JP3315175B2 - 3D measuring device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a three-dimensional measuring device for obtaining three-dimensional information of an object using a captured image. Thereby, recognition, analysis, and the like of the object can be performed.

[0002]

2. Description of the Related Art FIG. 5 shows a conventional three-dimensional measuring device (Hamano et al .: a method for extracting three-dimensional environmental information by voting into space, IEICE Transactions, Vol.J75-D-II, pp.342- Three
50, 1992). In the figure, reference numeral 51 denotes a space divided into voxels, 52 denotes a voxel, 53 denotes a line of sight, 54 denotes a feature point, 55 denotes a captured image, 56 denotes a camera lens center, 57 denotes a three-dimensional feature point, and 58 denotes a false point. Represents a three-dimensional feature point.

[0003] A space 51 in which an object is present is divided into voxels 52, and a certain value is added to voxels passing through the line of sight 53 of the camera.
Since the number 3 intersects a lot, the voxel 52 is extracted as the three-dimensional feature point 57 of the object.

Based on this principle, attempts have been made to devise a method of determining a value to be substituted into a voxel through which the line of sight passes, or to devise a method of determining a three-dimensional feature point from the voxel value (Kawato: Back projection twice). Extraction of 3D Information from Multiple Images by Image Method, IPSJ Research Report, CV-79-7, 1992
).

[0005]

However, in the conventional three-dimensional measuring device, a constant value is added to all voxels through which the line of sight passes, and three-dimensional feature points are simultaneously extracted. As shown in FIG. 5, the processing of the points interferes with each other to generate a false three-dimensional feature point 58 or the number of images required for measurement (ie, the number of viewpoints).
There was a problem that many were required.

[0006]

SUMMARY OF THE INVENTION In order to solve this problem, the present invention provides an apparatus for measuring the three-dimensional shape of an object using a captured image by using cameras from a plurality of viewpoints for the object. A means for inputting an image, a means for storing the obtained image, a means for extracting a feature point of an object from the image, and a plurality of viewpoints using camera parameters at the time of capturing the image and feature points on the image. Each feature point on the image from
Viewpoint corresponding means for obtaining a line of sight (the line passing through the characteristic points on the camera's lens center and an image) which is divided into voxels composed of three-dimensional pixel space in the presence of an object, the contains no visual axis for each line of sight There is an intersection between each line of sight from
Seeking cross voxels standing, means for determining the number of intersections of times at each intersection voxel, the resulting object from crossing frequency was 3
And a means for determining a dimensional feature point.

[0007]

According to the present apparatus, based on the property that one line of sight corresponds to one three-dimensional feature point, a voxel where each line of sight intersects with all lines of sight from viewpoints not including the line of sight, The number of intersections in the voxel is obtained, and one three-dimensional feature point corresponding to the line of sight is determined from the obtained number of intersections. Therefore, processing for each of the three-dimensional feature points does not interfere with each other, and false three-dimensional feature points are less likely to occur, and the number of images required for measurement, that is, the number of viewpoints can be reduced.

[0008]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a three-dimensional measuring apparatus according to one embodiment of the present invention. In the figure, reference numeral 10 denotes an imaging unit, 11 denotes an image memory unit, 12 denotes a feature point extraction unit, 13 denotes a visual line extraction unit, 14 denotes an intersection extraction unit, 15 denotes a three-dimensional feature point determination unit, and 16 denotes a control unit.

[0009] The imaging unit 10 inputs images from a plurality of viewpoints. The imaging unit 10 can be constituted by a plurality of cameras corresponding to the number of viewpoints for observing the object, or one movable camera. Hereinafter, an example in which a plurality of cameras are used will be described.

[0010] The image memory unit 11 accumulates images of each viewpoint taken by the imaging unit 10. The feature point extraction unit 12 extracts feature points from the image stored in the image memory unit 11.
Various features such as edges can be considered as feature points.
Any of them may be used. For example, “JFCanny: A computational approach to edge d
etection, IEEE Trans.Pattern Analysis and Machine
Intelligence, Vol. 8, pp. 679-698, 1986 ”. The method of extracting such feature points is well known, so detailed description is omitted here. I do.

The line-of-sight extraction unit 13 provides a camera lens center (x _o , y _o , z _o ) and a feature point (x _p , y) for each viewpoint.
_p, an equation of three-dimensional straight line passing through _{z p) (x, y,} z). That is, the following equation (1) is obtained.

