JP2005315846A - Three-dimensional information extraction device and method, computer-readable recording medium recording program, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To automatically extract three-dimensional information related to the height of a target and elements (for example, structure in ships) for constituting the target, regarding a three-dimensional information extraction device based on images related to the target of ships, or the like. <P>SOLUTION: For target images read by an image reading section 1, a main component analysis section 3 calculates the main component information, such as the main component direction, center of gravity, starting point, and ending point. A resampling section 4 normalizes target images, based on the main component information. A three-dimensional information calculating section 5 calculates the three-dimensional information related to the height of a target, the elements (for example, structure in ships) for constituting the target, based on the normalized target images. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a three-dimensional information extraction apparatus based on an image relating to a target such as a ship, and automatically extracts three-dimensional information related to the height of the target and elements constituting the target (for example, a ship structure). Regarding technology.

  FIG. 26 is a diagram illustrating the concept of the height component of the target included in the captured image. Since a perspective image is captured with a certain elevation angle (θ), the image includes a target height component. For example, the actual height of h1 appears as a height component h1 ′ on the image.

Conventionally, the “height” specification is calculated from an obliquely photographed image by the following procedures (1) and (2).
(1) The length (number of pixels) of the target height component is visually determined based on the density of the image. An example will be described below.

  FIG. 27 is a diagram showing a height component of a building. The light color portion of the building portion is determined as the height component, and the length (number of pixels) of the height component is counted.

  FIG. 28 is a diagram illustrating a height component of a ship. In the case of a ship, the height of the structure is the calculation target. Therefore, the length (number of pixels) of the height component of the structure is counted.

  (2) The actual “height” specifications are calculated by applying the length (number of pixels) obtained in (1) to the geometric relationship between the satellite and the target. FIG. 29 is a diagram illustrating the concept of calculating the height component. By dividing the length h1 ′ by Sinθ, the actual height h1 is obtained (more precisely, a predetermined magnification is multiplied).

  However, the above-described method can be applied only to an object having a simple shape such as a building, and is difficult to apply to a target having a complicated shape such as a ship. For example, in FIG. 28, it is difficult to visually determine the height of a ship, which is a target having a complicated shape. In addition, the height of the ship said here is the distance from the ship bottom to the top of the highest structure.

Another problem is that quantitative evaluation is difficult because it depends on human subjectivity. For example, when the number of pixels for the targets in FIGS. 26 to 28 is measured, variation by the measurer occurs. In particular, when the direction of the height component has an inclination with respect to the normal direction of the image as shown in FIG. 27, further quantitative evaluation becomes difficult.
JP 2003-187219 A JP 2000-020692 A JP 2001-200142 A

  The present invention was made in order to eliminate the above-described drawbacks of the prior art, and the object of the present invention is to automatically calculate target height information by a computer, thereby reducing the labor of measurement, It is to enable quantitative evaluation.

Furthermore, the target height and detailed three-dimensional information (length, width, height of the target structure, the start position of the structure, the end position of the structure) are automatically calculated by a computer. It is to be.

The three-dimensional information extraction apparatus according to the present invention is
(1) Principal component analysis unit for calculating at least the principal component direction of the target image (2) Resampling unit for normalizing the target image by the principal component direction (3) Normalized A height calculation unit that calculates the target height based on the target image.

  The height calculation unit determines a boundary between a horizontal plane and a vertical plane based on a difference in luminance, and calculates the height of the boundary as the height of the target.

  The target is a ship.

The three-dimensional information extraction method according to the present invention includes:
(1) Principal component analysis step for calculating at least the principal component direction of the target image (2) Resampling step for normalizing the target image by the principal component direction (3) Normalized A height calculating step for calculating a target height based on the target image.

A computer-readable recording medium on which a program according to the present invention is recorded,
It is a computer-readable recording medium that records a program for causing a computer as a three-dimensional information extraction apparatus to execute the following processing. (1) A principal component for calculating at least a principal component direction of a target image Analysis processing (2) Resampling processing for normalizing the target image based on the principal component direction (3) Height calculation processing for calculating the target height based on the normalized target image.

