JP2005173995A - Device and method for calculating depth, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To efficiently calculate the depth of a significant object which does not accompany increase in the calculation quantity or memory. <P>SOLUTION: This depth calculating device is configured of an input/output processing means 1 for storing processing results in a depth database 6, an image feature point calculation processing means 2 for calculating image feature points, a parallax quantity candidate calculating means 3 for calculating parallax quantity candidates, by using the image feature points calculated by the image feature point calculation processing means 2 and a parallax quantity calculation processing means 4 for deciding parallax quantity from the calculated parallax quantity candidates. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a technique for calculating depth by performing stereo viewing using an image whose horizontal axis is parallel projection.

  The following two methods are known for calculating the depth from a stereo image.

  Method 1 is a block in which a block-shaped partial image centered on a pixel in one image is cut out, an area having the highest similarity is obtained on the other image, and the depth is calculated using that area as a corresponding point. There are methods based on matching. In this method, all regions on the image or features such as shading are extracted, corresponding points are assigned to the feature points, and the depth is calculated for each pixel or region (see Non-Patent Document 1, for example).

Also, as method 2, a three-dimensional shape model such as a plane or a curved surface is prepared, and the degree of similarity obtained by the matching process is voted on a space defined by the shape model parameters, and the parameters are determined using the principle of majority vote. Thus, there is a method for restoring the shape (see, for example, Patent Document 1).
Japanese Patent Laid-Open No. 2003-256811 Takeshi Yasui, “Introduction to Image Processing in C Language”, Shosodo, November 20, 2000

  The above-described method 1 has a drawback that it takes a long time to perform the matching process for each block-shaped partial image. Further, since the depth is obtained for each point, priority cannot be given to the depth to be restored, and a depth that is less necessary is calculated. For this reason, the number of surfaces constituting the object becomes very large at the time of restoration, and it cannot be said that sufficient modeling is performed as it is. Further, the calculated depth has a lot of noise. For example, even when it is desired to restore the planar model, it becomes a complicated shape with irregularities, and it may be necessary to store extra data.

  Since the method 2 described above uses voting, it can be calculated from the depth of a highly necessary object. However, there is a problem that the calculation amount is large and it is difficult to extract the region of the object. In addition, since the number of dimensions of the space for voting increases when searching for 3D shapes, Hough transformation based on the principle of voting and majority voting for shape restoration (Takashi Matsuyama, Yoshinori Kuno, Satoshi Imiya “Computer Vision”, New Technology Communications, 1999 July 25th), the amount of calculation and memory increase in proportion to the power of the number of dimensions of the space for voting. In addition, in the voting space, voting trajectories from different figures may interfere and the shape restoration accuracy may decrease, but there is a problem that it is difficult to deal with the more dimensions.

  This invention is made | formed in view of this situation, The objective is to provide the depth calculation technique which solved the said subject.

  In order to solve the above problem, the present invention provides an image feature point calculation process for calculating an image feature point, a parallax amount candidate calculation process for calculating a parallax amount candidate by voting only on the image feature point, and a parallax amount It is characterized by having a parallax determination process in which the amount of parallax is determined from candidates and the depth in the image is determined.

  These processes were realized as a depth calculation device, a depth calculation method, and a program as follows.

  The depth calculation apparatus according to claim 1 is an apparatus that calculates a depth of an object in an image using a stereo image whose horizontal axis is parallel projection, and includes two stereo images obtained by photographing the object, and Input / output processing means for acquiring shooting parameters including the shooting interval distance between the vertical line images constituting this image and the camera installation angle at the time of stereo image shooting, and calculating image feature points from the two stereo images Image feature point calculation processing means for performing, parallax amount candidate calculation means for calculating a parallax amount candidate by voting using the calculated image feature points, and determining a parallax amount from the parallax amount candidates, and the parallax amount and the photographing Parallax amount determination processing means for calculating a depth from the parameter and attaching the depth to an image area corresponding to the parallax amount.