[0012]

(Equation 1)

FIG. 2 is a diagram for explaining the processing of the intersection extraction unit 14 shown in FIG. In the figure, 20 is a space divided into voxels, 21 is a voxel, 22 is a line of sight for finding an intersection, 2
3 is a line of sight from a viewpoint that does not include the line of sight for finding the intersection,
Indicates a feature point, 25 indicates a captured image, 26 indicates a lens center of the camera, and 27 indicates an intersecting voxel. The space divided into voxels is determined by giving both end points (x1, y1, z1) and (x2, y2, z2) of the diagonal and the number of divisions S.

The intersection extracting unit 14 divides the space in which the object is present into voxels 21, and voxels 27 at which all the sight lines 23 from a viewpoint not including the sight line intersect with one sight line 22, and the number of intersections. Ask for.

Various specific procedures for calculating the intersection voxel and the number of intersections are conceivable. 3 and 4 are flowcharts showing one procedure. The flow diagram viewpoint number i (i = 1, ..., N), number j (j = 1, ..., M i) the line of sight of each viewpoint by attaching a, j-th line of sight in the i-th viewpoint _Is expressed as L _ij , the calculation procedure of the intersection voxel and the number of intersections with respect to the line of sight L _pq is shown. (A) to (n) in the following description correspond to (a) to (n) shown in FIGS.

(A) The position of a voxel is represented by (x, y, z), and an array representing the number of intersections of the voxel is represented by v (x, y, z).
z). x, y, and z take values from 0 to S-1, respectively. First, all elements of the array v (x, y, z) are initialized to zero.

(B) The viewpoint number i is set to 1. (c) It is determined whether or not the line of sight L _pq for finding the intersection includes the viewpoint i. If yes, go to processing (m). If not included, the intersection voxel and the number of intersections with respect to the line of sight L _pq are obtained by the following processes (d) to (l).

(D) Set the line-of-sight number j at the viewpoint i to 1. (e) Let w (x, y, z) be an array representing the presence or absence of an intersection at the viewpoint i of voxel (x, y, z). All the elements of the array w (x, y, z) are initialized to “false”. Here, x, y, and z take values from 1 to S-1, respectively.

(F) Line of sight L_pqAnd gaze L_ijFind the intersection with
You. The obtained intersection is (x_c,y_c, Z _c). (g) If there is an intersection, execute processes (h) and (i), and
If not, proceed to processing (j).

(H) A voxel (x ' _, y', z ') corresponding to the intersection is obtained as follows. The value of x ′ is [S * | x _c −x1 | / | x2-x1 |]. The value of y ′ is [S * | y _c −y1 | / | y2-y1
|].

Let the value of z 'be [S * | z _c -z1 | / | z2
−z1 |]. Here, [] represents a Gaussian symbol. (i) Since voxel (x ′ _, y ′, z ′) has an intersection, w (x ′ _, y ′, z ′) is set to “true”.

(J) j is incremented by 1, and the next line of sight is processed. (k) The processes (f) to (j) are repeated until the process for all the sight lines L _ij of the viewpoint i is completed. When the processing is completed, the process proceeds to the next processing (l).

(L) For each voxel (x, y, z), it is checked whether w (x, y, z) is “true”. If “true”, the number of intersections v (x, y) is determined. , Z) is increased by one. After that, it proceeds to processing (m).

(M) i is incremented by 1, and the next viewpoint is processed. (n) Determine whether the processing has been completed for all viewpoints, and if not, return to processing (c) and repeat the same processing. When the processing for all viewpoints is completed, the processing for obtaining the intersection voxel and the number of intersections is ended.

The three-dimensional feature point determination unit 15 determines a three-dimensional feature point from the intersection voxels obtained from the intersection extraction unit 14 and the number of intersections obtained by the above processing. Various decision procedures are conceivable. For example, an intersection voxel having the maximum number of intersections is extracted, and if the value is equal to or greater than a given threshold, the extracted intersection voxel is set as a three-dimensional feature point.

The control section 16 controls the entire apparatus, and operates the respective sections as described below. First, a test object is placed in front of the imaging unit 10, and a human instructs the control unit 16 to calibrate the imaging unit 10. Control unit 16
Obtains an image of the test object from the imaging unit 10 and stores it in the image memory unit 11. Next, the control unit 16 obtains the relationship between the camera parameters and the cameras (the relationship between viewpoints) using the obtained images and stores the relationship in the control unit 16. Specific calculation procedures for camera parameters are described in “Iguchi, Sato: 3D Image Measurement, pp.91-99, Shokodo, 1990” and “Kanaya: Image Understanding,
pp. 79-82, Morikita Publishing Co., Ltd., 1990. By using the relationship between the camera parameters and viewpoints, a unified coordinate system (called the world coordinate system) is used.
The coordinates of the feature points and the coordinates of the center of the lens of the camera can be expressed.

