JPH02245610A - Object position measuring system by binocular stereoscopy - Google Patents

Abstract

PURPOSE:To achieve a higher speed of a corresponding processing by performing a binocular stereoscopy by the use of a plurality of cameras to search points corresponding to each other on a picture obtained with the respective cameras. CONSTITUTION:An area dividing section 10 divides right and left input images obtained by performing binocular stereoscopy into areas of the same texture independently and the areas as sides are labeled. Finally, the dividing section 10 outputs a label image as the entire image labeled. A structure generating section 20 judges an overlapping relation with other sides for each of the sides labeled by the division section 10 to infer a vertical relation of the individual sides over the entire image. A coarsely corresponding section 30 selects candidates of right and left sides to be corresponded from a structure of right and left images obtained by the generating section 20. A side corresponding section 40 performs a corresponding for each candidate of the right and left sides selected to calculate a mutual parallax. Then, based on the parallax obtained, a distance calculating section 50 calculates a distance between the cameras and the sides by a theory of trigonometrical survey.

Description

[Detailed Description of the Invention] [Summary] Performs binocular stereoscopic vision using multiple cameras, searches for mutually corresponding points on images obtained by each camera, and uses the principle of triangulation to determine the relationship between the camera and the object. With regard to object position measurement methods using binocular stereopsis to obtain the distance between The left and right images obtained by binocular stereoscopic vision from multiple cameras are each independently divided into areas with the same texture, and y9
A region dividing section labels each surface using one region as a surface, and for each surface labeled by the region dividing section, the overlapping relationship with other surfaces is determined, and the overlapping relationship of each surface is determined over the entire image. a structure generation unit that infers the vertical relationship; a rough correspondence unit that selects left and right plane candidates to be correlated from the structures of the left and right images obtained by the structure generation unit; The surface mapping section calculates the mutual parallax by mapping each surface candidate, and the surface mapping calculates the distance between the camera and the surface using the principle of triangulation based on the parallax obtained in the section. and a distance calculation unit that calculates the distance.

[Industrial application field]

The present invention relates to an object position measurement method that is used, for example, as a visual aid for robots and self-driving cars, or to determine the three-dimensional shape of an object. , relates to a binocular stereoscopic object position measurement method that searches for mutually corresponding points on images obtained by each camera and obtains the distance between the camera and the object using the principle of triangulation.

[Conventional technology]

FIG. 16 is a diagram showing the principle of conventional binocular stereopsis.

In the figure, left and right cameras 101 and 102 first obtain left and right images of the target object, and element extraction units 103 and 104 extract elements such as edges from these left and right images. Then, the mapping unit 105 performs mapping between the left and right elements for each element, and then the distance calculation unit 106 calculates the parallax for each of the correlated left and right elements,
From this parallax and camera parameters (inter-camera distance, focal length, etc.), the distance between the camera and the object is calculated using the principle of triangulation.

Here, the principle of the correspondence for each element by the correspondence unit 105 will be briefly explained based on FIG. 17. In addition, here, the scanning line at YW7o on the left and right images ML, Ms+! Consider the correspondence between edges above.

In the figure, if the number nL of edges on the scanning line 2 in the left image ML is 9 and the number n of edges on the scanning line 2 in the right image is 8, then nLXni is the edge association candidate. =72 combinations are possible. Search plane P for the surface covered by this combination of left and right edges
The intersection of edges on this search plane P will be called a node. Then, the problem of mapping is the search plane P
The above problem is to determine the route from node O to node M, and for this, for example, there is a method using dynamic programming. Katsuo Institute of Electronics and Communication Engineers Journal '85/4
Vol1. J68-D No. 4).

[Problem to be solved by the invention]

In the above-mentioned conventional binocular stereoscopic viewing technology, it is necessary to make correspondences between all the elements that satisfy the constraint conditions (Epipolar condition, etc.) between the left and right images. Therefore, when the number of elements n increases, There is a problem in that the amount of calculation increases in proportion to the amount of processing, and therefore it is not possible to achieve faster processing.

Furthermore, since the correspondence is made for the entire image, there is also the problem that if even one incorrect correspondence occurs, it will affect the subsequent correspondence.