The program according to the present invention is:
It is a program for causing a computer as a three-dimensional information extraction device to execute the following procedure: (1) Principal component analysis procedure for calculating at least a principal component direction of a target image (2) Depending on the principal component direction Resampling procedure for normalizing the target image (3) A height calculation procedure for calculating the target height based on the normalized target image.

  In the present invention, target height information can be automatically calculated. Further, it is possible to automatically calculate the three-dimensional information (for example, the length, width, height, structure start point position, structure end point position) of the target element having a complicated shape.

Embodiment 1 FIG.
Hereinafter, the present invention will be described based on embodiments shown in the drawings.

  FIG. 1 is a diagram illustrating a software configuration of the three-dimensional information extraction apparatus. The three-dimensional information extraction apparatus includes an image reading unit 1, an image binarizing unit 2, a principal component analyzing unit 3, a resampling unit 4, a three-dimensional information calculating unit 5, a calculation result writing unit 6, and a three-dimensional information display unit 7. A read image storage unit 11, a resampling image storage unit 12, and a three-dimensional information storage unit 13.

  Next, processing will be described. FIG. 2 is a diagram showing an overall processing flow. First, the image reading unit 1 performs image reading processing (S10). For example, a ship image obtained by perspective shooting as shown in FIG. 3 is read. A ship is an example of a goal.

  Then, the image binarization unit 2 performs image binarization processing (S20). By this processing, the target portion and other portions are determined, and the target portion is converted to 1 and the other portions are converted to 0.

  Next, the principal component analysis unit 3 performs principal component analysis processing (S30). By this processing, the target principal component direction (target direction), the center of gravity, the total length, the full width, the area, and the like are calculated. In particular, the direction of the ship is used. The subsequent processes are always performed in the normalized direction. The principal component analysis process (S30) will be described later with reference to FIGS.

  Then, the resampling unit 4 performs resampling processing (S40). In this processing, the principal component direction, the center of gravity, and the resolution (length per pixel) of each target are normalized and resampled by rotating, enlarging, and reducing the image. Thereby, the direction of the target is aligned in a predetermined direction, and the center of gravity position and the reduction rate are also aligned. The resampling image is re-stored in the temporary image area (size Ship_x × Ship_y).

  The three-dimensional information calculation unit 5 performs a three-dimensional information calculation process (S50). With this processing, target three-dimensional information is calculated from the image. FIG. 4 is a diagram illustrating an example of three-dimensional information. This process will be described later in detail with reference to FIG.

  The calculation result writing unit 6 performs a calculation result writing process (S60). By this process, the target three-dimensional information calculated in the three-dimensional information calculation process (S50) is written to the hard disk or the temporary area.

  Further, the three-dimensional information display unit 7 performs a three-dimensional information display process (S70). By this process, a target stereoscopic image can be created using the target three-dimensional information calculated in the three-dimensional information calculation process (S50).

Next, the three-dimensional information calculation process (S50) will be described. FIG. 5 is a diagram showing a three-dimensional information calculation processing flow. FIG. 6 is a diagram illustrating a software configuration of the three-dimensional information calculation unit. It has a scissors calculating unit 51, a structure calculating unit 52, and a structure part calculating unit 54. They also have the elements shown in the figure.

  First, the height of a side wall part is calculated | required from an image by the side wall calculation process (S501) by the side wall calculation part 51. FIG. In addition, only the part necessary for the subsequent processing is extracted from the image using the obtained information. This process will be described later with reference to FIG.

  Next, in the structure calculation process (S502) by the structure calculation unit 52, three-dimensional information of the structure (an example of an element constituting the target) is calculated. In addition, only the part necessary for the subsequent processing is extracted from the image using the obtained information. This process will be described later with reference to FIG.

  A loop process is performed for the number of structures (S503). In the structure part calculation process (S504) on the structure, the same process as the process on the structure in the structure calculation process (S502) is performed on the structure part on the structure. Thereby, the three-dimensional information of the structural part on the structure is calculated. The structure part calculation processing (S504) on the structure by the structure part calculation unit 54 will be described later with reference to FIG.

  The scissor calculation process (S501) will be described in detail. FIG. 7 is a diagram illustrating a side-calculation process flow. In the height calculation process (S5011), the height of the scissors portion is calculated. In this process, the target height is calculated by obtaining a statistical process (average value) for each line in the total length direction of the normalized ship. Further, in the process corresponding part extraction process (S5012), only the part necessary for the subsequent process is extracted using the result of the height calculation process (S5011). The process corresponding part extraction process (S5012) will be described later with reference to FIG.

