JPH0326986A - Marker detector - Google Patents

Abstract

PURPOSE:To accurately detect the center position of a marker by providing a means taking an image containing a marker, a lightness sum/difference calculation means, a dispersion calculating means and a means judging a point where lightness sum/difference is the first predetermined threshold value or more and each dispersion is the second predetermined threshold value or less. CONSTITUTION:A dummy 4 rides in a car 3 and a state of the collision of the car 3 having a plurality of markers 1 against a wall 5 is photographed with a high speed camera 7 and reproduced by a reproduction control part 13 to be stored in an image memory 17. The output of a filtering operation part 23 is judged by a primary candidate point extraction part 25 and a dispersion value is evaluated at the first predetermined threshold value or more by a secondary candidate point extraction part 29 and the final determination of a marker position is performed at the second predetermined threshold value or less by a marker position calculating part 33. By this method, the center position of a marker can be accurately detected.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Application Field) The present invention relates to a marker detection device used, for example, in automobile collision experiments.

(Prior Art) As shown in FIG. 13, for example, a conventional marker detection device of this type uses a high-speed camera to monitor the collision of a car 3 to which a plurality of markers 1 are affixed to a wall 5 in front of it. This recorded image is reproduced and monitored by the reproduction v1 setting 100, and the reproduced image is manually analyzed. Through this analysis, the behavior of the dummy doll riding in the car 3, the state of damage to the car, etc. can be ascertained, thereby helping to ensure the safety of the car. Note that the camera 7 is designed to move together with the car 3.

Furthermore, as a conventional method for detecting a special-shaped object that can be applied to this type of marker detection device, for example, the Hough conversion method for detecting a circle is well known.

In this method, the normal direction of an edge is determined at a point on the image where the edge strength is large, and voting is performed on a separately prepared image plane at a point that is a predetermined radius R away from the edge in the normal direction. This is the h method, which considers the many points as the center of a circle with radius R (Pattern Recogn1tlon I,
etters No. 7@, No. 1, pp. 37-43, 1988
Year, R. R. (Reference: DaVleS).

(Problems to be Solved by the Invention) The conventional mica detection device shown in FIG. There is a problem of large variations in ε.

Furthermore, when using Hough's transformation method, a circular shape looks like an ellipse when viewed from an angle, but there is a problem in that it cannot deal with such distortion of the circular shape.

The present invention has been made in view of the above, and its purpose is to provide a marker detection device that can automatically perform image processing and accurately detect the center position even if the marker is distorted. It is in.

[Structure of the Invention] (Means for Solving the Problems) In order to achieve the above object, the marker detection device of the present invention has the following features:
A marker force detection device that detects a marker of a specific pattern, comprising: an imaging means for imaging an image including the marker; and a brightness sum of a first group of pixels on a line segment in the image taken by the imaging means. , the second pixel group on the line segment perpendicular to the first pixel group
a brightness:fu difference calculating means for calculating a brightness sum difference obtained by subtracting the brightness sum of the pixel group; a variance calculating means for calculating each variance of the first and second brightness values; and a brightness sum difference calculating means. the calculated brightness sum difference is greater than or equal to a first predetermined threshold;
and determining means for determining that the point where each variance calculated by the variance calculating means is less than or equal to a second predetermined threshold value is the marker center.

(Function) In the marker detection device of the present invention, the first and second orthogonal
While calculating the picture of each brightness sum of the pixel group on the line segment,
The variance of each brightness value is calculated, and the point where the brightness sum difference is equal to or higher than a first predetermined threshold and each of the variances is equal to or lower than a second predetermined threshold is determined to be the center of the marker. .

(Example) Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is an illustration of a marker detection device according to an embodiment of the present invention. The marker detection device shown in the figure is applied to a collision photo analysis device that images and analyzes the collision state of a car, and a car 3 to which a plurality of markers 1 are attached collides with a wall 5, etc. in front of it. The situation is imaged by a high-speed camera 7 that moves with the car 3 and recorded in a high-speed photo recording device 9, and the recorded image is image-processed by a reproduction storage unit 11, one marker position extraction unit, and a search determination unit 35. Through this, when the car 3 collides with a wall 5 or the like, the behavior of the dummy 4 riding in the car 3, the state of damage to the car, etc. can be grasped, and this helps to ensure the safety of the car.

