7 removes isolated points by using, for example, a 3 × 3 logical filter, and extracts a connected pattern in which dots are continuous, such as 4 connected. FIG. 15 shows an example of the output value table 8 for a 3 × 3 input pattern. As shown in FIG. 16, the projection calculation circuit 9 obtains a horizontal projection for each row and a vertical projection for each column with respect to the binary pattern from which noise has been removed. The projection calculation circuit 9 is prepared by the number of colors to be identified. For the projection value calculation, for example, the "video rate projection calculation circuit" described in JP-A-63-140381 is used.

The center of gravity calculating circuit is a DSP (Digital Signal).
Processor) 10 is used. In Figure 17, DSP
10 shows an external RAM 11 and processing blocks of a program. As the center of gravity calculation method, for example, a minute object position measurement method is used. The results of horizontal projection and vertical projection obtained by the processing of FIG. 17 are stored in the external RAM 11 of the DSP 10. The maximum value and the minimum value of the moving area of the object from the host computer 12 are stored in the external RAM 11 as the projection effective section. Further, a value about twice the size of the object is stored in the area of the arbitrary section width of the external RAM 11 as the section width. Using the values of horizontal projection, vertical projection, projection effective section, and arbitrary section width, the cumulative value of the projection values of each section is calculated in sequence.

The section projection values calculated one after another are compared with the maximum section projection value calculated previously, and if larger than that, the section projection value becomes a new section maximum projection value. At the same time, the starting point and the ending point of the section are stored. By repeating this process, the section having the maximum projection value, the start point and the end point of the section are obtained and stored in the external RAM 11. Figure 1
8 shows the maximum projection value in the section.

In FIG. 18, as the set values from the host computer 12, the starting point is the 10th line, the ending point is the 97th line, and the section width is 4. The maximum section projection value is 312, the start point is 14, and the end point is 17. To determine the centroid calculation section, as shown in FIG. 19, the section width is expanded by adding an arbitrary section correction width registered in advance based on the information of the maximum projection section in which the accumulated projection values are maximum. If the projection effective section is exceeded, the start point or the end point of the projection effective section is set as the end point of the centroid calculation section. in this way,
By determining the center-of-gravity calculation section, the accuracy of the center-of-gravity deteriorates because the center-of-gravity calculation of the object A is performed only within the section (2) in (a) of FIG. It is possible to accurately calculate the center of gravity by fitting A into the center of gravity calculation section.

Next, the center of gravity calculation circuit calculates the center of gravity in the center of gravity calculation section determined as described above. In the horizontal projection, the start point of the reception calculation section is sth, the end point is edh, and in the vertical projection, the start point of the center of gravity calculation section is stv,
If the end point is edv and the cumulative values of the projection values of each section are SUMv and SUMh, the center of gravity can be calculated by the following equation.

[0011]

[Equation 1]

[0012]

However, in such a conventional object image analysis apparatus, the marker may be hidden and invisible depending on the movement of the object, and the movement may be stably captured. could not. On the contrary,
There is also a means to observe the movements of objects from multiple TV cameras from various angles, but not only does the total system cost become enormous, but when detecting the foot direction, the feet are hidden by pants or the like. There was a problem.

The present invention has been made in view of the above-mentioned problems of the prior art, and even in a place where an object is difficult to see,
An object of the present invention is to provide an image analysis device for an object, which can capture the movement of the object in a stable and accurate manner and can reduce the cost.

[0014]

FIG. 1 is a diagram for explaining the principle of the present invention. According to the present invention, a pressure sensor 22 that outputs a pressure distribution of an object as a load image, a load image is classified according to a density value of each pixel, a projection value is calculated in each classified group, and a calculated projection value is obtained. Position measuring means 23A for weighting and calculating the center of load as the center of gravity, and orientation calculating means 24A for calculating the principal axis of the moment centered on the calculated center of gravity and determining the orientation of the object and the change in direction are characterized by being provided. To do.

In the present invention, the load state recognition means 23B for recognizing the load state of the object by calculating the total load, the load shape and the spread of the load from the load image is provided, and the load center is determined from the load center. After that, the orientation of the object and the change in the orientation are obtained. Further, the present invention is characterized in that a superimposed image generating means 23C for generating an image in which the load images are superimposed is provided, and the load state of the object is recognized based on the superimposed images.