  In addition, the depth calculation apparatus according to claim 2 is characterized in that, in claim 1, the image feature point calculation processing means extracts, as the image feature point, a portion having a large change in gray value in a local region on the image. Features.

  According to a third aspect of the present invention, in the depth calculation apparatus according to the first or second aspect, the parallax amount candidate calculating unit shifts one image and superimposes corresponding image feature points of the two images. A necessary difference value is calculated, a vote is given to the difference value, and a candidate whose vote is equal to or greater than a threshold value is set as a parallax amount candidate.

  According to a fourth aspect of the present invention, in the depth calculation apparatus according to any one of the first to third aspects, when the parallax amount candidate calculating unit calculates the parallax amount candidates using the image feature points, It is characterized by expanding and contracting the distance.

  According to a fifth aspect of the present invention, in the depth calculation apparatus according to any one of the first to fourth aspects, the parallax amount determination processing unit occupies an image by performing processing of parallax amount candidates in order of the voted number of votes. The depth is calculated in order from an object with a large area ratio.

  The depth calculation method according to claim 6 is a method of calculating the depth of an object in an image using a stereo image whose horizontal axis is parallel projection, and includes two stereos obtained by photographing the object I / O processing step for acquiring shooting parameters including an image and a shooting interval distance between vertical line images constituting the image and a camera installation angle at the time of shooting a stereo image, and image feature points from the two stereo images An image feature point calculating step for calculating a parallax amount candidate calculating step for calculating a parallax amount candidate by voting using the calculated image feature point, determining a parallax amount from the parallax amount candidate, And a parallax amount determination processing step of calculating a depth from the imaging parameter and attaching the depth to an image area corresponding to the parallax amount.

  According to a seventh aspect of the present invention, in the depth calculation method according to the sixth aspect, the image feature point calculating step is to extract, as the image feature point, a portion having a large change in gray value in a local region on the image. Features.

  Further, in the depth calculation method according to claim 8, in the parallax amount candidate calculation step according to claim 6 or 7, when one image is shifted and the corresponding image feature points of the two images are overlapped. A necessary difference value is calculated, a vote is given to the difference value, and a candidate whose vote is equal to or greater than a threshold value is set as a parallax amount candidate.

  A depth calculation method according to a ninth aspect of the present invention is the depth calculation method according to any one of the sixth to eighth aspects, wherein the parallax amount candidate calculating step calculates a parallax amount candidate using the image feature points. It is characterized by expanding and contracting the distance.

  The depth calculation method according to a tenth aspect of the present invention is the depth calculation method according to any one of the sixth to ninth aspects, wherein the parallax amount determination processing step performs processing of the parallax amount candidates in the order of the voted number of votes, thereby including in the image It is characterized in that the depth is calculated in order from an object having a large area ratio.

  In addition, the program according to claim 11 is a computer program that enables the depth calculation device or the depth calculation method according to any one of claims 1 to 10 to be executed. Features.

  Thus, if the object in the image has a roughly constant depth, such as the depth to a building along the road in an urban area, etc., it is restored from the most occupied depth in the image by using voting. Is possible. Further, since the voting is performed only on the image feature points, the processing time can be shortened. In addition, by voting by extending or contracting several pixels between the vertical edges, it is possible to cope with the case where the distance between the vertical edges is expanded or contracted during photographing. Further, by giving priority to the parallax amount candidates based on the number of votes obtained in voting, it becomes possible to select the resolution of the depth to be restored.

  Note that the parallel projection referred to here is an expression that forms an image at the point where the parallel light radiated from the object intersects the projection plane. The size of the object in real space and the image It is proportional to the size of the object. In general, when shooting with an ordinary camera, it is represented by perspective projection and is not proportional to the size of the real space. Further, the image in which the horizontal axis here is parallel projection refers to an image in which the horizontal axis of the image is proportional to the real space.