SUMMARY OF THE INVENTION An object of the present invention is to speed up the mapping process and to minimize the effects of incorrect mapping.

[Means to solve the problem]

FIG. 1 is a diagram showing the principle configuration of the present invention, and FIG. 2 is an explanatory diagram thereof.

In Fig. 1, 10 independently divides the left and right human images obtained by binocular stereoscopic vision from multiple cameras into areas with the same texture (shading and pattern), and uses these areas as surfaces to divide each side into areas. This is an area dividing section that attaches labels. The area dividing unit 10 finally outputs a label image in which the entire image is labeled.

Reference numeral 20 denotes a structure generation unit that determines the overlapping relationship with other planes each time the area division unit 10 labels the area, and infers the vertical relationship of each plane over the entire image.

Reference numeral 30 denotes a rough correspondence unit that selects left and right plane candidates to be correlated from the structure of the left and right images (vertical relationship between individual planes) obtained by the structure generation unit 20.

Reference numeral 40 denotes a surface matching section that performs matching for each of the left and right surface candidates selected by the coarse matching section 30 and calculates mutual parallax.

Reference numeral 50 denotes a distance calculation unit that calculates the distance between the camera and the surface based on the parallax obtained by the surface mapping unit 40 and based on the principle of triangulation.

That is, in the present invention, first, as shown in FIGS. 2(a) and 2(b), when there are left and right images obtained from left and right cameras, these are independently divided into regions by the region dividing unit 1o and labeled. (1.2.3.4 etc. in the figure). Next, for these images, the structure generation unit 20 generates images in FIGS.
), perform structure generation to determine the vertical relationship of the surfaces. Next, as shown in FIG. 2(e), the rough correspondence unit 30 determines surface candidates to be mapped according to the vertical relationship between the left and right surfaces, and the resulting left and right surface candidates are On the other hand, the surface correspondence section 40 performs surface correspondence as shown in FIG. 2(f). Finally, based on the parallax, the distance between the camera and the surface is calculated using the triangle@quantity principle.

[For production]

According to the present invention, if there are overlapping surfaces when observing multiple objects (surfaces) with left and right cameras, each surface is divided by region division, and the vertical relationship of each divided surface is inferred by structure generation. By performing the mapping each time from the top surface to the bottom surface, mapping can be achieved at a speed proportional to the number of elements, and distance information and three-dimensional information of the object can be obtained at high speed. .

Furthermore, since correspondence is made for each area, even if an incorrect correspondence occurs, the effect will remain only on the relevant area and will not affect other areas.

On the other hand, if there are no overlapping surfaces when observing multiple objects (surfaces) with the left and right cameras, the vertical relationship of the surfaces is not extracted by structure generation, so ), and the speed of the mapping is the same as before, but by dividing the regions, the influence of incorrect correspondence will no longer be exerted on other regions.

〔Example〕

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

FIGS. 3 to 6 are block diagrams showing the configuration of an embodiment of the present invention with respect to the area division section 10, structure generation section 20, rough correspondence section 30, and surface correspondence section 40 shown in FIG. 1, respectively. It is a diagram. Each block will be explained in turn below.

(Region Dividing Unit 10) As shown in FIG. 3, the area dividing unit 10 includes a texture analysis unit 11 and a region labeling unit 2.

First, the texture analysis unit 11 divides the input image into regions with the same texture such as shading or pattern.

Next, in the region labeling section 12, regions with the same texture are called surfaces, and a different label is assigned each time.Finally, a label image in which the entire image is labeled is output (second (See Figures (a) and (b)).

(Structure Generation Unit 20) The structure generation unit 20, as shown in FIG. There is.

<Boundary Line Extraction Unit 21> The boundary line extraction unit 21 extracts the boundary lines of surfaces with the same label from among the label images obtained by the area division unit 10. The image obtained by extracting this boundary line is called a line drawing.

The label image and one stroke are images of the same size, and the boundaries of the same label correspond to the boundaries of line drawings.

Three-sided connection point extraction unit 22> The three-sided connection point extraction unit 22 constructs the line drawing.

A three-sided connection point where three boundaries intersect is extracted from each of the seven intersection points of the boundary lines.