Further, according to the present invention, the motion width calculating means 24B for calculating the maximum motion distance in the direction perpendicular to the direction vector of the object to calculate the motion width of the object, and the motion width of the object in advance. It is characterized in that a position shift calculating means 24C for calculating the position shift of the object from the given width of the object is provided.

[0017]

According to the object image analyzing apparatus of the present invention having such a configuration, the pressure sensor 22 is used to output the pressure distribution of the object as a load image, and the load image is image-processed to obtain the object image. By measuring the position, calculating the principal axis of the moment centered on the center of gravity of the load image, determining the change in the direction and direction of the object, and calculating the position shift of the object, not only the original pressure change but also the object It is possible to capture the movement of an object in a stable and accurate manner even in a place where is not visible.

Further, since a plurality of TV cameras and the like are not necessary, the cost can be reduced as a total system. As a result, it is effective for diagnosis of golf foams. Also, when calculating the change in the direction and direction of an object,
Since the object load state is recognized by obtaining the total load, load shape, and load spread of the load image, and the load state of the object is recognized from the image in which the load images are overlaid, the change in orientation and direction The calculation accuracy can be improved.

[0019]

Embodiments of the present invention will be described below with reference to the drawings. 2 to 12 are views showing an embodiment of the present invention. FIG. 2 shows an example applied to golf foam diagnosis. In FIG. 2, reference numeral 21 denotes a golfer (object), and pressure sensors 22 are provided immediately below both feet of the golfer 21. The pressure sensor 22 displays the pressure distribution that changes according to the swing of the golfer 21 as a load image (grayscale image) signal (NTSC signal).
Is output to the moving image analyzer 23.

The moving image analyzer 23 performs image processing on the position. That is, the moving image analyzer 23 finely measures the foot position at a sampling interval of 60 points / second. 24
Is a host computer, and the host computer 24 calculates the direction and displacement of the foot based on the measured foot position and the load image. The moving image analyzer 23 uses the image signal from the pressure sensor 21 as A
It is composed of an image memory for storing after the / D conversion, a video rate processor for performing projection processing from the A / D converted image, and a DSP for performing programmable arithmetic processing from the projection result.

That is, the moving image analyzer 23 classifies the load image from the pressure sensor 22 according to the density value of each pixel, calculates the projection value in each classified group, and weights the calculated projection value. The position measuring means 23A for applying the load center as the center of gravity, the load state recognizing means 23B for recognizing the load state of the object by calculating the total load, the load shape and the spread of the load from the load image and the load image are superimposed. It has a function as an overlay image generating means 23C for generating an image.

In addition, the host computer 24 calculates the principal axis of the moment centered on the center of gravity calculated by the position measuring means 23 to obtain the orientation and the change in the orientation of the foot.
4A, the movement width calculation means 24B for calculating the maximum distance in the direction perpendicular to the direction vector using the orientation of the foot calculated by the orientation calculation means 24A, and the width of movement of the foot, and the movement width of the foot are given in advance. It has a function as a positional deviation calculating means 24C for calculating the positional deviation of the foot from the width of the foot.

The analysis result of the host computer 24 is the monitor 2
5 and is printed by the video printer 26. First, the foot position measurement will be described. FIG. 3 is a diagram showing a side portion of the foot position meter. In FIG. 3, the pressure sensor 22 uses a high-sensitivity pressure-sensitive conductive rubber to change the tactile pressure distribution into electric resistance, and outputs the pressure distribution as a grayscale image at a video rate depending on its magnitude. The grayscale image is A / D converted by the A / D converter 27 in any of RGB. For the grayscale image, the projection value is calculated by the projection calculation circuit 30 in each group that is classified after the noise is removed by the noise removal circuit 29 according to the density value of each pixel by the clustering circuit 28,
The projection value is weighted and the load center is calculated by the DSP 31.

The load center is obtained by calculating the center of gravity of the foot pressure grayscale image. Here, the density value of a certain pixel (x, y) is f
If (x, y), the center of gravity G (gx, gy) is generally obtained by the following equation.

[0025]

[Equation 2]

That is, the denominator of gx is Σ in the domain of X.
_It can be calculated by multiplying and summing the result of horizontal projection that is _y f (x, y) and x, and the numerator of gx can be calculated by the total sum of horizontal projection. The same applies to gy. By doing so, the projection is realized by the video rate processor, the sum of products operation, and the summation / division are realized by the DSP 31, so that the center of gravity G (gx, gy) is obtained.
Is generally easily determined at the video rate.

