JP2006214893A - Computer software program for measuring three-dimensional shape of object using object measuring method and computer system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To accurately measure the object of a complicated shape by enabling positioning of images even when a sampling interval by a range sensor is wide. <P>SOLUTION: This computer software program includes a process for generating point group data including information of a distance of a depth direction up to measuring points of the object respectively by measuring the object from a plurality of different directions using the range sensor (a), a process for detecting as an edge by coupling a prescribed point group by a difference between the distances of the depth direction of adjacent points by processing each point group (b), a process for converting the edge into a characteristic line (c), and a process for outputting geometric conversion of the object by positioning by using the detected characteristic line (d). <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an object measurement method for obtaining three-dimensional shape data of an object by measuring the object from a plurality of directions using a range sensor, and a computer software program for executing the object measurement method.

  In general, use of a range sensor is one effective means for restoring a real space building on a virtual space. The range sensor is a measuring device for measuring the distance in the depth direction from the sensor position to each point on the object surface as digital data. By obtaining an image including distance information from the range sensor, the object shape can be acquired as a point group in a three-dimensional space.

  By the way, it is known that the point cloud obtained from the range sensor has a problem related to occlusion. The problem related to occlusion is a problem in which part of the imaging target is hidden behind other parts depending on the imaging direction and cannot be acquired as point cloud data. One way to solve the occlusion problem is to measure objects from a plurality of different positions and integrate them by aligning them in a three-dimensional space.

  As this alignment method, the coordinates of the obtained point cloud are directly used to calculate the optimal position between the point clouds and match the positions, and the position calculation is indirectly performed using feature points and feature lines. A method for aligning the position by applying has been proposed. In the method of calculating the position by directly using the point cloud, the optimum alignment is performed by directly using the coordinates of the point cloud. For this reason, a standard for alignment such as a point or a surface is required, and there is a drawback that a marker must be normally arranged as a reference point. When using this marker, it is necessary to calculate the alignment by placing a condition so that the same markers measured from other directions match.

  On the other hand, the method of positioning based on feature points and feature quantities is a method of grouping point groups on the same plane with tolerances to divide the point group into regions, and geometrically calculating from the intersection of the planes. A feature line is extracted, and alignment is performed using a normal vector of the plane and the feature line. Here, when alignment is performed using feature amounts, a direction vector, a feature line, or the like is used as a reference point. Since the alignment based on the feature amount does not require a marker, it is effective for expressing a portion or an object where the marker cannot be arranged.

  By the way, the alignment using the above characteristic lines is effective for buildings and objects such as office buildings with clear linear characteristics, but it is complicated such as many irregularities such as temples and shrines. There is a problem that it is not effective for an object having a shape.

  That is, in this method, when the intended planar shape cannot be obtained, such as when the measurement error of the point cloud is large or when the sampling interval is wide compared to the complexity of the shape, the method of dividing the region changes, and as a result The resulting feature line will also change. In such a case, when feature lines are obtained from point groups acquired from a plurality of directions, the feature lines obtained from the respective point groups do not match in the three-dimensional space. For example, when the sampling interval in the measured direction is wide, fine uneven portions in the actual surface shape are not divided into regions, and feature lines to be extracted are reduced. If alignment is performed as it is, there may be fewer feature lines that match, and alignment may be impossible. On the other hand, if the sampling interval is narrowed, the amount of measurement data increases, causing problems such as exceeding the amount of data that can be handled by a computer, and problems such as inability to measure over a wide range.

  Accordingly, an object of the present invention is to provide a measurement method that enables alignment of images even when a sampling interval by a range sensor is wide and accurately measures an object having a complicated shape.

  In order to achieve the above object, according to a first main aspect of the present invention, a range sensor is used to measure an object from a plurality of different directions, and each of the distances in the depth direction to the measurement point of the object is measured. Generating point cloud data including information; processing each point cloud data; detecting a point cloud by connecting a point cloud in which a difference in distance between adjacent pixels in a depth direction is equal to or greater than a predetermined value; And outputting the geometric shape of the object by aligning each point cloud data using the detected edge.