  The height calculation process (S5011) will be described in detail. First, the concept of processing will be described. FIG. 8 is a diagram illustrating an example of the luminance distribution of the ship image. As shown in the figure, the brightness on the side surface of the ship structure projected in the perspective image is higher (lower) than the upper surface. Therefore, in the present invention, the height of the side surface is determined from the difference in luminance between the deck surface (example of horizontal surface) and the side surface (example of vertical surface), and the height of the target (the structure of the target) is determined. calculate.

  FIG. 9 is a diagram illustrating an example of a perspective image and luminance average in the X-axis direction. In the perspective image as shown in FIG. 1A, when a statistical value (average value in this case) in the X-axis (horizontal axis) direction is calculated for each line of the Y-axis (vertical axis), a profile of the statistical value is drawn. The brightness average value is as shown in FIG. Since the length of the arrow in FIG. (B) is proportional to the height of the scissors (structure), the height can be calculated from a single image by measuring the length of the arrow portion.

  Next, specific processing will be described. FIG. 10 is a diagram illustrating a height calculation processing flow. Loop processing (S50111) is performed for the number of pixels in the Y-axis direction (I = 1 to the number of Y-axis pixels). In the luminance average value calculation process of S50112, the average luminance value in the X-axis direction is calculated. This process will be described later with reference to FIG.

  In the bottom surface luminance average determination process of S50113, it is determined whether I = 1 or not. If I = 1, the bottom luminance average setting process is performed in S50114. In this process, the luminance average of I = 1 is set as the luminance average value of the lower surface.

  In the height calculation start determination process in S50115, I> = 2, but it is determined whether or not, and in the luminance average value determination process (No. 1) in S50116, the I-th luminance average value is further converted to the bottom luminance average value. It is determined whether or not + α (predetermined threshold) or more. When I> = 2 and the I-th luminance average value is equal to or greater than the bottom surface luminance average value + α (predetermined threshold value), the luminance average value determination process (No. 2) of S50117 is performed. Otherwise, The processing shifts to the next Y-axis pixel processing loop.

  In the luminance average value determination process (No. 2) in S50117, it is determined whether the I-th luminance average value is equal to or greater than the (I-1) -th luminance average value. When the luminance average value of the Y-axis pixel (I) is larger than the luminance average value of the previous Y-axis pixel (I-1), the side pixel counter is incremented by the side pixel counter increment processing in S50118. .

  When all the Y-axis pixels have been processed, the height is calculated by multiplying the number of side pixels by a scale factor (length of one pixel) in the height calculation process of S50119.

The luminance average value calculation processing in S50112 will be described in detail. FIG. 11 is a diagram illustrating an X-axis direction luminance average value calculation processing flow.
As shown in S501212, loop processing is performed for the number of pixels in the X-axis direction. In the luminance sum calculation processing of S501212, the X = Jth and Y = Ith luminance (I, J) are sequentially added to the luminance sum SUM (I). Then, after the end of the loop processing, the luminance average value Mean (I) in the X-axis direction is calculated by dividing the luminance sum SUM (I) by the number of X-axis pixels in the luminance average value calculation processing in S501123.

  By this processing, for example, the “height” of the building and the “side height” of the ship as shown in FIG. 12 can be automatically calculated.

  Here, the process relevant part extraction process (S5012) will be described in detail. In the processing corresponding part extraction processing (full length direction) in S5012, only the portion necessary for the subsequent processing is extracted from the sea, the side fence, the deck surface (the structure side surface, the structure upper surface) divided in the above-described processing. . FIG. 13 is a diagram illustrating an image of the processing relevant part extraction processing.

  Next, the structure calculation process (S502) will be described in detail. FIG. 14 is a diagram illustrating a structure calculation processing flow. In the structure position calculation process of S5021, the structure feature length, width, structure start point position, and structure end point position are extracted. FIG. 15 is a diagram illustrating an image of the structure position calculation process.

  For example, the length, width, start point position, and end point position are extracted from the pixels in the axial direction.