  In addition, in order to attach the depth to the region of the image corresponding to the parallax amount, the depth corresponding to the vertical edge is added to the object corresponding to the vertical edge, and the object is sandwiched between the vertical edges. This includes adding a depth calculated from the parallax amount of the vertical edge to the object corresponding to the region.

  According to the present invention, by calculating the depth by voting only on the image feature points, it is possible to restore from a large object that is most important for the depth calculation, and it is very fast and stable and efficient. The effect which can implement | achieve a process is acquired. In addition, since the resolution of the depth to be restored can be selected, it is possible to adjust to an efficient data amount. This makes it possible to efficiently construct a three-dimensional city space in an actual urban area.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  In the embodiment of the present invention, the depth is calculated by stereo viewing using an image whose horizontal axis is parallel projection.

  Note that the parallel projection referred to here is an expression that forms an image at the point where the parallel light radiated from the object intersects the projection plane. The size of the object in real space and the image It is proportional to the size of the object. In general, when shooting with an ordinary camera, it is represented by perspective projection and is not proportional to the size of the real space. Further, the image in which the horizontal axis here is parallel projection refers to an image in which the horizontal axis of the image is proportional to the real space.

  The configuration of the depth calculation apparatus in this embodiment is shown in FIG. As shown in FIG. 1, the depth calculation apparatus receives processing from an operator, stores the processing result in the depth database 6, an image feature point calculation processing unit 2 that calculates image feature points, and an image feature. A parallax amount candidate calculation unit 3 that calculates a parallax amount candidate using the image feature points calculated by the point calculation processing unit 2, and a parallax amount calculation processing unit 4 that determines the parallax amount from the calculated parallax amount candidate; It has.

  In addition, the depth calculation apparatus stores an imaging database 5 that stores imaging parameters including a stereo image whose horizontal axis is parallel projection and a shooting interval distance between vertical line images constituting the image and a camera installation angle at the time of stereo image shooting. , And a depth database 6 that stores data in which depth information is added to an image to be processed.

  Here, in describing each unit of the depth calculation apparatus, first, the input / output processing unit 1 will be described. The input / output processing unit 1 inputs the processing target image from the image database 5, and the image feature point calculation processing unit 2, the parallax amount candidate calculation unit 3, and the parallax amount determination processing unit 4 calculate the depth of the processing target image. The result is stored in the depth database 6. This process is performed as follows in the procedure shown in FIG.

  (S21) The input / output processing means 1 includes, from the image database 5, the shooting interval distance between the two processing target images for performing stereo viewing and the vertical line images, and the camera installation angle at the time of stereo image shooting. The shooting parameters are acquired and stored in the buffer.

  (S22) The image feature point calculation processing means 2 performs image feature point calculation processing.

  (S23) The parallax amount candidate calculation unit 3 and the parallax amount determination processing unit 4 perform depth calculation processing.

  (S24) The input / output processing means 1 stores the calculated depth in the depth database 6.

  Next, the image feature point calculation processing unit 2 will be described. The image feature point calculation processing unit 2 calculates image feature points from the processing target image acquired by the input / output processing unit 1. This processing is performed as follows in the procedure shown in FIG.

  (S31) For each pixel (i, j) of the image to be processed, Sobel's edge detection operator in the x direction (supervised by the Image Processing Standard Textbook Editorial Board, “Image Processing Standard Textbook”, Image Information Education Foundation) Japan Society for the Promotion of Science, February 25, 1997, pp. 179) is applied to calculate the feature value F (i, j).

  However, the axis in parallel projection is the x direction, and the direction perpendicular thereto is the y direction.

  (S32) (i, j) having F (i, j) equal to or greater than the threshold is calculated as an image feature point.

  The Sobel edge detection operator is used to calculate the feature amount. The Laplacian filter (supervised by the Image Processing Standard Textbook Editorial Committee, “Image Processing Standard Textbook”, Foundation for Image Information Education, 1997 2) May 25, pp. 170) may be used. The threshold value of the feature amount may be set in advance as a fixed value, or may be determined using a statistically calculated value such as an average value or a deviation value of the feature amount.