  According to the present invention, a new technique for solving the above-described problems is provided by extracting feature amounts by a technique different from the conventional technique. That is, in one embodiment of the present invention, first, an edge-detected image is created from a point group that is a three-dimensional coordinate value having a correspondence relationship with each pixel of a distance image obtained by measurement, and a feature line is created. To extract. Next, the extracted feature line is segmented in a three-dimensional space. In final alignment, a pair in which line segments corresponding to each other obtained from different directions overlap in a three-dimensional space is selected, and an optimum geometric transformation is obtained by error evaluation.

  According to the present invention, feature lines that have been differentiated by a tolerant modeling method that takes into account an allowable error are used in order to enable alignment even for structures with rough surfaces or data with poor accuracy. Also, with this method, many feature lines can be extracted even when the sampling interval is wide, and alignment is also possible. Therefore, there are advantages such as being able to measure over a wide range and requiring less measurement data.

  According to a second main aspect of the present invention, there is provided a computer software program for measuring a three-dimensional shape of an object using a computer system, the storage medium being stored in the storage medium, Point cloud data acquisition unit that causes a computer system to generate point cloud data including distance information in a depth direction to a measurement point of an object based on information obtained by measuring the object from a plurality of different directions using a range sensor And a feature line that is stored in the storage medium, causes the computer system to process each point cloud data, and connects point clouds that have a distance difference in the depth direction between adjacent points to be detected as an edge. An acquisition unit, a feature line processing unit which is stored in the storage medium and causes the computer system to convert the edge into a feature line, and the storage medium The computer system further includes a geometric transformation output unit that outputs the geometric transformation of the object by aligning each point cloud data using the detected feature line. A computer software program is provided.

  According to such a configuration, a computer software program capable of executing the method according to the first aspect is provided.

  Note that other features and remarkable effects of the present invention will be apparent to those skilled in the art by referring to the following embodiments.

  Hereinafter, an embodiment of the present invention will be described by taking as an example a case where an existing building, topography, or the like is converted into three-dimensional data.

  FIG. 1 is a layout diagram of range sensors according to an embodiment of the present invention.

  In this embodiment, the imaging object 1 is a Japanese temple gate. And the 1st-4th range sensors 2a-2d are arrange | positioned so that this imaging target object 1 can be imaged from a different imaging direction. Each range sensor 2a-2d is connected to the object imaging system 4 of this embodiment. The range sensor in this embodiment includes a three-dimensional laser scanner, and is configured to be able to acquire distance data in the depth direction to the object surface by detecting the laser reflected by the imaging object. . The range sensor may be a currently known range sensor, or any range sensor that will be used in the future as long as it has a similar function.

  FIG. 2 is a schematic configuration diagram showing the object imaging system 4.

  In this system 4, a program storage unit 10 and a data storage unit 11 are connected to a bus 9 to which a CPU 5, a RAM 6, a range sensor interface 7, and an input / output device 8 such as a keyboard and a mouse are connected.

  The program storage unit includes a basic program 12 such as an OS, a point cloud data acquisition unit 13 that receives image signals from the range sensors 2a to 2d and outputs point cloud data, and a point cloud that constitutes an edge from each point cloud data. A feature line acquisition unit 14 that identifies and obtains a feature line of each point cloud data, a feature line processing unit 15 that converts the feature line into a line segment in a three-dimensional space, integrates and optimizes the feature line, and the feature An alignment processing unit 16 that aligns line groups obtained by different range sensors using lines, and a geometric conversion output unit 17 that outputs geometric conversion of each point group data based on the alignment; Have. The point cloud data, feature line data, and geometric transformation data processed by each of these components are stored in the data storage unit 11.

  The system 4 may actually be a general-purpose computer, and the program storage unit 10 and the data storage unit 11 may be an external storage device such as a hard disk. Each of the components 13 to 17 is stored in the program storage unit 10 in the form of computer software, and is appropriately called and executed on the RAM 6 by the CPU 5 to function as components of the present invention. ing.

  Hereinafter, the configuration of this system will be described together with its operation. FIG. 3 is a flowchart showing this processing step. In addition, S1-S6 in a figure respond | corresponds to the following steps S1-S6.

Acquisition of point cloud data

  The point cloud data acquisition unit 4 measures the imaging target 1 from different directions using the first and second range sensors 2a and 2b to obtain point cloud data (step S1). The point group has a correspondence between each pixel of the image obtained from the range sensors 2a and 2b and the three-dimensional coordinate value.

Feature line extraction

  Next, the feature line acquisition unit 14 extracts a feature line from the point cloud data.

