JPH06147830A - Three-dimensional position measuring equipment and measurement correcting method - Google Patents

Abstract

PURPOSE:To obtain highly accurate measurements while enhancing operability and shortening measuring time by detecting light emitted from an LED fixed to a coordinate teaching probe through a light emission detector and delivering a detection signal to an instrumentation controller. CONSTITUTION:When three-dimensional position of a measuring point P on a work 4 is measured, the measuring point 6 is touched by a supporting point 6 of a coordinate teaching probe 5 and then an LED switch 8 is depressed to light an LED 7. An LED emission detector 9 comprising a photodiode detects lighting of the LED 7 and delivers an emission sense signal to an instrumentation controller 10 thus notifying lighting of the LED 7. Immediately upon receiving the emission sense signal, the instrumentation controller 10 drives a three- dimensional position operating unit 3 to image the probe stereoscopically by means of TV cameras 1, 2. Three-dimensional position of each LED is then calculated based on brilliant point parallax of LED in the right and left stereoscopic images and three-dimensional position of a designated point 6 is calculated based on the three-dimensional position of each LED 7.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is used for measuring the three-dimensional shape of a workpiece or a tool in a machine tool such as an electric discharge machine, or for measuring the three-dimensional shape of the workpiece or CAD (Computer).
The present invention relates to a three-dimensional position measuring device used when inputting a three-dimensional shape when using such as Aided Design).
The present invention also relates to a three-dimensional position measurement result correction method in the above three-dimensional position measuring device.

[0002]

2. Description of the Related Art FIG. 5 is a block diagram of a three-dimensional position measuring apparatus. In the figure, reference numerals 1 and 2 are image pickup apparatuses such as TV cameras,
3 is an image processing device for calculating a 3D position based on a shift (parallax) between left and right images of stereo images from these TV cameras 1 and 2, and 4 is a 3D position measurement object (hereinafter referred to as a work). The work 4 has a recess 4a that cannot be imaged by the TV cameras 1 and 2. Reference numeral 5 is a coordinate teaching probe (measurement point teaching means).
Is a pointing point 6 for pointing a measuring point P on the surface of the work 4.
And a plurality of LEDs (point light sources) 7 that have a known fixed positional relationship with the designated point 6 in advance.

Next, the operation will be described. In FIG. 5, when the measurement point P of the work 4 is imaged, the designated point 6 of the coordinate teaching probe 5 is placed in contact with the measurement point P, and after the contact, the LED 7 is turned on. Then, with the LEDs 7 turned on, the TV cameras 1 and 2 capture a stereo image of the coordinate teaching probe 5 including the LEDs 7. The image pickup result is input to the image processing device 3, and the image processing device 3
In, the three-dimensional position of each LED is calculated based on the bright spot parallax in the left and right images of the stereo image of the LED 7. Since the positional relationship between each LED 7 and the pointing point 6 is known in advance, the three-dimensional position of the pointing point 6 is calculated from the three-dimensional position of each LED.

Further, FIG. 6 is a digital image picked up by an image pickup apparatus as shown in, for example, "Image Recognition Theory" by Shin Nagao (published by Corona, pp.23-28, 3rd printing, 1984, first edition). It is explanatory drawing explaining the conventional correction method of the geometric distortion of.

When an image is taken by an image pickup device such as a TV camera in the three-dimensional position measuring apparatus shown in FIG. 5, various geometric distortions such as perspective distortion and lens system distortion caused by photographing from an oblique direction. Occurs. Therefore, in order to perform accurate measurement, geometric correction of these distortions is performed. In the conventional three-dimensional position measurement result correction method, first, in each of the images obtained by the TV cameras 1 and 2, the measurement target point P (X,
Camera coordinates (x ₁ , y ₁ ) (x, Y, Z) (pointing point or position of each LED) on each imaging plane of the cameras 1, 2
₂ , y ₂ ) is obtained using posture parameters such as the position and orientation of the camera in the measurement space. Next, the following geometrical correction is performed on these camera coordinates. Then, the three-dimensional position is calculated based on the corrected camera coordinates.

