JP2576243B2 - Range image acquisition method and apparatus - Google Patents

Description

Description: FIELD OF THE INVENTION The present invention relates to a range finder, and more particularly to a range image acquisition method and apparatus in which a measurement error is small even if the surface reflectance of an object is not uniform and an image input time is short. Things.

(Prior Art) When operating an object by driving a robot arm or the like, it is necessary to grasp a three-dimensional position of the object. There is also a demand for performing three-dimensional measurement of an object such as a human body, which cannot be expected to remain stationary for a long time. At this time, for example, it is known to use a range finder capable of projecting a spectrum pattern and shortening an image input time to obtain a distance to an object plane for each pixel.

The range finder known as a conventional example will be described with reference to FIGS. 3 and 4.
7414, Japanese Patent Application No. 62-195379). In FIG. 4, a light beam collimated from the light source 1 through the lens 2 is split into red to blue-violet spectrum light by the diffraction grating 3 and projected on the object 5. A light having a wavelength λ is a point P (β from the Z-axis direction) on the object surface at an angle θ from the normal of the diffraction grating (α from the Z-axis direction).
At an angle position). This passes through a lens 61 in the camera 6 and is divided into two paths by a half mirror 62, and forms an image on sensors 63 and 64 having different spectral sensitivities, and is sensed at a position x on the sensor. This image is sent to the image calculation device 8 via the signal lines 65 and 66, and a distance image is obtained.

Assuming that the coordinates of the point P are (X, Z), the distance Z from the camera to the object can be obtained by Expression (1).

Where X ₀ is the X coordinate of the center of radiation, and Z ₀ is Z of the center of radiation.
Coordinates.

The value of the angle β can be known from the position x on the image sensor, but the value of the angle α is the wavelength λ of the sensed light, and the angle θ is obtained by the following equation. It can be obtained by knowing the angle.

sin θ ₀ −sin θ = ± n ^* λ / d (2) where θ ₀ is the angle between the light incident on the diffraction grating and the normal,
d is the interval between the diffraction gratings, and n is the order, which is usually 1.

Regarding the wavelength of the light reflected at each point of the object, the camera 6 is regarded as a color camera, two of which have different spectral sensitivities σ ₁ (λ) and σ ₂ (λ), and the point P is imaged. When the output of each pixel is I ₁ and I ₂ respectively, the output ratio Holds, so the function By determining σ ₁ (λ) and σ ₂ (λ) such that is a monotone function of λ, the wavelength λ can be obtained from the output ratio R.

λ = f ⁻¹ (R) (5) The wavelength λ thus obtained is used to obtain the angle α.

FIG. 3 shows an example of the configuration of the image calculation device 8. In FIG. 3, the two images input to the image calculation device 8 are respectively
The signals are converted by the A / D converters 81 and 82, and R of the equation (3) is obtained by the ratio image calculation device 83. Λ is obtained by the function operation unit 84 that performs the operation of Expression (5), and Expression (2) and Expression (1) are obtained.
Finally, a distance image can be obtained by the distance calculation unit 85 that performs the calculation of.

(Problems to be Solved by the Invention) When trying to know the three-dimensional position of an object with high accuracy using the above-described conventional technology, it is desired that the reflectance of points on the object be uniform. . If there is a point on the object with a low reflectance, the image output at that point will be small. This is A /
Since the quantization error at the time of the D conversion becomes large, the error when the pixel output ratio R is obtained by the equation (3) becomes large. For this reason, the position of a point having a large reflectivity can be obtained with high accuracy, but the measurement error becomes large at a point having a small reflectivity. Even if the reflectance of the object is uniform, the dynamic range of the sensor output is not fully utilized, so that the measurement error is also increased due to the quantization error.

An object of the present invention is to provide a range finder capable of obtaining a distance to an object for each pixel in real time with a small measurement error, excluding the above-described drawbacks of the range finder.

(Means for Solving the Problems) According to the first invention, a spectral pattern is projected on an object to be measured, and an image of the object is input by sensors having different spectral sensitivities. Take the largest of the plurality of images as the pixel output corresponding to a point, spatially modulate the brightness of the spectral pattern so that it can fully use the dynamic range of the sensor, and then re-enter the image input The distance image obtaining method is characterized in that the distance between the image input means and the object plane is obtained for each pixel by performing an operation between a plurality of images obtained by performing the above.

According to the second aspect of the present invention, the spectral pattern projecting means for projecting the spectral pattern into space, and spatially modulating the spectral pattern by placing the spectral pattern projecting means between the spectral pattern projecting means and the object to be measured. Pattern modulation means, image input means for performing image input by sensors having different spectral sensitivities, pattern modulation drive means for driving the pattern modulation means based on an image obtained from the image input means, and a plurality of obtained An image calculation means for performing calculation between images and obtaining a distance from the image input means to the object plane at each pixel.

(Operation) The brightness of a point on the object that is actually obtained is not the brightness of the imaging plane corresponding to the point, but a value obtained by A / D conversion of a sensor output placed on the imaging plane. Since a digital signal essentially has an error of ± 1, the following error dR is expected in the value of R obtained by the equation (3).

Here, the term enclosed in the first {} variation of R when I ₁ is changed by 1, and the second term corresponds to the variation of R when the I ₂ changes by 1 . Since each occurs independently, dR is the square root of the sum of squares of each change.

A / D converters usually used in color cameras have an 8-bit output. That it is, can be expressed by 2 ⁸ -1 brightness from 0. When a certain point on the object is imaged by the output of about 2 ⁷ sensors, errors of R is approximately the equation (6) 0.
You can ask for 14%. If the monotone function f (λ) of λ is a linear function, then F ⁻¹ (R),
Or, the value of α obtained therefrom fits in with the same deviation. Eventually, the measurement error can be obtained by the following equation.