  In this case, first, edge detection is performed from the point cloud data (step S2). This edge detection is performed by detecting a step whose distance in the depth direction is greater than or equal to a threshold value between adjacent points. In the present invention, this method is referred to as a depth edge method.

  That is, when performing alignment between images using feature lines, it is desirable to obtain as many feature lines as possible in the three-dimensional space from each measured direction. However, the feature detection method in the conventional image processing is a method based on pattern matching, and when a feature line is extracted based on this method, the feature line is characterized by fine irregularities on the surface of the shape. Since noise is generated, a necessary feature line may not be extracted well. This feature detection is performed by adjusting the threshold value. At this time, the more the characteristic lines are extracted by adjusting the threshold value, the more noise is generated. On the other hand, when noise generation is suppressed by reducing the threshold, the number of extracted feature lines is reduced, and there is a possibility that feature lines from different directions can be obtained. It will be lower.

On the other hand, in the present invention, information on the depth direction of the point cloud is obtained so that noise generation can be suppressed as much as possible, and as many as possible can be obtained by matching characteristic lines from different directions, particularly those useful for alignment. To find the feature line. That is, in the present invention, DepthEdge is defined. Depth Edge is a feature line corresponding to a portion where the difference between the depth values of adjacent pixels is large. Since the position (ix, jy) of each pixel corresponds to one point (x, y, z) of the point group, the feature line acquisition unit 14 sets the z value of the three-dimensional coordinate value corresponding to each pixel. Edges are detected by applying a threshold value to them. FIG. 4 is an image obtained in this way.

Line differentiation of feature lines

  Next, the feature line processing unit 15 differentiates the feature line (step 3).

  The feature lines extracted from the point group are expressed in units of pixels and depth values on the image. Therefore, this feature line is converted into a line segment in a three-dimensional space by the processing described below. Here, the number of connected pixels in four directions Cd (1 ≦ d ≦ 4) is given to each pixel as an element for determining a line segment. As shown in FIG. 5, the four directions are directions rotated by 45 degrees from the right direction on the two-dimensional image to the diagonally lower left. Cd for each direction d is the number of pixels that continue in that direction, and is the number of points that make up the line segment. The feature line processing unit 15 determines Cd of each pixel by applying the following algorithm.

(1) First, Cd is initialized. If a certain pixel P (ix, jy) is an edge portion, Cd = 1 is set for all directions d, and Cd = 0 is set otherwise.

(2) Search Cd from the pixel in the upper left corner of the image. For example, when d = 1, that is, when the search direction is the right side, the creation of the line segment L1 is started from a certain pixel P (ix, jy), and is set as the end point of the line segment. The number of connected components of the pixel P (ix + 1, jy) adjacent to the right side of P (ix, jy) is set to C′1. At this time, when the number of C′1 is 1 and the distance Dv between the two points in the three-dimensional space is shorter than the threshold Dis, C′1 = 1 + C1, and the pixel P (ix, jy of interest) C1 of 0) is set to 0.

(3) Next, the pixel to be referred to is P (ix + 1, jy), and similarly, the number of connected components of the right adjacent pixel and the distance between two points in the three-dimensional space are determined. If the number of connected components of the pixel to the right of the pixel of interest P (ix + n, jy) is 0 or the distance Dv between the two points is greater than the threshold Dis, the number of connected components is not added. The line segment creation is also finished. At this time, the points P (ix, jy) and P (ix + n, jy) are the end points of the line segment L1. The number of points constituting the line segment is the number of connected components of P (ix + n, jy), and in this case, (n + 1). Then, a new edge portion pixel is searched and the same processing is repeated.

Similarly, the feature line is differentiated by applying the linear differentiation algorithm in one direction described in (1), (2), and (3) to all four directions. The obtained line segment information is stored in the data storage unit 11 as feature line data.

Integration of line segments

  Next, the feature line processing unit 15 integrates the line segments converted from the feature lines (step S4).

That is, since the line segment created above extracts only components in a certain direction on the image, it is created as a discontinuous line segment in the case of the line segments A, B, C, and D shown in FIG. Will be. The feature line processing unit 15 derives an infinite straight line in a three-dimensional space on which the line segments are placed in order to integrate the line segments A, B, C, and D into one line segment, and arbitrarily two lines It is determined whether or not the minute satisfies the following two conditions at the same time, and if so, the two line segments are integrated.