  And the height of each structure is calculated by the structure height calculation process of S5022. This process is the same as the process of S5011. In the structure extraction process of S5023, only the structure necessary for the subsequent processes is extracted using the result of S5022. FIG. 16 is a diagram illustrating an image of the structure extraction process.

  Next, the structure part calculation process (S504) will be described in detail. This process is the same as the process performed on the structure in the structure calculation process (S502) shown in FIG. 14 except that the structure portion (example of lower component) on the structure (example of higher component). ) Therefore, a process (structure part position calculation process) corresponding to the structure position calculation process (S5021) and a process (structure part height calculation process) corresponding to the structure height calculation process (S5022) are performed. In these processes, information in the entire width direction of the structure part (height of the structure part + width of the structure part) is calculated. FIG. 17 is a diagram illustrating an image of the structure part position calculation process and the structure part height calculation process.

  A process (structure partial extraction calculation process) corresponding to the structure extraction calculation process (S5023) is performed. Using the information on the structure part extracted in the above process, a part necessary for the subsequent process is extracted. FIG. 18 is a diagram illustrating an image of the structure portion extraction process.

  By repeating the height calculation process in the same manner as described above with the structure portion thus extracted as a processing target, more detailed height information and width information of the structure portion can be separated. Further, by repeating the position calculation process for the processing target in the same manner, a detailed structure portion can be extracted. By repeating these processes, it is possible to know the target height having a complicated shape and detailed three-dimensional information of the ship.

  It is also possible to create a target stereoscopic image based on the three-dimensional information obtained as described above. FIG. 19 is a diagram illustrating an example of a stereoscopic image.

Finally, the principal component analysis process (S30) by the principal component analysis unit 3 will be described in detail.
This will be described in detail with reference to FIG. X1, x2,... Xn, y1, y2,... Yn, μx, μy, Δx1, Δx2,..., Δxn, Δy1, Δy2,. q2, L, and B are variables stored in a rewritable storage area, and s is a constant stored in the storage area.

  First, in the center-of-gravity calculation process (Step 2001), the average value μx of the X coordinate of the component pixel group and the average value μy of the Y coordinate are calculated, and the coordinates (μx, μy) of the center of gravity of the region composed of the component pixel group are obtained.

When the number of pixels in the component pixel group is n and the coordinates of each pixel are (x1, y1), (x2, y2),..., (Xn, yn), μx and μy are the following formulas (1) and (2) Calculated by
μx = (x1 + x2 +... + xn) / n (1)
μy = (y1 + y2 +... + yn) / n (2)

  The procedure of the center-of-gravity calculation process (Step 2001) is shown in FIG. In the figure, sigma_x, sigma_y, and n are variables stored in the temporary storage area. For each coordinate in the minimum rectangle (S2103), when included in the target ship (S2104), the coordinate values are summed (S2105), and as a result, sigma_x and sigma_y are obtained and divided by the summed number of coordinates n ( S2106), the x-coordinate μx and the y-coordinate μy of the center of gravity are obtained (S2107).

Next, a correlation matrix is obtained by calculating a correlation matrix (Step 2002 to Step 2004). Therefore, first, in the deviation calculation process (Step 2002), deviations (Δx1, Δy1) to (Δxn, Δyn) from the center of gravity of each pixel of the component pixel group are calculated. Each deviation corresponds to a coordinate value when the center of gravity (μx, μy) is regarded as the coordinate origin. Δxi and Δyi (1 ≦ i ≦ n) are calculated by the following formulas (3) and (4).
Δxi = xi−μx (3)
Δyi = yi−μy (4)

Next, in the deviation vectorization process (Step 2003), a vector Δx is obtained from the x component of the deviation, and a vector Δy is obtained from the y component of the deviation. Δx and Δy are defined by the following formulas (5) and (6).
Δx = (Δx1, Δx2,..., Δxn) (5)
Δy = (Δy1, Δy2,..., Δyn) (6)

Next, in a correlation matrix calculation process (Step 2004), a 2-by-2 correlation matrix C of Δx and Δy is calculated. Each component of C is calculated as follows. (・ Is a vector dot product operator.)
C11 = Δx · Δx (7)
C12 = Δx · Δy (8)
C21 = Δy · Δx (9)
C22 = Δy · Δy (10)

  FIG. 22 is a diagram showing details of the correlation matrix calculation process. Note that x1, x2,... Xn, y1, y2,... Yn, μx, μy, C11, C12, C21, and C22 are variables stored in a rewritable storage area.