  Next, the parallax amount candidate calculation unit 3 will be described. The parallax amount candidate calculating means 3 calculates parallax amount candidates using the image feature points on the two images calculated in the above procedure. This process is performed as follows in the procedure shown in FIG.

  (S41) Image feature point on one image and epipolar line of the other image (supervised by the Image Processing Standard Textbook Editorial Committee, “Image Processing Standard Textbook”, Japan Image Information Education Promotion Association, 1997 2 On the 25th of May, pp.272), the difference value of the x coordinate of the image feature point is calculated, and the number of votes is calculated for this difference value.

  (S42) The top N difference values that satisfy the threshold value for the number of votes are acquired and set as parallax amount candidates.

  Note that the depth is proportional to the difference between the x-axis coordinates of the object on both images, so the number of pixels when both images are image feature points on both images while shifting one image in the x-axis direction. What is necessary is just to count with respect to the shift amount. The number of votes obtained may be a simple count value or may be calculated as count / total count.

  Next, the parallax amount determination processing unit 4 will be described. The parallax amount determination processing unit 4 determines the parallax amount from the parallax amount candidates calculated in the above procedure. This process is performed as follows in the procedure shown in FIG.

  (S51) Delete the parallax amount candidate whose difference value does not fall within the predetermined range, and determine the parallax amount from the remaining parallax amount candidates.

  (S52) Based on the determined amount of parallax, the amount of parallax of the region sandwiched between the image feature points is determined.

  (S53) It is confirmed whether processing has been performed for all parallax amount candidates to be restored.

  If the depth range of the object to be restored is known to some extent, the range of parallax is known, so if it does not fall within the predetermined range, it is deleted and the number of votes obtained is high up to the resolution of the depth to be restored. The amount of parallax can be determined in the order of depth.

  In the following, the above-described processing procedure will be specifically described with reference to actual data.

  First, a stereo image in which the horizontal axis is parallel projection stored in the image database 5 and shooting parameters including a shooting interval distance between vertical line images constituting the image and a camera installation angle at the time of shooting the stereo image are set. Acquisition will be described by taking as an example a case where two line sensor cameras are mounted on a vehicle and an image of a building in an urban area is acquired while moving.

  FIG. 6 is a diagram for explaining the above-described image acquisition method. In FIG. 6, the two line sensor cameras 601 and 602 simultaneously start photographing the building 606 so that the optical axes of the two line sensor cameras 601 and 602 are symmetrical in the direction perpendicular to the moving direction 603. Install. The line direction 604 is perpendicular to the ground, and the angle formed between the direction perpendicular to the moving direction 603 and the optical axis of the line sensor cameras 601 and 602 is equal, and the camera installation angle is both θ (605).

  While moving the two line sensor cameras 601 and 602 in such a situation, a line image is captured at fixed distances using a rotary encoder. This distance is called a photographing interval distance. Note that a rotary encoder is a device that generates a specific signal at fixed distances by being attached to a vehicle or the like. In the present invention, a fixed distance can be freely set within a range of 1 to 100 mm. For example, a signal for releasing the shutter of the line sensor camera can be generated every 5 mm of distance travel.

  In this way, two stereo images are created by arranging the line images acquired from the respective line sensor cameras 601 and 602 in time series. Then, the obtained stereo image whose horizontal axis is parallel projection and the photographing parameters including the photographing interval distance between the vertical line images constituting the image and the camera installation angle θ (605) at the time of photographing the stereo image are obtained. This is stored in the image database 5.

  An example of a stereo image whose horizontal axis is parallel projection is shown in FIG. FIG. 7 shows an example of a cityscape where a building 711 and two telephone poles 712 and 713 are standing. The building 711 is configured with the same depth, and the telephone poles 712 and 713 are in front of the building 711. Reference numeral 703 denotes an image taken by the line sensor camera 601 tilted in the traveling direction, and this is called a right image. Reference numeral 704 denotes an image taken by the line sensor camera 602 tilted in the direction opposite to the traveling direction, and this is called a left image. The time axis direction of this image corresponds to the x axis 701 and the line direction corresponds to the y axis 702.