(1) The two line segments L1 and L2 are such that the point Pi constituting one line segment is adjacent to one of the other points in the vicinity of the eight directions on the image, and between the two points in the three-dimensional space. Is less than the threshold Dis.

(2) As shown in FIG. 7, all of the line segments L1 and L2 that form the line segment L1 and L2 within the radius R from the infinite straight line L12 (t) on which the two line segments are combined in the three-dimensional space. If a point exists.

  When the above two conditions are satisfied, the feature line processing unit 15 derives an infinite straight line in the three-dimensional space by the following method. That is, first, an infinite straight line on which a point on the line segment is placed is L (t), and a distance from an arbitrary point Pi constituting the line segment to L (t) is d (Pi, L (t)). . Then, an infinite straight line is defined such that the distance between the infinite straight line L (t) and the point group Pi having an error is minimized.

For this definition, this embodiment introduces an evaluation formula (EL) that represents the sum of squares of the distance between the infinite line and the point group, as shown in formula (1).

  Here, n is the number of points constituting the line segment. The feature line processing unit 15 calculates an approximate solution by the least square method and determines an infinite straight line L (t) that minimizes EL. The initial value of L (t) uses the slope of a straight line connecting the midpoint of the line segment and the end points of the line segment. After unifying line segments, there are many line segments with very few points. In this case, it is difficult to determine whether the line segment is effective for alignment or invalid noise. Therefore, in the present invention, the line segments are arranged in descending order on the basis of the number of points constituting the line segment, and the top tens are used for alignment.

  Information on the line segments integrated in this way is stored in the data storage unit 11 as feature line data.

Line segment optimization

  Next, the feature line processing unit 15 performs line segment optimization processing (step S5).

  In the alignment method of the present invention, it is required that some line segments selected from the line segment group more accurately represent the shape features. Therefore, optimization of the line segment is realized by integrating the line segment used for alignment with the points of other adjacent edge portions.

  After optimizing the line segment, the number of vertices constituting the line segment increases. Specifically, optimization is performed according to the following procedure.

(1) A point P ′ having the closest distance between two points is searched for in the edge portions adjacent to all the points Pi constituting the line segment L on the Depth Edge image.

(2) From the infinite straight line L ′ (t) on which the new line segment L ′ including the point P ′ is located, if all the points constituting L ′ exist within the radius R, the point P ′ is defined as the line segment. Add to L.

The operations (1) and (2) are repeated for each line segment until there are no adjacent edge portions.

  In this way, the line segment is optimized, and the information is stored in the data storage unit 11 as feature line data.

Alignment

  Next, the alignment processing unit 16 derives the relative positional relationship between the point groups acquired from the plurality of sensor positions by using the line segment obtained above (step S6).

  This alignment is performed sequentially between two sets of line segments. That is, first, align the line segments obtained from any two directions, synthesize the two line segments, and then select one each from the remaining line segments and perform alignment and synthesis. Go.

  In the alignment of line segments from two directions, the three-dimensional geometric transformation in which one coordinate system is transferred to the other coordinate system is considered, and corresponding line segments between two line groups match as much as possible in the three-dimensional space. Find the optimal geometric transformation. Specifically, an error evaluation function for a line segment pair is defined, and a geometric transformation that minimizes the value is obtained by the least square method. At this time, in general, it can be said that alignment with higher accuracy is possible by selecting as many line segment pairs as possible.

  However, as described above, when the line segment group obtained from each direction includes an allowable error and a sufficient number of line segments cannot be obtained, a line that corresponds well between the line segment groups from the two directions can be obtained. It is often difficult to find the minute pair itself. Therefore, in the present invention, at least two pairs of line segments to be selected are selected. However, it chooses so that a pair may not become parallel. Further, for example, a line segment that appears differently depending on the measurement direction, such as a line segment corresponding to a silhouette in the central axis direction of a thick cylinder, is not used as an object for alignment. Next, an evaluation value based on the correspondence relationship in the three-dimensional space of the line segment other than the selected line segment pair is also calculated, and the geometric transformation is determined in consideration of the value. In the alignment algorithm of the present invention, the optimal geometric transformation of the selected line segment pair is determined by repeating the following four steps.