The geometrical correction method will be described below.
In FIG. 6, the grid points (x, y) of the corrected undistorted coordinate system are general points (x ^' , y ^' ) on the distorted coordinate system.
Suppose that it falls into, and the relational expression is given by the following expression. _{^{x '= h 1 (x,}} y) y' = h 2 (x, y) (1) Here, h _1, h ₂ is, (x, y) mapping relationship from the (x ^{', y')} Is a function that represents. If h ₁ and h ₂ are, for example, linear expressions, they are expressed as x ^′ = ax + by + cy ^′ = dx + ey + f (2), and the shape distortion that often appears in a television image is approximately x ^′ = k ₁ x {1 + k ₂ (x ² + y ² )} y ^′ = k ₁ y {1 + k ₂ (x ² + y ² )} (3) Generally, various distortion factors are present in an image, and it is difficult to theoretically determine these factors. Therefore, distortion correction is performed using a general polynomial expression such as the following equation. _{^{x '= Σa ij x i y}} j y' = Σb ij x i y i (4) i, or take to what extent higher the j will depend on the nature of the image, as defined by the scope of the often i + j ≦ 5 .

The determination of the coefficient of the coordinate conversion equation as described above is performed by using a specific point on the image whose coordinates on the ideal image without distortion are known as a reference point. Since there are six coefficients in a linear expression such as Expression (1), at least three reference points need to be coordinated to determine these. Further, in the case of taking up to the third-order term in the equation (4), the number of coefficients becomes 20, so that it is necessary to correspond the coordinates of 10 reference points. In general, more coordinate points are taken, and a more stable conversion equation is obtained on average by a method such as the least square method.

[0008]

Since the conventional three-dimensional position measuring device is constructed as described above, it is necessary to judge whether or not the LED is turned on by image processing. Therefore, the pointing point of the teaching probe must be placed in contact with the measurement point and the LED must be kept on for a long time with the LED turned on for a long time, resulting in poor operability.

Further, when the measurement is performed using an image pickup device such as a camera, there is a problem that the measurement cannot be performed accurately due to the distortion of the lens and the displacement of the image pickup element. Therefore, it was necessary to make a correction, but since the conventional measurement result correction method is performed as described above, it is necessary to make a correction for each image pickup device, and thus a large amount of labor is required to determine the coefficient of the correction formula. It is difficult to determine the optimum order when the correction formula is expressed by a polynomial.
Only the error that can be approximated by the correction formula can be corrected, and the three-dimensional position is not taken into consideration because the correction is made by the coordinates of the measurement target point on the imaging plane, so that the accurate correction cannot be performed.
There was a problem such as.

The present invention has been made in order to solve the above-mentioned problems, and it is possible to improve the operability and shorten the measurement time by shortening the holding time of the teaching probe during measurement. In addition, even when the result measured by the three-dimensional position measuring device includes an error that cannot be approximated by the correction formula, the three-dimensional position measuring device that enables easy and accurate measurement, and a more accurate three-dimensional position measuring device. It is an object of the present invention to provide a correction method that can obtain high measurement results.

[0011]

A three-dimensional position measuring apparatus according to the present invention detects the light emission of a point light source by using the point light source light emission detection means of the measurement point teaching means for detecting the light emission of the point light source. The three-dimensional position calculation means is driven based on the light emission detection signal sent from the light emission detection means.

Further, the three-dimensional position measuring apparatus according to the present invention uses at least two image pickup devices for obtaining a stereo image of the measurement target point, and a two-dimensional image pickup coordinate system in the stereo image, in which the two-dimensional position of the measurement target point is measured. Two-dimensional coordinate calculating means for calculating each coordinate, three-dimensional coordinate calculating means for calculating the three-dimensional coordinate of the measurement target point in the three-dimensional space coordinate system from the two-dimensional coordinate, and the two-dimensional coordinate and the three-dimensional coordinate It is learned in advance by using a plurality of reference points so that at least one of them is input and a measurement error between the three-dimensional coordinates calculated by the three-dimensional coordinate calculating means and the true three-dimensional coordinates of the measurement target point is output. The neural network measurement result correcting means is provided.

Further, when learning is performed in advance by using a plurality of reference points in the neural network measurement result correcting means, when a learning data set is created using the data of the reference points, the learning data having a larger learning error,
It is advisable to carry out learning of the neural network such that the learning data set is included in a large amount.

[0014]

In the three-dimensional position measuring apparatus according to the present invention, the point light source light emission detecting means detects the light emission of the point light source attached to the measuring point teaching means, and the detection signal is sent to the measurement control means, whereby the high speed 3 The calculation for calculating the dimensional position is started.