Next, the object to be measured is placed in front of the imaging unit 10, and a human instructs the control unit 16 to perform measurement. Control unit 16
Instructs the imaging unit 10 to operate and store images from each viewpoint in the image memory unit 11. Next, the control unit 1
According to the instruction of No. 6, the feature point extracting unit 12 extracts feature points from the obtained image, and outputs the image number (the image number corresponds to the viewpoint) and the coordinates of the feature point to the visual line extracting unit 13.
Next, in response to an instruction from the control unit 16, the line-of-sight extraction unit 13 inputs the relationship between the camera parameters and the viewpoint from the control unit 16, inputs the image number and the feature point coordinates from the feature point extraction unit 12, And the line-of-sight equation are output to the intersection extraction unit 14.

Next, the control unit 16 uses the first line of sight of the first viewpoint as the line of sight for calculating the intersection.
, And the intersection extraction unit 14 outputs the intersection voxel and the number of intersections to the three-dimensional feature point determination unit 15. And the control unit 1
6 operates the three-dimensional feature point determination unit 15 to output a three-dimensional feature point corresponding to the first line of sight of the first viewpoint.

Next, the control unit 16 uses the second line of sight of the first viewpoint as the line of sight for calculating the intersection,
, And the intersection extraction unit 14 outputs the intersection voxel and the number of intersections to the three-dimensional feature point determination unit 15. Then, the control unit 16 operates the three-dimensional feature point determination unit 15 to output a three-dimensional feature point corresponding to the second line of sight of the first viewpoint. When the three-dimensional feature points corresponding to all the lines of sight are calculated in the same manner, the control unit 16 terminates the operation of the entire apparatus.

Although the present invention has been described in detail with reference to the embodiments, it is needless to say that the present invention is not limited to the above-described embodiments and can be variously modified without departing from the scope of the invention. .

[0031]

As described above, according to the present invention,
Based on the property that one line of sight corresponds to one three-dimensional feature point, a voxel at which each line of sight intersects with all lines of sight from viewpoints not including the line of sight and the number of intersections at each intersection voxel are determined. Since one three-dimensional feature point corresponding to the line of sight is determined from the obtained number of intersections, processing for each of the three-dimensional feature points does not interfere, and false three-dimensional feature points are less likely to occur. Further, in order to suppress false three-dimensional feature points, the number of images required for measurement, that is, the number of viewpoints can be reduced.

[Brief description of the drawings]

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

FIG. 2 is a diagram for explaining processing of an intersection extraction unit shown in FIG. 1;

FIG. 3 is a flowchart showing a procedure for calculating an intersection voxel and the number of intersections in the embodiment of the present invention.

FIG. 4 is a flowchart showing a procedure for calculating an intersection voxel and the number of intersections in the embodiment of the present invention.

FIG. 5 is a diagram illustrating the principle of a conventional technique.

[Explanation of symbols]

 REFERENCE SIGNS LIST 10 imaging unit 11 image memory unit 12 feature point extraction unit 13 gaze extraction unit 14 intersection extraction unit 15 three-dimensional feature point determination unit 16 control unit 20 space divided into voxels 21 voxel 22 gaze to find intersection 23 gaze to find intersection Line of sight from viewpoints not included 24 Feature points 25 Captured image 26 Lens center of camera 27 Intersecting voxel

Continuation of front page (58) Fields investigated (Int.Cl. ⁷ , DB name) G01B 11/00-11/30 G06T 1/00 G06T 7/00

Claims (1)

(57) [Claims] An apparatus for measuring a three-dimensional shape of an object using a captured image, means for inputting an image to the object from a plurality of viewpoints using a camera, and means for storing the obtained image And means for extracting the feature points of the object from the image, and using the camera parameters at the time of capturing the image and the feature points on the image to correspond to each feature point on the image from multiple viewpoints.
Means for determining a line of sight that divides the space existing of the object voxels including three-dimensional pixel, each from the perspective without the visual axis for each line of sight
Find the intersection voxel where the intersection of the line of sight and the line of sight exists.
Because, means for determining the number of intersections of times at each intersection voxel, three-dimensional measurement apparatus comprising: a means for determining the three-dimensional feature point of the object from the crossing frequency obtained.