The grayscale image from the pressure sensor 22 is (1),
When the calculation is performed by the equation (2), n labels are attached to the density value f (x, y), and the weight corresponding to each label is reflected in the product-sum calculation performed in the DSP 31. Specifically, when measuring weight shift, we want to calculate the left and right center of gravity, so
In FIG. 4, image data A to C are assigned to calculate the center of gravity of the right foot, and image data D to F are assigned to calculate the center of gravity of the left foot.
The foot pressure image of each foot was given three labels according to the density value, and the DSP 31 weighted corresponding to each label as shown in FIG.

[0028]

[Equation 3]

Next, a method of calculating the foot direction will be described. The position of the foot is the center of gravity calculated by the position measuring unit.
The direction of the foot is determined by regarding the foot pressure image as a distribution image centered on the center of gravity of the foot and calculating the moment. Depending on the individual, the foot pressure image may be obtained only on the toes, heels, or a part of the left and right sides.
Confidence is defined so that only reliable and accurate foot direction can be calculated. The details will be described below.

First, a method of calculating the foot direction will be described. In the coordinate system of FIG. 6, the moment of inertia M of the straight line x = ytan Θ passing through the center of gravity is as follows.

[0031]

[Equation 4]

The principal axis of inertia that minimizes M is (xcos Θ
Let −ysin Θ) ² be F _M, and calculate Θ that satisfies dF _M / d Θ = 0 in the formula shown below.

[0033]

[Equation 5]

Where M _xx is the moment about the x axis,
M _yy is the moment about the y axis, and M _xy is the moment about the y = x axis, which is calculated by the following formula.

[0035]

[Equation 6]

Solving for dF _M / dΘ = 0 gives:

[0037]

[Equation 7]

When the direction of the foot is an angle α with the X axis, the angle α is calculated by the following equation.

[0039]

[Equation 8]

Thus, it becomes possible to calculate the points of the foot direction as shown in FIG. The image in FIG. 7 is a foot pressure image in a standard swing, but in an extreme case such as only the toes or the heel as shown in FIG. 8, there is no information to calculate the foot direction. It must be recognized that it is missing and the moment calculation is meaningless. Therefore, the reliability of the moment is defined, and only the toes and the heels are rejected. The reliability consists of three items, and rejects if even one condition is not satisfied. (1) Size of load The total load is large. That is, A = ΣΣf (x, y)> th1 (2) The shape of the load is not circular. That is, (L ² / A) × π / 4> th2, where (MAXY-MINY)> (MAXX-MIN
X), L = MAXY-MINY, otherwise L = MAXX-MINX (see FIG. 9) (3) Spread of load Large spread. That is, (MAXY-MINY) ² + (MAXX-MINX) ² > th3 In each point, only the toes and the heels are mixed with the toes and the heels in a series of swings. Therefore, an image in which the images of the respective points are superimposed is generated, the center of the load is calculated from the center of gravity from the images, and the moment about the center of gravity is calculated to calculate the orientation of the legs of the series of swings. Figure 10
Shows the result of calculating the foot orientation of the superimposed image. The superimposed image is generated by adding the density value of each frame from the start time to the end time of the foot movement for each pixel in the moving image analyzer.

Basically, in a series of swings, there is no displacement of the feet, only weight movement. However, a bad swing also causes a foot displacement, which needs to be quantified and diagnosed. Here, the width of the foot is calculated, and it is determined that the foot is displaced when the width is larger than the standard. The foot width is calculated from the maximum value of the distance between the vector and the pixel of the load image, using the orientation of the foot calculated from the above-mentioned superimposed image.
A specific method is shown below in FIG. 11, and in FIG.
If the equation of the line representing the direction of the foot is x = ay + C,
The distance d from a certain point (x1, y1) to the straight line is expressed by the following equation.

[0042]

[Equation 9]

If the standard foot width is L and each pixel inside the foot is included between the straight line x = ay + C1 and the straight line x = ay + C2, in order to quickly obtain the above d, x1> ay1 + C1 or x1 <ay1 +
Only d that satisfies C2 is calculated. Find the maximum value of the point satisfying x1> ay1 + C1 as L1 and calculate x1 <ay1 + C
When the maximum value of the points satisfying 2 is found as L2, the width of the foot can be calculated by L1 + L2. The displacement of the foot is obtained as follows, and when it is larger than a certain threshold value th4, it is determined that the displacement of the foot has occurred.