Here, a specific example of the method for extracting the three-plane connection points will be explained based on FIG. 7.

A three-sided connection point, which is a connection point between three boundary lines, appears in four types of shapes as shown in FIG. 7(a). So first,
By scanning the line drawing using four types of templates having the same shape as these and performing matching, the positions of the three-sided connection points on the image are determined.

When this position is determined, a label (,L) is attached to the connection point.

For all three-sided connection points obtained by the above matching,
As shown in Fig. 7(b), three boundary lines with each connection point (consider connection point J3 as an example) as an end point each reach other connection points (connection points J2, J6, JT). 4 neighboring pixels are traced up to 4. The three boundaries traced in this way (Jz-Jz, Jl, J3-1, JT
Each boundary line (between Jl) is held as a pixel column, and is held as boundary line information together with the label of the connection point that is the end point. Furthermore, the connection between adjacent connection points via a boundary line is also recorded as one of the information for the next process.

Next, based on the coordinate position of the three-sided connection point on the image, the labels of the three sides making up the three-sided connection point are determined from the label image. Specifically, as shown in FIG. 7(C), starting from an arbitrary pixel (•) around the pixel (0) of the three-sided connection point, surrounding pixel values are checked clockwise. The first three types of labels that appear at this time are three-sided labels that constitute the three-sided connection point.

Next, find out which faces make up which boundaries.

For this purpose, as shown in FIG. 7(d), the three-sided label constituting each connection point is, for example, Jl (AI A3 A
I ) is expressed in the form of a list of three terms. Then, the surfaces forming adjacent connection points are compared. For example, when comparing the list of connection point J3 and its adjacent connection point J2, it is found that their common planes are A1 and A3. From this, it can be seen that either plane A1 or A3 is a plane forming the boundary line J2J3 J& or J2 Jl JT. Similarly, when comparing the list of connection point J3 and its adjacent connection point J6, since their common planes are A and AI, one of these planes AI and AI is the boundary line J2 J:l J6 Or JL, Jl J
It can be seen that this is a surface that constitutes T.

From these two results, as shown in FIG.
3 Jb , J2 Jl JT , and Jb Jl Jq.

Finally, for example, the boundary line cad (c, d are the labels of the connection points) formed by the surface B at the three-face connection point a is defined as (B c
d), for example, the connection point J3 is the seventh
As shown in figure (f), (AI J2 Jb) (A:
I J2 JT ) (AI Jb J)). This is expressed for all three-sided connection points and passed to the next T-type connection determination section 23 (FIG. 4) as three-sided connection point information.

<T-type connection determination unit 23> The T-type connection determination unit 23 first regards the three-sided connection points as three straight lines, and performs linear approximation for each. and,
The three straight lines thus obtained are called a T-type connection, and the overlapping relationship between the three surfaces is determined by the way this connection is made.

These processes will be specifically described below based on FIG.

First, from the boundary line information obtained by the three-sided connection point extracting section 22, the boundary line extending from each connection point is approximated by a straight line, and a straight line extending directly from the connection point is determined. As a linear approximation method, for example, Duda and Hart's splitting method can be used (Duda, R,0, and P,E,Har
t: “Pattern Recognition”
and 5 scene analysis, Ne
(see York, Wiley, 1973). - As an example, let us describe the case where this method is executed for connection point J3. In this case, first, as shown in FIG. 8(a), straight lines (Jl J2 , Jy J6 , J
3J? ). Then, the connection point J3 and the connection points J2, J
6. Based on the boundary line information between JT and JT, find the point farthest from the above straight line (Jl, J2 in FIG. 8(a)), and compare the distance between them with a predetermined darkness value. If the distance is greater than the darkness value, the point J
Connect I J2 and each connection point J2, J3, and J6. Next, the newly created straight line J3 J”, J” J2, J3
For J' and J' J6, find the farthest point in the same way as above, compare the distance between them with the threshold, and if it is greater than the darkness value, connect a new straight line (see Figure 8). (
C)). In this way, by repeating the same process, the connection point J is finally reached as shown in FIG. 8(d).
Three straight lines extending directly from 3 J3 J", J3 J'
, J3 obtains J7. These three straight lines are called a T-type connection, and their intersection is called a T-type connection point.