The reproduction storage section 1-1 is composed of a reproduction control section 13, an A/D conversion section] 5, and an image memory 17, and one marker position extraction section includes a filter pattern generation section 21-, a filtering calculation section 23, It consists of a primary candidate point extraction section 25, a variance calculation section 27, a secondary candidate point extraction section 29, a clustering section 31, and a marker position calculation section 33. It consists of three parts.

In the marker detection device shown in FIG. 1, an image taken by a high-speed camera 7 of a collision test T in which a car 3 on which a dummy doll 4 is riding and a plurality of markers 1 are attached collides with a wall 5 is a high-speed image. The recorded image is recorded in the speed photographic recording device 9, and is reproduced by the reproduction control section 13 of the reproduction storage section 11, and converted into a digital image signal by the A/D conversion section 15, resulting in image data I (x, y). The image is stored in the image memory 17 as .

Image data I (x, y) stored in image memory 17
is supplied to the marker position extraction unit]9, where marker position extraction processing is performed. Note that the target marker 1 here is
As shown in FIG. 2, this is a very commonly used target, and there are no restrictions on its size, color, or direction.

In the marker position extracting unit 19, in order to use it for filtering to extract marker 1 from the image data I (x, y) stored in the image memory 17, a filter pattern similar to the shape of marker 1 is generated in the filter pattern generation $21. Create a new pattern.

That is, the filter pattern generating section 21- generates the marker 1
In order to obtain an extraction result that is independent of the direction, a plurality of filter patterns as shown in FIG. 3, four filter patterns A, B, C and D in FIG. 3, are generated. These filter patterns each have four 13x13 image memories F A (X + V) + Fs (X.V
), Fc (X+V)+Fo(x,y), and in FIG. 3, the pixel indicated by 1 stores "2", the pixel indicated by the mouth stores "1", and the other pixels store "2". 0"
By remembering this, it becomes effective.

Image data I (x, y) stored in image memory 17
is supplied to the filtering calculation section 23 together with the filter pattern generated from the filter pattern generation section 21, and the pattern of marker 1 is extracted from the image data.

The principle of the filtering operation in the filtering operation unit 23 will be explained by taking as an example a case where a marker 1 having the direction shown in FIG. 2 is extracted. In the marker 1 shown in Fig. 2, bright points, that is, pixel openings, are distributed on a straight line in the direction of 45@, and dark points, that is, pixel 1, are distributed on a straight line that is orthogonal to this straight line. . Therefore,
Among the filter patterns generated by the filter pattern generating section 21, the filter pattern A shown in FIG. 3 is used to apply a filter, and the difference P in the brightness sum is calculated by the following equation.

P-1 total brightness of dark part mouth - total brightness of dark part 1 1 (1) This difference P between the calculated brightness sums shows a prominent value at the center position of the marker.

Therefore, the filtering operation in the filtering operation unit 23 is performed by calculating the above equation (1) for the corresponding points of the filter pattern A and the original image data I (x, y), as shown in FIG. , the data of the point corresponding to the center of the filter pattern is the filter output data JA
It is output as (X, y).

That is, the filter output from the filtering calculation unit 23 is obtained by performing the calculation shown in the following equation so that each filter is applied to the original image 1 (x, y).

However, F A (x, y)-2, F A (x'.y'
)-1 However, F a (x.y)-2. F a (x'
．． y')-1 However, F c (x, y)-2, F c
(x', y')-1 However, F o (x.y)-2, F
D (x', y'')-1 The filter output data JA (X, y) from the filtering calculation unit 23 is supplied to the primary candidate point extraction unit 25, where it is determined whether the filtering result is an outstanding value or not. This is done by comparing the filter output data J^ (x, y) with a first predetermined threshold Thl (determined by experiment etc.) and
When y)≧Thl, PA (X+ y)-1, and when PA (X. y)<Thl, PA
(x,y)-0, and the resulting point with PA (X.y)-1 is called a primary candidate point.