  First, C11, C12, and C22 are initialized to 0 (S2201). For each coordinate in the minimum rectangle (S2202), if included in the target ship (S2203), the difference between x and μx is obtained and the difference is temporarily Store, calculate the square of the difference, temporarily store the square value, and add the square value to C11. Further, the difference between y and μy is obtained, the difference is temporarily stored, the difference between x and μx, the difference between y and μy is added, the product is temporarily stored, and the product is added to C12. Further, the square of the difference between y and μy is obtained, the square value is temporarily stored, and the square value is added to C22 (S2204).

Next, calculation of eigenvalues and eigenvectors (S2005, S2006) will be described. First, in the full length direction calculation process (Step 2005), the maximum eigenvalue λ1 of the correlation matrix C and the corresponding eigenvector (p1, q1) are calculated. The maximum eigenvalue λ1 can be calculated by the following equation. (Note that sqrt (X) represents the square root of X.)
λ1 =
(C11 + C22 + sqrt ((C11−C22) 2 + 4 · C12 · C21)) / 2 (11)

Using λ1 obtained by the above equation (11), p1 and q1 can be calculated by the following equation.
p1 = C12 / sqrt (C12 · C21 + (C11−λ1) 2)
(12)
q1 = (λ1-C11) / sqrt (C12 · C21 + (C11-λ1) 2)
(13)

  The direction of (p1, q1) obtained here is the full length direction of the region composed of the candidate pixel component pixel groups. This is because the variance of the position on the straight line of each component pixel projected onto a straight line passing through the center of gravity (μx, μy) (one center is the positive direction and the other is the negative direction) is the direction of the straight line This is because it becomes maximum when the values coincide with the direction of (p1, q1). In this way, the traveling direction of the ship candidate region can be obtained by q1 / p1.

Similarly, in the full width direction calculation process (Step 2006), the minimum eigenvalue λ2 of the correlation matrix C and the corresponding eigenvector (p2, q2) are calculated. The minimum eigenvalue λ2 can be calculated by the following equation.
λ2 =
(C11 + C22-sqrt ((C11-C22) 2 + 4 · C12 · C21)) / 2 (14)

P2 and q2 can be calculated by the following equation using λ2 obtained by equation (14).
p2 = C12 / sqrt (C12 · C21 + (C11−λ2) 2)
(15)
q2 = (λ2−C11) / sqrt (C12 · C21 + (C11−λ2) 2)
(16)

  The direction of (p2, q2) is the full width direction of the ship candidate region. This is because the dispersion of the position of each component pixel projected onto the straight line passing through the center of gravity (μx, μy) is minimized when the direction of the straight line matches the direction of (p2, q2). Further, the full width direction calculated in this way has a property that it is always orthogonal to the full length direction obtained above.

  FIG. 23 is a diagram illustrating a calculation process of eigenvalues and eigenvectors. Note that C11, C12, C22, λ1, λ2, divisor11, divisor12, divisor21, divisor22, p1, q1, p2, and q2 are variables stored in a rewritable storage area.

  First, λ1 and λ2 are obtained (S2301). The difference between C11 and C22 is obtained, the difference is temporarily stored, the square of the difference is obtained, the square value of the difference between C11 and C22 is temporarily stored, the square of C12 is further obtained, and the square value of C12 is temporarily stored. To do. Add the square value of the difference between C11 and C22 and the square value of C12, temporarily store the sum, find the square root of the sum, temporarily store the square root, add C11, C22 and the square root, and the result Is stored as λ1. Also, C11 and C22 are added, the above-mentioned square root is subtracted from the sum, and ½ of the result is stored as λ2.

  Next, the divider 11, the divider 12, the divider 21, and the divider 22 are obtained (S2302). Subtract λ1 from C11, temporarily store the difference, obtain the square of the difference between C11 and λ1, and temporarily store the square value. The square value is added to the square value of C12, the sum is temporarily stored, the square root of the sum is obtained, and the square root is stored as the divider 11.