Next, using the T-type connection which is the result of the linear approach described above, the angle formed by adjacent boundary lines at each connection point is determined. And among those angles, πrad (・180 @)
If we select the closest one and find the boundary line to which the two straight lines forming the angle belong, this boundary line belongs to the top surface. For example, regarding the connection point J3, as shown in FIG. 8(e), two straight lines J3 J2 and J3 J', two straight lines J3
J2 and J3J? , and two straight lines J3 J' and J3 J?
Find the angle formed by at, a2, a
3, since al is closest to πrad, the boundary line J2 J3 Jb is determined to be the boundary line belonging to the upper surface, as shown in FIG. 8(f).

Furthermore, a label for the top surface is obtained from the three-face connection point information obtained by the three-face connection point extraction unit 22 and the boundary line belonging to the above-mentioned top surface. For example, for connection point J3, three-sided connection point information (J3: (AIJZ
J4) (AzJzJt) (AsJ6J7)) is obtained, and as shown in Figure 8), the boundary line Jz J3
Since J6 is known to be the boundary line belonging to the upper surface, from these facts, the surface A1 is determined as shown in Figure 8 (g).
It can be seen that these are the upper surfaces of surfaces A3 and 'A5. However, the vertical relationship between surface A3 and surface A is unknown at this point. These pieces of information are expressed in the form of a three-item list as shown in FIG. 8 (5), and are transferred to the next surface relationship search unit 24 as T-type connection point information. In addition, in Fig. 8 (5), when the overlapping relationship of three surfaces is expressed as (a b c), surface a is the upper surface of surface bSc, and the relationship between surface C and surface C is unknown. shall be shown.

Plane relationship search unit 24> The plane relationship search unit 24 converts the overlapping relationship of the three planes at each T-shaped connection point obtained by the three plane connection determination unit 23 into an overlap relationship between two planes, and then The overlapping relationship between the images is propagated to all surfaces existing in the image, and the overlapping relationship in the entire image is determined. These processes will be specifically described below based on FIG. 9.

First, from the T-type connection point information as shown in FIG. 9(a),
As shown in Figure 9 (b), all the information is extracted for each surface, and from this, the upper surface of each surface is determined as shown in Figure 9 (C), and even though there is an adjacent relationship, the relationship is unknown. Find the surface that is. In addition, the Book of Kings, Figures 9 (b) and (C)
In , the arrows indicate an overlapping relationship (the head side of the arrow is the top surface), the solid line indicates the same surface, and the dotted line indicates a surface that is in an adjacent relationship but whose vertical relationship is unknown. FIG. 9 (
Output the dihedral amount relationship information as shown in d). In other words, in creating this dihedral relationship information, first,
For each surface, provide a column for describing the top surface (top surface column) and a column for describing adjacent surfaces with unknown relationships (unknown relationship column), and set N U L L as the initial value in each column. I'll keep it.

Thereafter, the first item (that is, the top surface) of the three-item list of T-type connection point information (see FIG. 9(a)) is registered as the second and third items of the two-sided amount relationship information. Subsequently, after the registration of the top surface is completed, for each surface, among the adjacent surfaces that are not registered in the registered top surface, if that surface is also not registered as a top surface (for example, surfaces A4 and A , etc.), the adjacent surface is registered in the unrelated douen.

Here, in the two-sided amount relation information shown in FIG. 9(d) above, if we look at surface A, for example, surface A+ is higher than surface A5, and Aa is the upper surface and is adjacent to surface A4. However, the overlapping relationship is unclear. Also,
For example, when looking at surface A, there is NULL in the top column.
Therefore, it can be seen that surface A+ does not have an upper surface.

Next, we summarize aspects that overlap and have unknown relationships. For example, the 9th
Assuming that the relationship between surfaces A4 and A5 is unknown as shown in FIG. 9(C), they are integrated as shown in FIG. 9(e). Then, by summarizing these relationships, Figure 9 (f)
As shown in the figure, the upper surfaces of both sides are registered as one surface. However, the aspect that overlaps with both (here As)
is registered as one surface. Information obtained by correcting the two-plane amount relationship information in this way is called inter-plane relationship information, and this information is used as the structure of the left and right images (see FIGS. Transfer to (Figure).