That is, each primary candidate point PA (X, 3'),
PB (x, y) , Pc (x, y) r Po
<x, y) is as follows.

? If the marker position is output as is, there is a risk of incorrect extraction. This is because an area that happens to have the same brightness distribution as the marker is surprised. Therefore, in order to narrow down the marker positions, the primary candidate point extraction unit 2
5 is supplied to the distributed calculation unit 27, and this distributed calculation unit 2
7, the variance VAI(X
I Y) and the variance vA■(x.y
) are calculated respectively. Variance VAI calculated as a result
(X.y). Both V^2(X, y) have small values in the true marker part.

Calculation of variance is PA (X,y)-1,P3 (X
,y)-1,Pc (x,y)-1,Po (x,
It is obtained by calculating the following equation for the point y)-1.

V ^yi (x++ y+) The primary candidate point obtained by the primary candidate point extraction unit 25 is
A (x,y)- 2 V^2(X y+) Vc+(X yI) However, F A(!', y')= 1 V8+(X+ r Y+) However, F c
(x=y)-2 VC2 (xI, YI) However, F a
(x.y)-2? Il ■ ( X y+ ) The variance value calculated by the primary candidate point extraction unit 25 is supplied to two secondary candidate point extraction units, and the variance value is evaluated. For Th2 (determined by experiment, etc.), P A (X + y)-1, VA 1 (X, y):=Th2, and vA2
This is done by setting QA(x,y)=1 when (x,y)≦Th2, and setting QA(x,y)-0 otherwise. QA (X
, y) -1 is called a secondary candidate point.

That is, each two candidate points Q^ (X, Y), QB(x,
y) , Qc (x, y) , Qo (x, y
) is as follows.

1 - P A (x, y) - 1 Th V
A+ (x, y)≦7JQA (x, y) −
fp VAI (1.1>≦Th,0...
・other 1. - P B (x,y)-1 and V
al (x+ y)≦Th,Q [l (x, y)
1 and VJ (x+!)≦Th,
0-other 1-P c (x,y)-1 and vc+ (x
, y)≦Th, Q c (X , y) 1 #k VC
2 (Zl 7 )≦Tb20...other 1 - P o (x,y)-1 and VJ
(x,y)≦Th2Qo(x,y) -
ka'lVD2 (L y)≦Th,0...oth
After being processed by the two secondary candidate point extraction units, the true marker positions are further narrowed down by the clustering unit 31.

This clustering process is performed according to the flow shown in FIG.

In FIG. 5, first, as described above, the primary candidate point extraction unit 25 extracts the 1st and secondary candidate points PA (X, y)-1 (step 11.0), and then the secondary candidate point extraction unit 2
9, the secondary candidate point QA (X, y
)-1 (Stethop 120). Then, filter output JA (X
. be exposed. This process is performed using the secondary candidate point QA (
x, y) = 1 until there are no more points (step 1
50).

FIG. 6(a) is a flowchart showing this clustering process in more detail. In the same figure, first clear the X and y coordinate values Rx (v) and Ry (v) of the markers, and set the subscript V representing the number of markers to 1 (v=1).
, set the area for accumulating the maximum value S of the filter output data JA (x.y) to 0 (S-0), and further set the image memo, Ri1.
Set addresses y and X of 7 to 0 respectively (y-0,
x=0) (steps 200-240). Then, the point at the address (x, y) of the image memory 17 set in steps 230 and 240 is the secondary candidate point Q^ (X, y)-1
It is checked whether it is (step 250). If it is not a secondary candidate point, return to steps 230 and 240 and repeat the process for all values from 0 to the maximum value for each of X and y until the secondary candidate point QA (X, y) is no longer equal to 1. When there are no more next candidate points, the process ends (step 320).