Further, the three-dimensional position measuring apparatus according to the present invention corrects the measurement result by using the neural network, so that the non-linear error such as the distortion of the lens is easily and accurately corrected.

In the above neural network, when learning is performed in advance using a plurality of reference points, learning data having a larger error is included in the learning data set so that the learning is performed. It is possible to output the learning data with less error.

[0017]

【Example】

Example 1. An embodiment of the present invention will be described below with reference to the drawings. In FIG. 1, 1 to 7 and P are exactly the same as those in the conventional device. 8 is an LED lighting switch attached to the probe 5, 9 is an LED light emission detection device, 1
Reference numeral 0 is a measurement control device.

Next, the operation will be described. In FIG. 1, when measuring the three-dimensional position of the measurement point P of the work 4,
After the pointing point 6 of the coordinate teaching probe 5 is placed in contact with the measurement point P, the LED lighting switch 8 is pressed to turn on the LED 7. By pressing the switch and turning on the LED in this way, it is possible to measure the three-dimensional shape of a easily scratched metal or a soft object such as cloth. Then, the LED light emission detection device 9 including a photodiode detects the lighting of the LED, sends a light emission detection signal to the measurement control device 10, and notifies the measurement control device 10 that the LED has lighted. The measurement control device drives the three-dimensional position calculation device 3 immediately after receiving the light emission detection signal and drives the three-dimensional position calculation device 3
Captures a stereo image of the coordinate teaching probe 5 with the TV cameras 1 and 2. Then, the three-dimensional position of each LED 7 is calculated based on the bright spot parallax of the LEDs in the left and right images of the stereo image. LE attached to teaching probe 5
Since the positional relationship between D7 and the designated point 6 is known in advance, LE
The three-dimensional position of the designated point 6 can be calculated from the three-dimensional position of D7, and as a result, the three-dimensional position of the measurement point P is obtained. At this time, if three or more LEDs 7 having a known positional relationship with the pointing point 6 are attached to the teaching probe 5,
The indicating point 6 does not have to be arranged on a straight line.

By the way, in the above-mentioned embodiment, the TV cameras 1 and 2 and the LED light emission detection device 9 are provided with a filter which transmits only the wavelength of the light emitted by the LED 7 attached to the teaching probe 5, so that the influence of ambient light is exerted. It is possible to configure a three-dimensional position measuring device that is not affected by the above. Also, L
By using an LED that emits light having an infrared wavelength, for example, a wavelength of 950 nm, for the ED 7, it is possible to obtain a three-dimensional measuring device that is less affected by ambient light.

Although the three-dimensional position calculation device 3 and the measurement control device 10 are separately configured in the above embodiment, they may be configured as one device. Furthermore, the luminescence detection device 9
May be configured as one device including.

Example 2. Next, a three-dimensional position measuring device equipped with a correction means will be described. FIG. 2 shows a three-dimensional position measuring apparatus according to another embodiment of the present invention, in which 1 and 2 are image pickup apparatuses (TV cameras), 31 is camera coordinate calculation means (two-dimensional coordinate calculation means), and 32 is three-dimensional coordinates. Calculation means, 33 is 3
It is a dimensional coordinate correction means (measurement result correction means).

Next, the operation will be described. The camera coordinate calculation means 31 obtains the camera coordinates (x ₁ , y ₁ ) and (x ₂ , y ₂ ) of the measurement target point in the cameras 1 and 2 from the stereo images taken by the TV cameras 1 and ₂ . The three-dimensional coordinate calculation means 32 uses the camera coordinates (x ₁ , y ₁ ) and (x ₂ , y ₂ ) obtained from the camera coordinate calculation means 31 and the attitude parameters of the camera to determine the three-dimensional coordinates (X _c , _Yc , _Zc ) is calculated. The three-dimensional coordinate correction means 33 is composed of a neural network, and has camera coordinates (x ₁ , y ₁ ), (x ₂ ,
y ₂ ) and the three-dimensional coordinates (X _c , Y _c , Z _c ) of the measurement target point obtained from them are input, and the true three-dimensional coordinates of the measurement target point and the three-dimensional coordinate calculation means 32 are obtained ( X _c , Y _c ,
The error (measurement error) (e _x , e _y , e _z ) with respect to Z _c ) is output. Here, for example, a multilayer feedforward type neural network is used as the neural network. The neural network acquires an input-output relationship by learning in advance using a plurality of reference points with known three-dimensional coordinates in the measurement space. That is, the three-dimensional coordinates of the reference point i are (X _ti , Y _ti , Z _ti ) and the camera coordinates obtained by the camera coordinate calculation means are (x _1i , y _1i ) (x _2i ,
y _2i ), where the coordinates obtained by the three-dimensional coordinate calculation means 32 are (X _ci , Y _ci , Z _ci ), (x _1i , y _1i ), (x
_2i , y _2i ), (X _ci , Y _ci , Z _ci ) are input, and (e _xi ,
e _yi , e _zi ) = (X _ti −X _ci , Y _ti −Y _ci , Z _ti −
Learning is performed in advance so as to output Z _ci ). Then, (X _c , Y _c , Z _c ), (ex _x , e _y , obtained as described above)
From (e _z ), (X _c + e _x , Y _c + e _y , Z _c + e _z ) is obtained as a three-dimensional position measurement result.