(Rough correspondence unit 30) As shown in FIG.
, Epipolar condition determination section 32, and brightness distribution determination section 33
The overall process flow is shown in the 10th section.
As shown in the figure.

Structure Searching Unit 31> The structure searching unit 31 finds faces (labels) that can be matched based on the structures of the left and right images obtained by the structure generating unit 20 (i.e., inter-surface relationship information indicating the overlapping relationship between the faces). (step at in FIG. 10). On this occasion,
Surfaces are combined starting from upper levels and working their way down to lower levels so that the order of the structure is reflected. In addition, the range of combinations is such that combinations of the same level are performed preferentially, and if there is a combination rejected by the Epipolar condition determination unit 32 or brightness distribution determination unit 33, which will be described later, the range of combinations is increased by that amount. shall be expanded.

Ebi-polar condition determination section 32> The Ebi-polar condition determination section 32 extracts the region corresponding to the label specified by the structure search section 31 from the left and right label images, applies the Ebi-polar condition to the region, and determines this condition. Determine whether the conditions are satisfied. The above Epipolar conditions include finding the maximum and minimum parts of the left and right surfaces in the Y-axis direction (step a2 in Figure 10), and determining whether there is a shared part between them (step a3 in Figure 10).
) is applied. Here, if the epipolar condition is satisfied, the label is sent to the next brightness distribution determining section 33, whereas if it is not satisfied, information indicating that it has been returned to the structure searching section 31 is returned.

Luminance Distribution Determining Unit 33> The brightness distribution determining unit 33 first extracts the region corresponding to the label satisfying the above epipolar condition from the left and right manual images, and obtains the luminance information of the region (step a4 in FIG. 10). Then, from this luminance information, the average values IL and IR of the luminance of the left and right regions are calculated (step as in FIG. 10).
). Subsequently, the difference (l IL -[Rl) is found and compared with the threshold value δ to determine the similarity of the brightness distributions of the left and right regions (step a6 in FIG. 10).

Here, if IIL IRI < δ, that is, if the left and right luminance distributions are similar, the left and right surfaces are considered as matching candidates and sent to section 4° for the next surface matching (10th
In step a7) of the figure, on the other hand, if they are not similar, information to the effect that it has been returned to the structure search unit 31 is returned.

(Plane correspondence section 40) As shown in FIG.
and a parallax calculating section 44, and the overall processing flow is shown in FIG.

<Bend Point Search Unit 41> The bend point search unit 41 searches for bend points where the slope of the boundary line suddenly changes, in order to segment the boundary of the surfaces into straight line segments for the left and right surface candidates obtained by the rough correspondence unit 30. Explore (
Step b+) in FIG.

At this time, the change in the slope of the boundary line is determined by, for example, the -curvature ("Pattern Information Processing" by Makoto Nagao, Corona Publishing, p.

82, etc.), and the point where the absolute value of the curvature is at least a certain value and locally maximum or minimum is defined as a bending point. This process is performed independently for the left and right surfaces.

Corresponding segment generation unit 42> The corresponding segment generation unit 42 first divides the boundary line of the surface at the bending point obtained by the bending point search unit 4L.
Step b2) in the figure, a straight line is applied between adjacent bending points instead of the divided boundary line (step bz in Figure 11). At this time, the error between the straight line and the boundary line between the bending points, that is, the distance between the point on the boundary line that is farthest from the straight line and the straight line, is compared with a predetermined threshold ε (first
Step b4 in Figure 1). Here, if the error is smaller than the threshold ε, the straight line is set as a segment (step b5 in FIG. 11), whereas if the error is greater than the dark value ε,
At the point on the boundary line that is farthest from the straight line, divide the straight line further and apply a new straight line. A series of straight line segments is obtained by repeating this process for the boundaries between the left and right surfaces.

Next, this straight line segment is compared left and right, and if there is no end point of the corresponding segment on the same Ebibora line, the segment is further subdivided to obtain left and right corresponding segments (Step B6 in Figure 11). ).