If the secondary candidate point QA (X, y) = 1 is detected in step 250, the filter output data J^(x, .y
) is checked to see if the value obtained by subtracting the filter output data JA (x, y) corresponding to the addresses x, y of the memory I7 from the maximum value S is greater than 0, and Find the point where the filter output data JA (X.y) is maximum (step 260). Let the filter output data JA (x, y) at this maximum point be the maximum value S, and let the addresses x, y of the image memory 17 in this case, that is, the x, y coordinates of the marker position in this case, be t and U, respectively. (step 270). These coordinates t and u are respectively defined values Rx (V) and Ry (
v) (step 280) X of this confirmed marker
and the coordinate Xm obtained by subtracting a predetermined constant α from the y coordinate
While calculating Rx (v) - α and Y-Ry (v) - α (step 290), as shown in FIG. 6(b), this locus II
and Y-Ry (v) from the coordinates X-Rx (v)
+a and Y−Ry(V)+α, all secondary candidate points QA(x,y) are set to 0 (step 30
0).

Then, the number of markers V is updated by incrementing it by 1 (step 310), and the process returns to step 220 to repeat the same process.

As a result of the above clustering process, only the point with the highest filter output data JA (X, y) is detected among the secondary candidate points, and its neighboring points are removed to zero. 7th
The figure is a table showing arrays Rx (v) and Ry (v) generated from the results of such clustering processing.

For the marker positions determined by the clustering process in the clustering unit 31, the final determination of the marker positions is further performed in the marker position calculation unit 33. This is the marker position (x
This is achieved by performing the following calculation on 3×3 pixels in the vicinity of i, y i).

The coordinates obtained in this way are converted to the true marker position (xi,
yf).

Above formula (2). (3) Two-dimensional histogram JA
The center of gravity position near 3X3 of (X.y) is calculated,
This makes it possible to increase the accuracy of marker position extraction. That is, marker positions can be calculated on the order of pixels or less.

More specifically, the final determination process of the marker position is performed as shown in the flowchart of FIG.

In FIG. 8, first, the number of markers V. is set to 1, this number of markers v0 is set to V, and it is checked whether Rx (v) and Ry (v) are O (steps 410 to 4).
30). Rx (v). If Ry (v) is not 0, calculate the center of gravity position of JA (X, y) within the 3X3 neighborhood (step 440). Then, the number of markers V is increased by +1, the process returns to step 430, and Rx (v), R
Repeat until y(v)=0 is detected (step 45
0). In step 430, Rx (v)=Ry (v)=
If 0 is detected, the marker number ■ is stored as vO, and the process ends (step 460). Note that the above processing is for filter A, and filters B, C,
If D is applied, the above process is started from the position of step 420.

The original image 1 (x, All markers in y) can be extracted as shown in FIG.

In other words, the result obtained by the above processing (X (v)Y
(V) ) and let the extraction result be Ki(x,y). Here, Ki (x, y) is the i-th original image] elephant I
t (x, y), Ki (x, y) 1 is plotted at the extracted point, and Ki (X.y> -a is plotted at the other points).

Furthermore, since X (v) and Y (v) are not integer data, when plotting K t (x+y), they are plotted as integer data closest to it. The result produced in this way is shown in FIG.

The processing flow of one marker position extraction unit explained above is the first
This is what is shown in Figure 0.

The marker extracted by the marker position extraction unit 19 is supplied to the search determination unit 35, and first, the correspondence search unit 37 correlates the l-th marker extraction result with the i+1-th marker force extraction result, and determines where the marker is. Determine whether it has moved.

This process is performed by creating a distance d between the i-th marker extraction result Ki (x, y) and the i+1-th marker extraction result Kt+1 (x, y), which are the two marker extraction results, as shown in FIG. This is done by matching the points within the range. In FIG. 11, i+1, indicated by rxJ, is the closest to the i-th marker extraction result Ki, indicated by "・" within "O".
Search for the th power extraction result Ki+1.

Assuming that the time interval between the acquisition of the two images is sufficiently short, the amount of movement of the marker is considered to be very small. Processing is accomplished.

In FIG. 12, the coordinates x, y are set to O, and it is checked whether the marker extraction result Ki (x, y) for these coordinates is 1, and the marker position is detected (step 710).
, 720). Let this detected mechanical force position be (x, y). Note that the process of step 720 is performed for all of x-0 to maximum value and y-o to maximum value, and the process ends (steps 720, 810).