  Here, the ends of the building 711 shown in the right image 703 are set to 705 and 706, and the ends of the same building 711 shown in the left image 704 are set to 707 and 708. In later processing, 705, 706, 707, and 708 are set. Depth is calculated using a feature corresponding to an edge such as the end of the building 711.

In addition, although 709 on the right image 703 and 710 on the left image 704 are photographed at the same location, if the respective coordinates are (x _A , y _A ) and (x _B , y _A ), Depth L is

It is represented by That is, the depth is proportional to the difference value (parallax amount).

  Next, processing in the image feature quantity calculation unit 2 will be described. The image feature amount calculation means 2 performs image feature point calculation processing on the above-described processing target image to calculate image feature points. Here, the well-known Sobel edge detection operator in the x direction is applied to the processing target image to calculate the edge intensity of each pixel (i, j). Then, the distribution of the edge strength is calculated, a threshold value is calculated from the calculated distribution, and a pixel having edge strength that satisfies the threshold value is calculated as an image feature point. This image feature point is called a vertical edge.

  FIG. 8 illustrates the result of calculating the image feature points of the stereo image of FIG. In FIG. 8, the x-axis of the image is 801, the y-axis is 802, the right image is 803, the left image is 804, and the line segment obtained by extracting the vertical edge of the end of the building 711 shown in the right image 703 of FIG. Line segments obtained by extracting the vertical edges at the ends of the building 711 shown in the left image 704 are designated as 807 and 808, respectively.

  Next, the parallax amount candidate calculation unit 3 will be described. The parallax amount candidate calculating means 3 calculates a difference value between vertical edges, which are image feature points of both images calculated in the above procedure. That is, while shifting the right image pixel by pixel in the x-axis direction on the left image, when both pixels are vertical edges, the difference value obtained by shifting the right image is voted. That is, if both (i, j) of the right image and (i + n, j) of the left image are vertical edges, +1 is voted for the difference value n. Here, the count obtained by voting is set as the number of votes. Then, the top several difference values satisfying the threshold value for the number of votes are acquired and set as parallax amount candidates. Note that the distance between the vertical edges may be expanded or contracted at the time of shooting. Therefore, the voting may be performed by extending or contracting several pixels between the vertical edges.

  FIG. 9 shows a state in which a parallax amount candidate is obtained using the result of calculating the image feature points of FIG. In FIG. 9, when moving by the difference value n, the vertical edges of the buildings including the vertical edges 905 and 906 at both ends of the building on the right image 903 respectively change the vertical edges 907 and 908 at both ends of the building on the left image 904. It shows a state in which the degree of coincidence between the right image and the left image is maximized, matching the vertical edge of the building. Here, the number of votes to the difference value corresponding to the depth of the building is the largest, and the number of votes to the difference value corresponding to the depth of the telephone pole is the second largest.

  Next, the parallax amount determination processing unit 4 will be described. The parallax amount determination processing unit 4 determines the parallax amount from the parallax amount candidates calculated in the above procedure. With respect to the parallax amount with the highest number of votes, a depth for this parallax amount is given to the vertical edges on both images that overlap when the parallax amount is shifted. Next, the depth of the region sandwiched between the vertical edges to which the depth corresponding to the parallax amount with the highest number of votes is obtained for each y coordinate is determined. That is, the regions on the original image sandwiched between the vertical edges are compared, and if the similarity between the regions satisfies the threshold, the depth is determined with respect to the parallax amount given to the vertical edges sandwiching the region. . Next, the same processing as described above is performed for the parallax amount having the second highest number of votes.