If the surfaces observed by the left and right cameras are in the same three-dimensional space, then if a three-dimensional plane is drawn from three corresponding points on the left and right images, then All other corresponding points should also be projected onto that three-dimensional plane. Projecting corresponding points onto a three-dimensional plane in this way is called plane constraint, and conversely, the plane observed by the left and right cameras is
If they are different planes in the dimensional space, even if other corresponding points are projected onto a three-dimensional plane spanned by three points that are considered to correspond, they will not be distributed on that three-dimensional plane. Using this principle, it is possible to determine whether the surfaces on the left and right images are the same surface in three-dimensional space.

Figure 12 shows a three-dimensional plane (a
x+by+cz=1) is observed with the left and right cameras, and the relationship between the corresponding points and the three-dimensional plane is shown below.

If the coordinates of a point P on a three-dimensional plane observed in the coordinate system of the camera C on the left are P (x, y, z), and the coordinates at Z=0 are A (X, Y), then the following is obtained. . On the other hand, if the distance between cameras is U, then Similarly, the coordinates of point P when observed in the coordinate system of the right camera C° are P(x',
', z'), and the coordinates at z'-o are B(X'',
If Y゛), then it becomes . So, by substituting equation 0 into the plane equation, we get: Therefore, from the formula, it can be expressed as. Therefore, from ■2, X' =
(1+ua)X+ubY+uc...■ is obtained.

Therefore, if three corresponding points are found, the equation of the plane is determined. The coordinates of the three points are (LY+) and (Xz
, Y2 ), (X3 , Y3 ) and solve the ■ equation, u u
u However, it becomes y 11 aX+bY+c, and the plane equation (ax+by+cz=1) is determined.

Based on these facts, the plane constraint unit 43 first determines a three-dimensional plane from the three corresponding points (step b in FIG. 11).
7). To determine whether other corresponding points are constrained on this plane, use the above equation 0 to project the corresponding points on the right onto the coordinate system on the left (step b8 in Figure 11), and then The deviation between the points is determined, and by comparing the average value of deviations at a plurality of corresponding points with the darkness value, it is determined whether the plane constraint is satisfied (step bq in FIG. 11).

Here, if the plane constraint is satisfied, the left and right segments are sent to the next parallax calculation unit 44, while if the plane constraint is not satisfied, the rejected segment is sent to the structure search unit 31 (Fig. 5). Return information.

Disparity Calculation Unit 44> The disparity calculation unit 44 calculates the disparity each time from the left and right segments that have passed through the plane restraint unit 43 (steps b + o in FIG. 11). As shown in Fig. 13, the parallax is
The difference (η
, -ξX) at the point (ηx, yo).

(Distance Calculation Unit 50) The distance calculation unit 50 calculates the distances Rt and Ri between the viewpoints of the left and right cameras and the object surface. For this purpose, first, the left and right camera coordinate systems and the camera head coordinate system are defined as shown in FIG. Here, the distance between the two viewpoints (baseline length) is 2a, the angle formed by both optical axes (convergence angle) is 2α, and for simplicity, the image of the object is projected parallel to the camera head coordinate system. shall be taken as a thing. Then, in Figure 15,
Since △ABC and ∆DEF are similar and DE DF AB 2a, the following holds true. Here, considering the transformation from the camera coordinate system to the camera head coordinate system in order to find the DF, (ηg,
77y) 4 (-a +77x /cosα,
0. 1. )(ξ8.ξ,)→(a+ξ, /co
scr, O, ξy ), On the other hand, the distances Lo and Ro when the left and right optical axes cross the X axis of the camera coordinate system and the χ axis of the camera head coordinate system can be expressed as follows. .

1,, = 77x j, anα Ro = ξXtancr From the above, the distances RL, R, l between the left and right camera coordinate systems and the optical axis to the object are: For the left camera: (1) If η8 is 0, then RL = DE LO (2) η≧
If it is 0, RL = DE + L.

For right camera: (1) If ξ8×O, then RR=DE+R.

(2) If ξ8≧0, RR=DE−R,) can be found.