In the next step 730, other coordinates X' and y'' are set to O, respectively, and a constant β is substituted for the distance 11D. Note that the constant β is set to a sufficiently large value. The force extraction result Ki+1 (
It is checked whether X=, V-') is 1 and the marker position is detected (step 750). Note that the processing in step 750 is similar to step 720 in that X′ and y
′ for all values from O to the maximum value.

Check whether the distance between the two marker positions (x, y) and (x-.y-) detected in steps 720 and 750 is smaller than D (step 760). As a result of the check in step 760, if the distance between them is less than or equal to D, replace D with the distance between (x, y) and (x-.y-), and replace X' and y'' with X and Y
(steps 770, 780). Ki−←1 After searching all the points of (x”, y−)−1, check whether D is less than or equal to the threshold d, and if it is less than or equal to the threshold d, ( +y) and (
X, Y) represent the same marker, and a straight line connects both coordinates (steps 790, 800). Note that straight line drawing may be performed using the DDA algorithm. Ki (x
, y)-1 and terminate this process (
step 810).

Note that in step 790, if the calculated D is larger than the threshold value d, it is determined that the marker imaged at KL (x, y) is hidden at Ki+1 (x, Y). Through this process, a matching result as shown in FIG. 11 is obtained.

When the three end processing determining units determine that the series of processes described above have been completed, the playback control unit 13 outputs an instruction to play the next frame, that is, a “one frame advance signal”, and the next frame is output. A series of processes of extracting markers and associating them is performed on the same frame.

[Effects of the Invention] As described above, according to the present invention, the difference between the brightness sums of the pixel groups on the first and second orthogonal line segments is calculated, and the variance of each brightness value is calculated. , the lightness sum difference is the first
The marker position is determined to be the center of the marker at the point where each variance is less than or equal to the second predetermined threshold. This can be done automatically through image processing, making it possible to improve work efficiency, and even if the marker is distorted, such as in a car crash experiment, the center position of the marker can be accurately detected.

[Brief explanation of drawings]

FIG. 1 is a diagram showing the structure of a marker detection device according to an embodiment of the present invention, FIG. 2 is a diagram showing the shape of a marker used in the marker detection device of FIG. 1, and FIG. 3 is a diagram showing the shape of a marker used in the marker detection device of FIG. Figure 4 shows the filter pattern used in the marker detection device.
FIG. 5 is an explanatory diagram showing the IE essentials of clustering processing in the clustering unit used in the marker detection device of FIG. 1;
FIG. 6 is a flowchart showing the clustering process of the clustering unit, FIG. 7 is a diagram showing the results of the clustering process of the clustering unit, and FIG. 8 is a flowchart showing the clustering process of the clustering unit. FIG. 8 is a flowchart showing the clustering process of the clustering unit. A flowchart showing the final position determination process, FIG. 9 is an explanatory diagram showing the marker extraction process by the marker position calculation unit used in the marker detection device of FIG. 1, and FIG. 10 is the marker detection device of FIGS. 11 is an explanatory diagram showing the association of markers in the marker position calculation section, and FIG. 12 is a flowchart showing the operation of the marker position extraction section used in the marker detection device of FIG. 1. FIG. 13 is a flowchart showing the operation of the search section, and is an explanatory diagram showing conventional marker detection processing. 1... Marker 7... High speed camera 9... High speed photo recording device 11... Reproduction storage section 19... Marker position extraction section 21... Filter pattern generation section 23... Filtering calculation Section 25...Primary candidate point extraction section 27...Two variance calculation sections...Secondary candidate point extraction section 31...Clustering section 33...Marker position calculation section 37...Correspondence search section

Claims (1)

[Claims] (1) A marker detection device that detects a specific pattern of markers, including an imaging device that captures an image including the marker, and a brightness of a first pixel group on a line segment in the image captured by the imaging device. a brightness sum difference calculating means for calculating a brightness sum difference by subtracting a brightness sum of a second pixel group on a line segment perpendicular to the first pixel group from the sum; a variance calculation means for calculating each variance and a brightness sum difference calculated by the brightness sum difference calculation means is equal to or greater than a first predetermined threshold;
and determining means for determining that a point where each variance calculated by the variance calculating means is less than or equal to a second predetermined threshold value is the marker center.