  FIG. 10 shows how the depth of a region sandwiched between vertical edges, which are image feature points, is determined. In FIG. 10, 1001 indicates the x-axis of the image, 1002 indicates the y-axis of the image, 1003 indicates the right image, and 1004 indicates the left image. Reference numerals 1011, 1012, and 1013 denote the building 711 and the telephone poles 712 and 713 in FIG. First, when the right image 1003 is shifted in the x-axis direction by the parallax amount n having the largest number of votes, a depth corresponding to the parallax amount n is given to the vertical edge overlapping the right image 1003 and the left image 1004. . That is, depths corresponding to the parallax amount n are given to the vertical edges 1005 and 1006 and 1007 and 1008. Next, the depth of the region sandwiched between the vertical edges that overlap when shifted by the amount of parallax n is determined. Reference numeral 1009 denotes an area sandwiched between 1005 and 1006 to which a depth corresponding to the parallax amount n having the highest number of votes on the right image 1003 is given. Reference numeral 1010 denotes an area between 1007 and 1008 to which a depth corresponding to the parallax amount n on the left image 1004 is given. When the similarity between the region P and the region Q satisfies the threshold, a depth corresponding to the parallax amount n corresponding to the same depth as 1005, 1006, 1007, and 1008 is given. Note that a difference in pixel value or a correlation value can be used as the similarity.

  In this way, the depth can be given to a building having a depth corresponding to the parallax amount n having the highest number of votes. Next, the same process as described above is performed for the parallax amount with the second vote count.

  Depth should be determined in the order of the amount of parallax with the largest number of votes in this way, and if the depth of the building with the largest occupied area in the image is constant, the number of votes will increase, so the largest building will be restored in order There is an effect that can be done.

  Further, when restoring to a fine detailed shape, the number of parallax amount candidates for determining the depth may be increased, and when only the global shape is restored, the number of parallax amount candidates may be reduced. For example, if the captured cityscape is mostly composed of a telegraph pole in front of the sidewalk and a building in the back of the sidewalk, and the depth of the building is almost constant, the two types of depth of the telegraph pole and the building can be identified. All you need to do is restore the depth of the second vote. In this way, it is possible to select according to the level of the detailed shape for which the resolution of the depth to be restored is to be obtained.

  By voting only on the image feature points and calculating the depth as described above, the processing time can be shortened and the object having the largest occupied area in the image stably, that is, the most important It is possible to recover from a large object that seems to be.

  In the present invention, some or all of the processing functions of the apparatus shown in FIG. 1 are configured as a program and realized using a computer, or the processing procedures shown in FIGS. 2 to 10 are configured as a program. It can be executed by a computer. In addition, a computer-readable recording medium such as a flexible disk, MO, ROM, or memory card can be used to store a program for realizing the processing function of each unit by the computer or a program for causing the computer to execute the processing procedure. It can be recorded on a CD, a DVD, a removable disk, etc., stored, provided, and distributed via a communication network such as the Internet.

The schematic block diagram of a depth calculation apparatus. The process flowchart of this invention. The flowchart of an image feature point calculation process means. The processing flowchart of a parallax amount candidate calculation means. The processing flowchart of a parallax amount determination processing means. Explanatory drawing of an image acquisition method. The figure which shows an example of the acquired image. The figure which shows calculation of an image feature point. The figure which shows calculation of a parallax amount candidate. The figure which shows the determination of the depth of the area | region pinched | interposed into the image feature point.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Input / output processing means 2 ... Image feature point calculation processing means 3 ... Parallax amount candidate calculation means 4 ... Parallax amount determination processing means 5 ... Image database 6 ... Depth database 601 ... Camera 602 ... Camera 603 ... Movement direction 604 ... Line direction 605 ... Camera angle 606 ... Building 701 ... x axis 702 ... y axis 703 ... Right image 704 ... Left image 705 ... Building edge 706 ... Building edge 707 ... Building edge 708 ... Building edge 709 ... One on the image Location 710: One location on the image 711 ... Building 801 ... x-axis 802 ... y-axis 803 ... Right image 804 ... Left image 805 ... Line segment extracted from the vertical edge of the building edge 806 ... Extract vertical edge at the building edge 807 ... Line segment obtained by extracting the vertical edge of the building edge 808 ... Line segment obtained by extracting the vertical edge of the building edge 901 ... x-axis 902 ... y-axis 903 ... right image 904 ... Left image 905 ... Vertical edge at the end of the building 906 ... Vertical edge at the end of the building 907 ... Vertical edge at the end of the building 908 ... Vertical edge at the end of the building 1001 ... X-axis 1002 ... Y-axis 1003 ... Right image 1004 ... Left image 1005 ... Vertical edge 1006 ... Vertical edge 1007 ... Vertical edge 1008 ... Vertical edge 1009 ... Region P
1010 ... Area Q
1011 ... Building 1012 ... Telegraph pole 1013 ... Telegraph pole