Once the distance between the camera and the object surface is determined in this way,
The three-dimensional coordinates of the object surface can be determined by coordinate transformation from the left and right camera coordinate systems to the camera head coordinate system.

The coordinate transformation from the left camera coordinate system to the camera head coordinate system is as follows.

On the other hand, the coordinate transformation from the right camera coordinate system to the camera head coordinate system is as follows.

By using either of these, three-dimensional information about an object can be obtained.

As described above, according to this embodiment, a plurality of objects (
If there are any overlapping surfaces when the surfaces (surfaces) are observed with the left and right cameras, the area dividing section 10 divides each surface, and the structure generating section 20 infers the vertical relationship of each divided surface. Therefore, the coarse correspondence section 30 and the surface correspondence section 40 perform correspondence from the top surface to the bottom surface each time, thereby realizing correspondence at a speed proportional to the number of elements, and distance information and the cubic object of the object. You can quickly extract the original information. Further, since the correspondence is made for each area, even if an incorrect correspondence occurs in the area, the effect will be limited to the corresponding area and will not affect other areas.

On the other hand, if there are no overlapping surfaces when multiple objects (surfaces) are observed with the left and right cameras, the vertical relationship of the surfaces is not extracted by the structure generation unit 20, so the brute force of the left and right regions (
(the square of the number of elements),
The speed of matching is the same as before, but because area division is performed, there is no need to worry about incorrect matching affecting others.

(Effects of the Invention) According to the present invention, it is possible to significantly speed up the matching process, especially when there are surfaces in the input image that overlap with each other. Even if the
The negative effects of incorrect responses can be minimized.

[Brief explanation of drawings]

Figure 1 is a diagram showing the principle configuration of the present invention, Figures 2 (a) to (f) are diagrams explaining the principle of the present invention, Figure 3,
FIG. 4, FIG. 5, and FIG. 6 respectively show a region dividing unit 10 and a structure generating unit 20 according to an embodiment of the present invention. Rough correspondence part 3
0, and a block diagram showing the specific configuration of the surface correspondence section 40. FIGS. h) is a diagram for explaining the processing of the T-type connection determination section 23, FIGS. 9(a) to (f) are diagrams for explaining the processing of the surface relationship search section 24, and FIG. 11 is a flowchart showing the overall processing flow of the surface mapping section 40; FIG. 12 is a diagram showing the principle of plane constraint by the plane constraint section 43; FIG. 13 is a diagram showing the principle of calculating parallax by the parallax calculation unit 44, FIG. 14 is a diagram showing the camera head coordinate system and left and right camera coordinate systems, and FIG. 15 is a diagram showing the principle of distance calculation by the distance calculation unit 50. 16 is a diagram showing the principle of conventional binocular stereoscopic vision, and FIG. 17 is a diagram showing the principle of conventional correspondence. DESCRIPTION OF SYMBOLS 10... Area division part, 11... Texture analysis part, 12... Area labeling part, 20... Structure generation part, 21... Boundary line extraction part, 22... Three-sided connection point Extraction unit, 23... T-type connection determination unit, 24... Surface relationship search unit, 30... Rough correspondence unit, 31... Structure search unit, 32... Ebipolar condition determination unit, 33... - Brightness distribution determination section, 40... Plane correspondence section, 41... Bend point search section, - Corresponding segment generation section, - Plane constraint section, - Parallax calculation section, - Distance calculation section.

Claims (1)

[Scope of Claims] A region dividing unit (10) that independently divides left and right images obtained by binocular stereoscopic vision of a plurality of cameras into regions having the same texture, and labels each surface using the regions as surfaces.
and a structure generation unit (20) that determines the overlapping relationship with other planes for each plane labeled by the area division unit (10) and infers the vertical relationship of each plane over the entire image. And from the structure of the left and right images obtained by the structure generation unit (20),
Rough correspondence part (3
0), and a surface matching section (30) that performs matching for each of the left and right surface candidates selected by the coarse matching section (30) and calculates mutual parallax.
40), and a distance calculating section (50) that calculates the distance between the camera and the surface based on the principle of triangulation based on the parallax obtained by the surface matching section (40). Object position measurement method using stereoscopic vision.