In the second embodiment, the camera coordinates (x ₁ , y ₁ ) are input as inputs to the three-dimensional coordinate correction means 33.
(X ₂ , y ₂ ) and the three-dimensional coordinates (X _c , Y _c , Z _c ) of the measurement target point obtained from them are used, but the camera coordinates (x ₁ ,
y ₁ ), (x ₂ , y ₂ ) only, or three-dimensional coordinates (X _c ,
Only Y _c , Z _c ) may be input.

In the second embodiment, the three-dimensional coordinate correction means 33 is constructed by using the multilayer feedforward type neural network, but the input / output nonlinear mapping such as the neural network having the feedback connection or the radial basis function is learned. Other methods that can be obtained by

Further, in the second embodiment, the camera coordinates obtained from the camera coordinate calculation means 31 are used as they are as the camera coordinates of the measurement target point and the reference point at the time of learning, but these coordinates are shown in the conventional example. It goes without saying that more accurate results can be obtained by substituting the geometrically corrected coordinates instead.

Example 3. In the above-mentioned three-dimensional position measuring apparatus, when learning of a neural network is performed using learning data based on a plurality of reference points, if learning is performed using each learning data evenly, an error in some learning data may cause another learning error. There were problems such as the error being larger than the data error, and the error being large when unlearned data was input. The third embodiment shows a measurement result correction method that solves such a problem and enables more accurate measurement. FIG. 3 is a block diagram showing a learning method used when learning the neural network used for the correction method, and 11 is a learning data storage unit for storing a plurality of learning data created by using the data of the reference points, 12 Is a learning data control unit that creates a learning data set using learning data, 13 is an error evaluation unit that evaluates learning errors, and 33 is a neural network. Further, FIG. 4 is a flowchart showing the learning method.

First, in step ST1 of FIG. 4, a learning data set used for learning of the neural network is created based on the learning data stored in the learning data storage unit 11 of FIG. The learning data i stored in the i-th learning data storage unit 11 is the input A _i = (a _i1 , a _i2 , ..., _Aim ) and Ai of the neural network.
The value to be output when the input is Bi = (b _i1 , b _i2 , _...
.., b _in ). Here, m and n are the numbers of inputs and outputs of the neural network, respectively. For example, A
_i is the camera coordinates and the three-dimensional coordinates x ₁ , y obtained from them
_It consists of seven values of ₁ , x ₂ , y ₂ , X _c , Y _c , and Z _c (m =
7) and Bi are the measurement errors e _x , e _y , and e _z between the true three-dimensional coordinates and (X _c , Y _c , Z _c ) obtained by the three-dimensional coordinate calculation means.
It consists of three values (n = 3). When the learning data set is created in step ST1, immediately after the learning is started, the learning data set is created such that all the learning data stored in the learning data storage unit 11 are included one by one, and after the learning progresses to some extent, , The learning data set includes at least one or more all the learning data stored in the learning data storage unit, and the output of the neural network and the learning data i
It is created such that the larger the error (learning error) of B _{i is,} the more the learning data i is included. Next, in step ST2,
A counter variable I for counting the number of times of learning using the learning data set created in step ST1 is initialized to 0.

In step ST3, learning data used for learning is selected from the learning data set. At this time, it is selected so that each learning data included in the learning data set is selected with equal probability. In step ST4, step ST
Neural network learning is performed using the learning data selected in 3. In step ST1, the learning data having a larger learning error is included in the learning data set more. Therefore, by selecting the learning data in this way,
The larger the learning error is, the more the learning data is learned.