When the estimated increase the error ignoring -Z ₀ is the second term of the numerator, at most d [alpha] / {(tan of this formula formula (1)
−tanα) * cos ² α｝ times, and if the diffraction grating, the object, and the camera are in a positional relationship where the inside of ｛｝ almost satisfies 1, the measurement error should still be within the same deviation. . However, when another point on the object is a ^one-4 about the sensor output of the eighth current point, R is calculated to have an error of about 1.1 percent as well, the measurement error is about 8 It turns out that it becomes twice.

In such a dark point, the measurement error can be reduced by brightening the spectrum pattern again and increasing the sensor output.However, simply increasing the brightness of the light source will reduce the range of the sensor output at the originally bright point. You can't do it because it's over. For this purpose, the problem is solved by spatially modulating the spectral pattern, illuminating dark spots with dark patterns and illuminating dark spots with bright patterns, and effectively utilizing the dynamic range of the sensor output.

The larger of the _two sensor outputs I ₁ and I ₂ corresponding to the point P on the object is taken. When this want to I _max, the brightness of the spectral pattern may be increased by the following magnification M.

I _max is a value that makes full use of the capability of the A / D converter, for example, 8
If it is a bit, set it to 255. However, the full range has a problem of the accuracy of the device for controlling the brightness, and it is realistic to take a slightly smaller value, for example, 200th.

This method can be realized by, for example, placing a liquid crystal display used in a small television or the like on a path through which a spectral pattern passes, and spatially modulating the pattern by the above magnification. Since this scan is very simple and the reaction of the liquid crystal is sufficiently fast, it is possible to measure even an object with a non-uniform reflectivity even at a low reflectivity without losing the advantage of the short image input time of the prior art. It is possible to obtain a range finder capable of suppressing an error low. Further, the measurement error can be improved even for an object having a uniform reflectance.

(Embodiment) An embodiment of the present invention will be described with reference to FIG. 1 and FIG. In FIG. 1, a light beam collimated from a light source 1 through a lens 2 is split by a diffraction grating 3 into red to blue-violet spectrum light. This spectrum light is projected on an object 5 after passing through a liquid crystal filter 4 which is one of the pattern modulation means. The light having the wavelength λ is projected onto a point P (located at an angle β from the Z-axis direction) on the object surface at an angle θ from the normal (an angle α from the Z-axis direction). This is camera 6
The light passes through a lens 61 inside, is divided into two paths by a half mirror 62, forms an image on sensors 63 and 64 having different spectral sensitivities, and is sensed at a position x on the sensors. The sensor output is sent to A / D converters 81 and 82 in the image processing device 8 via signal lines 65 and 66 and is converted into digital signals.
Sent via 87. FIG. 2 shows a configuration example of the liquid crystal filter driving device. In the two transmitted images, M is calculated for each point on the object by the magnification image calculation unit 71 that performs the calculation of the equation (8) in each pixel, and the liquid crystal filter driving unit 72 uses the signal line Driving the liquid crystal filter 4 via 72 spatially modulates the spectral pattern. Thereafter, a distance image with a small measurement error can be obtained from the image input again by the image calculation device 8.

The liquid crystal filter 4 and the liquid crystal filter driving device 7 in the above example can realize the present invention without any trouble as long as they satisfy the above-mentioned functions.

(Effect of the Invention) According to the above invention, it is possible to realize a range finder capable of obtaining a distance to an object for each pixel in real time with a small measurement error. As a result, in an environment having various reflectivities and an environment where such objects are mixed,
The dimensional position can be obtained with high accuracy. In addition, various applications are expected such as accurate measurement can be performed on a target such as a human body that cannot be expected to be stationary for a long time.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, FIG. 2 is a block diagram of a liquid crystal filter driving device which is one of pattern spatial modulation means used in the embodiment of the present invention, and FIG. FIG. 4 is a block diagram of a range finder known as a conventional example. DESCRIPTION OF SYMBOLS 1 ... Light beam, 2 ... Lens, 3 ... Diffraction grating, 4 ... Liquid crystal filter, 5 ... Object, 6 ... Camera, 7 ... Liquid crystal filter driving device, 8 ... Image processing device, 61 ... lens,
62: Half mirror, 63, 64: Sensor, 65, 66: Sensor output, 71: Magnification image calculation unit, 72: Liquid crystal filter drive unit, 73: Liquid crystal filter drive output, 81, 82 ... … A / D converter, 83… ratio image calculation unit, 84… function calculation unit, 85…
Distance calculation unit, 86,87 ... Digital image output.

Claims (2)

(57) [Claims] 1. A method of projecting a spectral pattern onto an object to be measured, inputting an image of the object by sensors having different spectral sensitivities, and obtaining a pixel output corresponding to each point on the object, the largest pixel output among a plurality of images. After taking and spatially modulating the brightness of the spectral pattern so that it can use the full dynamic range of the sensor,
A distance image obtaining method, wherein the distance between the image input unit and the object plane is obtained for each pixel by performing an operation between a plurality of images obtained by performing the above image input again. 2. A spectrum pattern projection means for spatially projecting a spectrum pattern, and a pattern modulation means for spatially modulating a spectrum pattern between the spectrum pattern projection means and an object to be measured. When,
Image input means for inputting an image by sensors having different spectral sensitivities, and based on an image obtained from the image input means,
Pattern modulation driving means for taking the largest of a plurality of images as a pixel output corresponding to each point on the object and driving the pattern modulation means so that it can fully use the dynamic range of the sensor; And a calculation unit for calculating a distance between the image input unit and the object plane at each pixel.