  Subtract λ1 from C22, temporarily store the difference, obtain the square of the difference between C22 and λ1, and temporarily store the square value. The square value is added to the square value of C12, the sum is temporarily stored, the square root of the sum is obtained, and the square root is stored as divisor 12.

  Subtract λ2 from C11, temporarily store the difference, obtain the square of the difference between C11 and λ2, and temporarily store the square value. The square value is added to the square value of C12, the sum is temporarily stored, the square root of the sum is obtained, and the square root is stored as the divider 21.

  Subtract λ2 from C22, temporarily store the difference, obtain the square of the difference between C22 and λ2, and temporarily store the square value. The square value is added to the square value of C12, the sum is temporarily stored, the square root of the sum is obtained, and the square root is stored as the divider 22.

  When both the divider 11 and the divider 12 are 0, or when both the divider 21 and the divider 22 are 0 (S2303), it is determined that the direction of the eigenvector is indefinite (S2304). In other cases, p1 and q1 are obtained, and p2 and q2 are further obtained.

If divider11> divider12 (S2305), C12 is divided by divider11, the result is stored as p1, subtracts C11 from λ1, temporarily stores the difference, the difference is divided by divider11, and the result is q1. Store (S2306). If divisor11 <divider12 (S2305), subtract C22 from λ1, temporarily store the difference, divide by divisor12, store the result as p1, divide C12 by divisor12, and set the result as q1 Store (S2307).

  If divisor21> divider22 (S2308), C12 is divided by divisor21, the result is stored as p2, C11 is subtracted from λ2, the difference is temporarily stored, the difference is divided by divisor21, and the result is q2. Store (S2309). If divisor21 <divider22 (S2308), subtract C22 from λ2, temporarily store the difference, divide the difference by divider22, store the result as p2, divide C12 by divisor22, and set the result as q2 Store (S2310).

Next, length / width calculation processing (S2007, S2008) will be described. First, in the total length calculation process (Step 2007), the total length L is calculated. The total length L is calculated according to the following formula.
L = (max (Δxi · p1 + Δyi · q1) +
max (−Δxi · p1−Δyi · q1)) · s (17)

  However, i takes 1 to n (n is the number of pixels in the component pixel group). Moreover, s is the pixel spacing of the satellite image data, and is the length of one side of one pixel of the satellite image data. The total length L is the maximum distance between two pixels of the component pixel group projected on a straight line passing through the center of gravity (μx, μy) and parallel to the total length direction.

Next, the full width B is calculated in the full width calculation process (Step 2008). The total width B is calculated by the following formula.
B = (max (Δxi · p2 + Δyi · q2) +
max (−Δxi · p2−Δyi · q2)) · s (18)
However, i takes 1 to n (n is the number of pixels in the component pixel group). The full width B is the maximum distance between two pixels of the ship pixel group projected on a straight line passing through the center of gravity (μx, μy) and parallel to the full width direction.

  FIG. 24 is a diagram illustrating a length / width calculation process. Note that min_z1, max_z1, min_z2, max_z2, z1, z2, x, y, μx, μy, p1, q1, p2, q2, length, and width are variables stored in a rewritable storage area.

  First, min_z1, max_z1, min_z2, and max_z2 are initialized to 0 (S2401). For each coordinate in the minimum rectangle (S2402), if included in the target ship (S2403), z1 and z2 are calculated (S2404). When z1 is smaller than min_z1 (S2405), z1 is stored in min_z1 (S1406), when z1 is larger than max_z1 (S2407), z1 is stored in max_z1 (S2408), and when z2 is smaller than min_z2 ( S2409), z2 is stored in min_z2 (S2410). If z2 is larger than max_z2 (S2411), z2 is stored in max_z2 (S2412).

z1 and z2 are calculated by the following processing. Subtract μx from x to temporarily store the difference, subtract y from μy to temporarily store the difference. Then, p1 is multiplied by the difference between x and μx to temporarily store the product of p1 · (x−μx), and q1 is multiplied by the difference between y and μy to temporarily store the product of q1 · (y−μy). Then, the product of p1 · (x−μx) and the product of q1 · (y−μy) are added, and the sum is stored as z1. Also, p2 is multiplied by the difference between x and μx to temporarily store the product of p2 · (x−μx), and q2 is multiplied by the difference between y and μy to temporarily store the product of q2 · (y−μy). Then, the product of p2 · (x−μx) and the product of q2 · (y−μy) are added, and the sum is stored as z2.