Claims (11)

An apparatus for calculating the depth of an object in an image using a stereo image whose horizontal axis is parallel projection,
Input / output processing means for acquiring shooting parameters including two stereo images obtained by shooting an object, a shooting interval distance between vertical line images constituting the image, and a camera installation angle during stereo image shooting;
Image feature point calculation processing means for calculating image feature points from the two stereo images;
Parallax amount candidate calculating means for calculating a parallax amount candidate by voting using the calculated image feature points;
Parallax amount determination processing means for determining a parallax amount from the parallax amount candidates, calculating a depth from the parallax amount and the shooting parameter, and attaching the depth to an image region corresponding to the parallax amount. Depth calculation device characterized by. The image feature point calculation processing means includes:
The depth calculation apparatus according to claim 1, wherein a portion having a large change in gray value in a local region on the image is extracted as the image feature point. The parallax amount candidate calculating means includes:
Calculating a difference value required to superimpose corresponding image feature points of the two images by shifting one image;
3. The depth calculation apparatus according to claim 1, wherein voting is performed for the difference value, and a candidate whose parallax amount is equal to or greater than a threshold value is set as a parallax amount candidate. The parallax amount candidate calculating means includes:
The depth calculation apparatus according to claim 1, wherein when calculating a parallax amount candidate using the image feature point, a distance between the image feature points is expanded and contracted. The parallax amount determination processing means includes
The depth calculation according to any one of claims 1 to 4, wherein the depth calculation is performed in order from an object having a large area ratio in the image by processing the parallax amount candidates in order of the voted number of votes. apparatus. A method for calculating the depth of an object in an image using a stereo image whose horizontal axis is parallel projection,
An input / output processing step for acquiring shooting parameters including two stereo images obtained by shooting an object, a shooting interval distance between vertical line images constituting the image, and a camera installation angle at the time of shooting a stereo image;
An image feature point calculating step for calculating an image feature point from the two stereo images;
A parallax amount candidate calculating step of calculating a parallax amount candidate by voting using the calculated image feature points;
A parallax amount determination processing step of determining a parallax amount from the parallax amount candidates, calculating a depth from the parallax amount and the shooting parameter, and attaching the depth to an image region corresponding to the parallax amount. Depth calculation method characterized by The image feature point calculation step includes:
The depth calculation method according to claim 6, wherein a portion having a large change in gray value in a local region on the image is extracted as the image feature point. The parallax amount candidate calculating step includes:
Calculating a difference value required to superimpose corresponding image feature points of the two images by shifting one image;
8. The depth calculation method according to claim 6, wherein voting is performed on the difference value, and a candidate whose parallax amount is equal to or greater than a threshold value is set as a parallax amount candidate. The parallax amount candidate calculating step includes:
The depth calculation method according to claim 6, wherein when calculating a parallax amount candidate using the image feature points, a distance between the image feature points is expanded and contracted. The parallax amount determination processing step includes:
The depth calculation according to any one of claims 6 to 9, wherein the depth calculation is performed in order from an object having a large area ratio in the image by performing processing of parallax amount candidates in the order of the voted number of votes. Method.   A program characterized in that the depth calculation device or the depth calculation method according to any one of claims 1 to 10 is described by a computer program and can be executed.

Images