[0044]

[Equation 10]

FIG. 12 shows the results with and without foot displacement. It can be seen from the overlay image that the spread of the distribution is large when the foot position shift is large,
As shown in FIG. 12B, the quantification was performed such that the larger the position, the larger the positional deviation. As described above, since the load image is analyzed using the pressure sensor 21,
Not only the original pressure change but also the movement of the object can be calculated at a place where the object is difficult to see.

In this case, since the load state of the object is recognized based on the total load of the object, the shape of the load, and the spread of the load, the direction and displacement of the object can be calculated more accurately. In addition, since an image in which the load images are superimposed is generated and the load state is recognized, the direction of the object,
The shift calculation can be performed more accurately. Further, since it is not necessary to install a plurality of TV cameras from various angles and observe the movement of the object in a place where the object is difficult to see, the cost can be reduced.

[0047]

As described above, since the load image is analyzed by using the pressure sensor, not only the original pressure change but also the direction and displacement of the object can be calculated even in a place where the object is difficult to see. Therefore, the movement of the object can be stably and accurately captured. Moreover, since a plurality of TV cameras and the like are not required, cost can be reduced. As a result, it is effective for diagnosis of golf foams.

[Brief description of drawings]

FIG. 1 is an explanatory view of the principle of the present invention.

FIG. 2 is a diagram showing an embodiment of the present invention.

FIG. 3 is a diagram showing a foot position measuring unit.

FIG. 4 is a diagram showing a centroid calculation unit.

FIG. 5 is an explanatory diagram of calculating the center of gravity of a grayscale image.

FIG. 6 is an explanatory diagram of calculation of a foot direction.

FIG. 7 is a diagram showing the position and orientation of the foot

FIG. 8 is a diagram showing a toe load and a heel load.

FIG. 9 is an explanatory diagram of the shape of load

FIG. 10 is a diagram showing a superimposed image.

FIG. 11 is an explanatory diagram of foot position shift calculation.

FIG. 12 is a diagram showing a foot position shift.

FIG. 13 is a diagram showing a conventional example.

FIG. 14 is a diagram showing a table of a color extraction circuit.

FIG. 15 is a diagram showing a table of a noise removing circuit.

FIG. 16 is an explanatory diagram of projection calculation of color markers.

FIG. 17 is an explanatory diagram of a centroid calculation process.

FIG. 18 is an explanatory diagram of a maximum projection value in a section.

FIG. 19 is an explanatory diagram of determination of a centroid calculation section.

[Explanation of symbols]

 21: Golfer 22: Pressure sensor 23: Moving image analyzer 23A: Position measuring means 23B: Load state recognizing means 23C: Overlay image generating means 24: Host computer 24A: Orientation calculating means 24B: Motion width calculating means 24C: Position shift Computing means 25: Monitor 26: Video printer

Claims (4)

[Claims] 1. A pressure sensor (22) for outputting a pressure distribution of an object as a load image, and the load image is classified according to the density value of each pixel, and a projection value is calculated in each classified group and calculated. A position measuring means (23A) for weighting the projection value and calculating the center of load as the center of gravity, and a direction calculating means (24A) for calculating the principal axis of the moment centered on the calculated center of gravity and obtaining the orientation of the object and the change in direction. An image analysis apparatus for an object, comprising: 2. A load state recognition means (23B) for recognizing a load state of an object by calculating a total load, a load shape and a spread of the load from the load image, and calculating the load center from the load state as a center of gravity. The object image analyzing apparatus according to claim 1, wherein the direction and the change in the direction of the object are obtained after the processing. 3. A superimposed image generating means (23C) for generating an image in which the weighted images are superimposed is provided, and the load state of the object is recognized based on the superimposed images. An image analysis apparatus for an object according to claim 1 or 2. 4. Motion width calculation means (24B) for calculating the maximum motion distance in the direction perpendicular to the direction vector of the object to calculate the motion width of the object, and the motion width of the object given in advance. 4. The object image analyzing apparatus according to claim 1, further comprising a position deviation calculating means (24C) for calculating a position deviation of the object from the width of the object.