Step A. Initial position setting

Step B. Determination of rotational movement transformation

Step C. Determination of translation transformation

Step D. Evaluation after alignment

Steps A, B, and C have weighting parameters (described later). Based on the evaluation of step D, the optimum weighting parameters are obtained by repeating step A to step D. After all the geometric transformations are determined, the line segments are aligned from two directions using the geometric transformation (step E).

  Hereafter, each process is demonstrated in detail based on the flowchart of FIG.

STEP A: Initial position setting

  In the present embodiment, the Newton method is used as a method of solving an equation formulated by the least square method. However, in the Newton method, the solution may diverge depending on the initial value. Therefore, in this step, two sets are selected from the N pairs as preprocessing for determining the conversion of the rotational movement and the parallel movement, and the initial position is set so that they approach each other. Setting the initial position prevents solution divergence. The algorithm for setting the initial position determines right-handed coordinate axes in two measurement directions, and applies the affine transformation to match the two coordinate axes.

The coordinate axis determination method is defined as follows. That is, the infinite straight lines formed by the _two line segment pairs DL ₁ and DL ₂ used for alignment measured from one direction are defined as DL ₁ (t) and DL ₂ (t). At this time, the origin G of the coordinate axis is set to the intersection of two straight lines or the closest point. Then, the Y axis is set to the slope v ₁ of the straight line DL ₁ (t). Next, the direction of the outer product of v ₁ and the slope v ₂ of the straight line DL ₂ (t) is taken as the Z axis, and finally the direction of the outer product of v ₁ and the Z axis is taken as the X axis.

  The above operation is also performed on two line segments obtained from the other direction, and finally, affine transformation is performed so that the respective coordinate axes coincide.

  STEP B: Determination of rotational movement conversion

In this step B, conversion is determined so that each pair of line segments selected for alignment is parallel. In the coordinate system with the position of the sensor when measured from one point as the origin, the rotation angles of the respective axes are R _x , R _y , and R _z, and rotational movement is performed using the respective rotation matrices.

In a plurality of selected line segment pairs, let L ₁ (t) and L ₂ (t) be an infinite straight line on which a set of line segments travel, and let P be a point through which the infinite straight line L ₁ passes. Consider a case where the infinite straight lines L ₁ (t) and L ₂ (t) are parallel. As described above, the distance from the point P on the infinite line L ₁ (t) to L ₂ (t) is represented by d (P, L (t)). Therefore, the distance from the end points P ′ ₁ and P ′ ₂ to L ₂ (t) of the line segment after rotation, and the points P ′ ₃ and P ′ ₄ to L ₂ (t) far from the point P ′ ₀ Equations (2) and (3) are defined using the distance to From FIG. 9, the conditional expression in which two infinite straight lines are parallel is expressed by Expression (4).

Therefore, the error evaluation function ER is expressed by Equation (5).

However, the weighting parameter α satisfies the equation (6).

  A transformation that minimizes the expression (4) is obtained by the least square method. When determining the conversion of the next translation, it is obtained in the same manner using the error evaluation function.

  STEP C: Determination of translational transformation

In this step, translation is performed, and transformation is determined so as to minimize the distance between the infinite straight lines L ₁ (t) and L ₂ (t) on which the line segments L1 and L2 are respectively drawn. If the translation amounts are T _x , T _y , and T _z , the converted point P ′ is as follows.

As shown in FIG. 9, let L ′ ₁ (t) be an infinite straight line after the line segment L ₁ is translated. Similarly to the above, using the points P ′ ₁ , P ′ ₂ , P ′ ₃ , and P ′ ₄ on L ′ ₁ (t), the conditional expression that minimizes the distance between the pairs selected for alignment is an expression: (10)

Therefore, the error evaluation function ET is expressed by Equation (11).

However, the weighting parameter β satisfies Expression (12).

  STEP D: Evaluation after alignment

  Each alignment step uses weighting parameters and is set by evaluating the line segment after alignment.

In this evaluation step, first, for each pair that can be considered from the two line segment groups to be aligned, the line segment is evaluated for error one by one. At this time, the sum of the error evaluation values corresponding to the geometric transformation obtained in each step is defined as E defined by Expression (13).

  Based on the sum E of error evaluation values, the line segment after alignment is evaluated using the following two reference values.