At step ST5, 1 is added to the variable I,
In step ST6, it is determined whether or not I exceeds the upper limit N of the number of times of learning in the same learning data set. If not, the process returns to step ST3 to continue learning. When it exceeds, it progresses to step ST7. In step ST7, the sum of the errors (learning errors) between Bi in all the learning data stored in the learning data storage unit and the outputs when Ai forming a pair with Bi is input to the neural network is the allowable error E _sum. If it is within E _{sum, the} learning is ended, and if it is larger than E _sum , the process returns to step ST1 to continue the learning.

In step ST7 of the third embodiment, the sum of learning errors is used to determine whether or not the learning is to be ended. It may be a condition such that the learning error of the maximum data is within the allowable error E _max .

[0031]

As described above, according to the present invention, the light emission of the point light source is detected by using the point light source light emission detection means for detecting the light emission of the point light source of the measuring point teaching means, and the light emission is sent from the point light source light emission detection means. Since the three-dimensional position calculating means is driven based on the emitted light sensing signal, it is possible to detect the emitted light at high speed, and therefore, the operability is excellent and the device can be inexpensive.

The two-dimensional coordinates of the measurement target point and the above-mentioned 2
At least one of the three-dimensional coordinates calculated from the three-dimensional coordinates is input, and a plurality of reference points are used in advance so as to output the measurement error between the calculated three-dimensional coordinates and the true three-dimensional coordinates of the measurement target point. Since the three-dimensional position measuring device is configured to correct the measurement result by using the learned neural network, there is an effect that highly accurate measurement can be performed.

Further, when learning is performed in advance by the neural network measurement result correcting means using a plurality of reference points, when the learning data set is created using the data of the reference points, the learning data having larger learning error Since the learning data set is included in a large amount in the learning data set so that the learning of the neural network is performed, there is an effect that an output with a small error can be output even for unlearned data.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing a three-dimensional position measuring apparatus according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a three-dimensional position measuring apparatus according to a second embodiment of the present invention.

FIG. 3 is a block diagram showing a learning data control system according to a third embodiment of the present invention.

FIG. 4 is a flowchart showing a learning data control method according to a third embodiment of the present invention.

FIG. 5 is a configuration diagram showing a conventional three-dimensional position measuring device.

FIG. 6 is an explanatory diagram of a three-dimensional position correction method in a conventional three-dimensional position measuring device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Imaging device 2 Imaging device 3 Three-dimensional position calculation device 5 Measuring point teaching means 6 Pointing point 7 Point light source 9 Point light source emission detection device 10 Measurement control device 11 Learning data storage unit 12 Learning data control unit 13 Error evaluation unit 31 2 Dimensional coordinate calculating means 32 Three-dimensional coordinate calculating means 33 Neural network measurement result correcting means

Claims (3)

[Claims] 1. A measuring point teaching means comprising a pointing point indicating a measuring point and a plurality of point light sources having a known fixed positional relationship with respect to the pointing point, and a stereo of the measuring point teaching means. At least two image pickup devices for obtaining images, three-dimensional position calculation means for calculating the three-dimensional position of the measurement point based on stereo images from these image pickup devices, and point light source light emission detection for sensing light emission of the point light source A three-dimensional position measuring apparatus comprising: a means and a measurement control means for driving the three-dimensional position calculating means based on a light emission sensing signal sent from the point light source light emission detecting means. 2. A two-dimensional coordinate calculating means for calculating two-dimensional coordinates of the measurement target point in a two-dimensional imaging coordinate system in the stereo image, wherein at least two imaging devices obtain a stereo image of the measurement target point. A three-dimensional coordinate calculating means for calculating the three-dimensional coordinate of a measurement target point in a three-dimensional spatial coordinate system from the two-dimensional coordinate, and at least one of the two-dimensional coordinate and the three-dimensional coordinate are input to calculate the three-dimensional coordinate. The neural network measurement result correcting means learned in advance using a plurality of reference points is provided so as to output a measurement error between the three-dimensional coordinates calculated by the means and the true three-dimensional coordinates of the measurement target point. Dimensional position measuring device. 3. When the learning is performed using a plurality of reference points in advance by the neural network measurement result correcting means according to claim 2, a learning error of a learning error is generated when a learning data set is created using the data of the reference points. A method for correcting a three-dimensional position measurement result in which a larger learning data is included in the learning data set so that the learning of the neural network is performed.