  Finally, min_z1 is subtracted from max_z1 and the difference is stored as length, and min_z2 is subtracted from max_z2 and the difference is stored as width (S2414).

  The three-dimensional information extraction apparatus is a computer, and each element can execute processing by a program. Further, the program can be stored in a storage medium so that the computer can read the program from the storage medium.

  FIG. 25 is a diagram illustrating a hardware configuration example of the three-dimensional information extraction apparatus. In this example, an arithmetic device 2501, an image input device 2502, a memory 2503, a hard disk 2504, and a display device 2505 are connected to the bus. The program is stored in, for example, the hard disk 2504, and is sequentially loaded into the arithmetic device 2501 and processed while being loaded in the memory 2503.

It is a figure which shows the software structure of a three-dimensional information extraction apparatus. It is a figure which shows the whole processing flow. It is a figure which shows the example of a ship image. It is a figure which shows the example of three-dimensional information. It is a figure which shows a three-dimensional information calculation process flow. It is a figure which shows the software structure of a three-dimensional information calculation part. It is a figure which shows a scissors calculation process flow. It is a figure which shows the example of the luminance distribution of a ship image. It is a figure which shows the example of the brightness | luminance average of a X-axis direction. It is a figure which shows a height calculation process flow. It is a figure which shows a X-axis direction luminance average value calculation processing flow. It is a figure which shows the height of a building, and the height of the side of a ship. It is a figure which shows the image of a process applicable part extraction process. It is a figure which shows a structure calculation process flow. It is a figure which shows the image of a structure position calculation process. It is a figure which shows the image of a structure extraction process. It is a figure which shows the image of a structure part position calculation process and a structure part height calculation process. It is a figure which shows the image of a structure part extraction process. It is a figure which shows the example of a stereo image. It is a figure which shows the flow of a principal component analysis process. It is a figure which shows the processing flow of calculation of a gravity center. It is a figure which shows the processing flow of calculation of a correlation matrix. It is a figure which shows the processing flow of calculation of an eigenvalue and an eigenvector. It is a figure which shows the processing flow of calculation of length and width. It is a figure which shows the hardware constitutions of a three-dimensional information extraction apparatus. It is a figure which shows the concept of the height component of the target contained in a picked-up image. It is a figure which shows the height component of a building. It is a figure which shows the height component of a ship. It is a figure which shows the concept of height component calculation.

Explanation of symbols

1 image reading unit, 2 image binarization unit, 3 principal component analysis unit, 4 resampling unit, 5
3D information calculation unit, 6 calculation result writing unit, 7 3D information display unit, 11 read image storage unit, 12 resampling image storage unit, 13 and 3D information storage unit, 51 lateral flaw calculation unit, 52 structure Calculation part, 54 Structure part calculation part.

Claims (6)

A three-dimensional information extraction device having the following elements: (1) a principal component analysis unit that calculates at least a principal component direction of a target image; (2) a resampling unit that normalizes the target image by the principal component direction (3 ) A height calculator that calculates the target height based on the normalized target image. The three-dimensional information extraction apparatus according to claim 1, wherein the height calculation unit determines a boundary between the horizontal plane and the vertical plane based on a difference in luminance, and calculates the height of the boundary as the height of the target.   The three-dimensional information extraction apparatus according to claim 1, wherein the target is a ship. A three-dimensional information extraction method having the following elements: (1) a principal component analysis step for calculating at least a principal component direction of a target image (2) a resampling step for normalizing the target image with the principal component direction (3 ) A height calculating step for calculating the target height based on the normalized target image. A computer-readable recording medium that records a program for causing a computer serving as a three-dimensional information extraction apparatus to execute the following processing. (1) Principal component analysis processing for calculating at least a principal component direction of a target image. Resampling process for normalizing the target image according to direction (3) Height calculation process for calculating the target height based on the normalized target image. A program for causing a computer to be a three-dimensional information extraction apparatus to execute the following procedure: (1) a principal component analysis procedure for calculating at least a principal component direction of a target image; Sampling procedure (3) A height calculation procedure for calculating the target height based on the normalized target image.