(1) The total number of matching pairs (cnt) among all possible pairs. However, if the sum E ′ of error evaluations for a certain pair is equal to or less than the threshold EmAx, it is determined that the pair matches.

(2) A sum E obtained by performing error evaluation on the N line segment pairs used for alignment, which is defined by Expression (13) when determining the geometric transformation as described above.

  How to set weighting parameters

  In the above, the weighting parameters α and β are used. As a specific setting method of this parameter, first, the values of all parameters are set to 0.5 as initial values. Next, the optimal parameter value is determined by repeatedly applying the following algorithm in the order of the parameters α and β. In the following, the parameter α is described, but the same applies to β.

(1) Steps A to D are performed once with α = 0.5, and E0 and cnt0 that are initial values are derived.

(2) For α ₁ and α ₂ defined by Equation (14) and Equation (15), the number of matching pairs cnt ₁ and cnt ₂ and the sum E ₁ and E ₂ of error evaluation values are obtained. However, Δα is a step size, and k is the number of iterations.

(3) The initial value cnt ₀ is compared with cnt ₁ and cnt ₂ obtained in (2), the parameter having the largest number of pairs is selected, and the value is set as a new initial value.

(4) If the initial value is not updated or if the expression (16) is satisfied with respect to α ₁ and α ₂ , the process ends. Otherwise, k is incremented by one and the process proceeds to the process (2) in this section.

  However, σ is a minute value.

Geometric transformation output

  Next, the geometric transformation output unit 17 outputs the initial position setting, the rotational movement transformation, and the translational transformation obtained by the alignment processing unit as a geometric transformation. Thus, the point cloud data acquired by the range sensor can be integrated as one 3D image data. In this embodiment, the geometric transformation output unit 17 outputs three-dimensional image data obtained by converting the point cloud data obtained by the four range sensors by the geometric transformation.

  Examples thereof will be described below.

1 Measurement environment

  In this example, the south gate of the outer wall of Shiba Castle, Morioka City, Iwate Prefecture, which was restored in 1997, was measured. The restored south gate is a large gate next to Heijo and the Suzaku Gate in Nara City, Nara Prefecture, with a frontage of 15m and a height of 11.1m.

  The equipment used was LMS-Z210 manufactured by RIEGL. The main specifications of LMS-Z210 are as follows.

  Measurement distance range: 2m to 350m

  Line scanning range: ± 40 degrees

  Frame scanning range: 0 degrees to 333 degrees

2 Measurement data

  The range sensor can output three types of images. A distance image showing the distance from the range sensor to the building in RGB. Received light intensity image showing the intensity of reflected light at the point irradiated with laser in gray scale. Color image displaying RGB of laser irradiation point. The range sensor can output a three-dimensional coordinate value corresponding to each pixel of the image as a point cloud.

3 Experimental results

  FIGS. 10 and 11 show the results of applying the method of measuring the front side of the south gate from the front and the right side and performing line segmentation on the feature lines proposed in this embodiment. The black line in the figure is a line segment.

  The threshold values used for line differentiation were Dis = 2.0 and R = 1.0. As can be seen from the figure, distinct feature lines such as the outline of the shape and the edge of the wall are differentiated.

  FIGS. 12A and 12B show examples before and after applying optimization after line segmentation of feature lines. However, each line segment was inversely transformed onto a two-dimensional image. When attention is paid to the area surrounded by the dotted line in FIG. 12B, it can be seen that the length of the line segment increases.

  The results of alignment using the line segments shown in FIGS. 10 and 11 are shown in FIGS. However, since there were few pairs that were expected to match, two pairs were used for determining the geometric transformation and were manually selected. FIG. 13 shows line segments after alignment, and the black line segment and the white line segment are line segments when measured from the front and the right side, respectively. Also, the part where the line is thick is the selected line segment. FIG. 14 shows the point group after alignment. It took about 9 seconds to complete the alignment.

  In the present invention, a method for aligning point clouds obtained by measuring from a plurality of directions has been proposed. First, edge detection was performed using point clouds, and feature lines were extracted as edge portions. Then, the feature line is converted into a line segment in consideration of an allowable error by using a threshold value in a three-dimensional space. The line segment was optimized using the edge detection image. Finally, we selected a pair of line segments that are expected to be aligned and automatically obtained a geometric transformation that would move them to the optimal position. The alignment algorithm is divided into four stages: initial position setting, determination of rotational movement conversion, determination of parallel movement conversion, and evaluation after alignment.

  In addition, this invention is not limited to the said one Embodiment, A various deformation | transformation is possible in the range which does not change the summary of invention.

  For example, in the above-described embodiment, the line segment pair is automatically selected. However, the line segment to be a characteristic line to be aligned by the operator may be manually determined. In the above example, four range sensors are used. However, the present invention is not limited to this, and one range sensor may be arranged in four places in order to acquire an image. Also, the number is not limited to four, and may be two or more.

The schematic diagram which shows the apparatus structure which concerns on one Embodiment of this invention. Similarly, the schematic diagram which shows schematic structure. Similarly, the flowchart which shows a process. The figure which shows a Depth-Edge image similarly. Similarly, the figure which shows the direction of a connection pixel. Similarly, the schematic diagram for demonstrating the process of a discontinuous line segment. Similarly, the schematic diagram for demonstrating the process of integration of a line segment. Similarly, the flowchart which shows the alignment of point groups. Similarly, the schematic diagram for demonstrating alignment of two line segments. Similarly, the schematic diagram for demonstrating the line differentiation of the characteristic line. Similarly, the schematic diagram for demonstrating the line differentiation of the characteristic line. Similarly, the schematic diagram for demonstrating optimization of a line segment. Similarly, the schematic diagram which shows the line segment after alignment. Similarly, the schematic diagram which shows the point group after alignment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Imaging object 2a-2d ... 1st-4th range sensor 4 ... Object imaging system 5 ... CPU
6 ... RAM
7 ... IF for range sensor
DESCRIPTION OF SYMBOLS 8 ... Input / output device 9 ... Bus 10 ... Program storage part 11 ... Data storage part 12 ... Basic program 13 ... Point cloud data acquisition part 14 ... Feature line acquisition part 15 ... Feature line processing part 16 ... Registration processing part 17 ... Geometric Conversion output section

Claims (8)

a) using a range sensor to measure an object from a plurality of different directions, and generating point cloud data including information on the distance in the depth direction to the measurement point of the object for each;
b) processing each of the point group data, and connecting the point groups having a distance difference between adjacent points in the depth direction of a predetermined value or more to detect them as edges;
c) converting the edge into a feature line;
d) outputting the geometric transformation of the object by aligning the point cloud data using the detected feature line;
The object measuring method characterized by having. The method of claim 1, wherein
In the step (c), when a plurality of edges coincide with a certain straight line within a threshold value, the edges are integrated and converted into one feature line. . The method of claim 1, wherein
The step (d)
Searching for feature lines at the same position for each set of feature lines acquired from two directions;
Calculating a geometric transformation for alignment from the pair of feature lines obtained above;
A method characterized by comprising: The method of claim 1, wherein
The method (d) includes a step of generating three-dimensional shape data by applying the geometric transformation to data of an object measured from the plurality of different directions. A computer software program for measuring a three-dimensional shape of an object using a computer system,
A storage medium;
Point cloud data stored in this storage medium and including information on the distance in the depth direction to the measurement point of the object based on information obtained by measuring the object from a plurality of directions using a range sensor in the computer system A point cloud data acquisition unit for generating
Feature line acquisition unit that is stored in this storage medium, causes the computer system to process each point cloud data, and connects point clouds that have a difference in distance in the depth direction between adjacent points to be detected as an edge. When,
A feature line processing unit stored in the storage medium and causing the computer system to convert the edge into a feature line;
A geometric transformation output unit that is stored in the storage medium and outputs the geometric transformation of the object by aligning the point cloud data with the detected feature line in the computer system;
A computer software program characterized by comprising: The computer software program according to claim 5, wherein
The feature line acquisition unit, when a plurality of edges match a certain straight line within a threshold value, integrates the edges and converts them into one feature line. Software program. The computer software program according to claim 5, wherein
The geometric transformation output unit includes:
For each set of feature lines acquired from two directions, a search is made as to whether or not the feature lines are at the same position, and a geometric transformation for alignment is calculated from the obtained pair of feature lines. A computer software program characterized by being. The computer software program according to claim 5, wherein
The geometric transformation output unit includes a step of generating three-dimensional shape data by applying the geometric transformation to data of an object measured from a plurality of different directions. .

Images